Thematic discussion 4: Respecting, protecting and promoting human rights - Global Dialogue on Artificial Intelligence Governance - Day 2

40 intervenants
Résumé

Résumé

Cette discussion, convoquée par le HCDH et la Banque mondiale dans le cadre d'un Dialogue mondial de l'ONU sur la gouvernance de l'IA, s'est concentrée sur l'intersection entre l'intelligence artificielle, les droits humains, la transparence, la responsabilité et la supervision humaine. Le Haut-Commissaire Volker Türk a ouvert les débats en établissant un parallèle entre le déploiement de l'IA et la réglementation pharmaceutique, s'interrogeant sur la rigueur avec laquelle l'IA est traitée , et avertissant que l'IA alimente déjà la surveillance de masse, la désinformation et les biais fondés sur le genre . Il a soutenu que le droit international des droits humains fournit un cadre contraignant pour gouverner l'IA et que la réglementation ne doit pas être perçue comme un obstacle à l'innovation, mais plutôt comme le fondement de la confiance publique .

Les panélistes du premier panel ont mis en lumière les dimensions genrées des préjudices causés par l'IA : Sima Bahous a noté que 44 % des systèmes d'IA évalués présentent des biais de genre et que jusqu'à 99 % des hypertrucages (deepfakes) en ligne ciblent des femmes . Samuel Arias Arzeno a souligné que les systèmes judiciaires doivent être renforcés pour traduire les principes des droits humains en recours effectifs , tandis qu'Amal El Fallah Seghrouchni a soulevé le défi particulier de la responsabilité dans les systèmes d'IA agentiques comportant des milliards de paramètres, où la responsabilité causale est extrêmement difficile à établir . Sonia Livingstone a présenté les données du panel scientifique faisant état d'une multiplication des violations des droits humains, notamment une augmentation exponentielle des contenus pédopornographiques générés par l'IA et des défaillances systématiques des modèles d'IA en matière de sécurité dans les environnements destinés aux enfants .

Le deuxième panel a examiné des solutions concrètes et l'inclusion des voix marginalisées. Nighat Dad a soutenu que la diligence raisonnable en matière de droits humains est appliquée de manière inégale selon les marchés, des évaluations significatives n'étant menées que là où le droit de l'UE l'exige, créant ainsi un régime de droits à deux vitesses déterminé par la géographie . Alvitta Ottley a identifié un « décalage d'évaluation », notant que les indicateurs d'ingénierie tels que la vitesse et la précision ne correspondent pas aux questions sociétales relatives à la protection des droits humains . Felipe Paullier a souligné que les jeunes sont les utilisateurs les plus actifs de l'IA, mais qu'ils ne participent presque jamais aux prises de décision, et a appelé à leur inclusion significative dans les cadres nationaux de gouvernance de l'IA .

Les interventions du public, émanant de gouvernements et de la société civile, ont renforcé ces thèmes. Le Brésil a présenté son Statut numérique pour les enfants et les adolescents, qui impose des mesures de protection de l'enfance dès la phase de conception , tandis que la Pologne et la République de Corée ont souligné la nécessité de stratégies nationales d'IA centrées sur l'humain et de cadres juridiques contraignants . Des représentants de la société civile, notamment d'AccessNow, de l'ICNL et de la Digital Rights Foundation, ont appelé à l'établissement de lignes rouges pour l'IA, à des évaluations obligatoires de l'impact sur les droits humains, et au financement des organisations de la société civile du Sud global en tant que systèmes d'alerte précoce face aux préjudices liés à l'IA .

La discussion s'est conclue par un large consensus selon lequel la gouvernance de l'IA doit intégrer les droits humains dès la phase de conception, garantir la responsabilité tout au long du cycle de vie de l'IA, et inclure de manière significative les enfants, les femmes et les communautés du Sud global - non pas simplement en tant que sujets de protection, mais en tant que participants actifs à l'élaboration de la technologie qui gouverne de plus en plus leurs vies .

Points clés

Objectif général

La discussion est un dialogue mondial convoqué par l'ONU sur la gouvernance de l'IA, co-dirigé par le HCDH et la Banque mondiale, réunissant des gouvernements, des organisations internationales, la société civile et des experts techniques afin d'examiner comment l'intelligence artificielle peut être développée, déployée et réglementée de manière conforme au droit international des droits humains. La session vise à identifier des cadres de gouvernance concrets, des mécanismes de responsabilité et des garanties - en particulier pour les groupes vulnérables tels que les femmes, les enfants et les communautés du Sud global - et à créer une dynamique en faveur d'une réglementation de l'IA contraignante et centrée sur les droits humains.

---

Principaux points de discussion

- La nécessité urgente d'une réglementation de l'IA fondée sur les droits humains, par analogie avec d'autres secteurs réglementés. Plusieurs intervenants ont soutenu que l'IA doit être soumise à la même surveillance rigoureuse que celle appliquée aux médicaments, aux véhicules et aux aéronefs. Le Haut-Commissaire aux droits de l'homme, Volker Türk, a ouvert les débats en demandant si la société accorde à l'IA le même soin qu'aux nouveaux médicaments, lesquels font l'objet de 10 à 15 ans d'essais cliniques avant leur autorisation de mise sur le marché . Il a averti que l'IA est déployée « à une vitesse vertigineuse » et alimente déjà la surveillance de masse, la désinformation et les biais de genre . Il a soutenu qu'une réglementation guidée par les droits humains est « la solution la plus évidente » et que la présenter comme un compromis avec l'innovation est un faux choix . Ce thème a été repris par Sima Bahous d'ONU Femmes , par le représentant de l'IDLO Mark Cassayre , et par la ministre irlandaise Niamh Smyth .

- Les préjudices genrés et intersectionnels causés par l'IA, et les lacunes en matière de responsabilité qui permettent leur persistance. Les intervenants ont présenté de nombreuses preuves que l'IA nuit de manière disproportionnée aux femmes, aux filles et aux communautés marginalisées. Les recherches d'ONU Femmes ont révélé que 44 % des systèmes d'IA évalués présentent des biais de genre, et que près d'une femme défenseure des droits humains sur quatre interrogées avait subi des violences en ligne assistées par l'IA . Jusqu'à 99 % des hypertrucages et des images sexuelles manipulées en ligne ciblent des femmes . Nighat Dad de la Digital Rights Foundation a souligné que la diligence raisonnable en matière de droits humains est appliquée de manière inégale selon les marchés - des évaluations significatives n'ont lieu que là où la loi l'exige (principalement dans l'UE), tandis que les mêmes systèmes sont déployés dans le Sud global sans contrôle équivalent . Elle a appelé à des évaluations obligatoires de l'impact sur les droits des femmes et des enfants, menées avant le déploiement, avec la participation des communautés concernées . Esther Eghobamien-Mshelia du Comité CEDAW a réaffirmé que la CEDAW constitue un cadre essentiel obligeant les États à parvenir à l'égalité substantielle et à s'attaquer aux obstacles structurels renforcés par l'IA .

- La vulnérabilité particulière des enfants dans les environnements médiatisés par l'IA et l'insuffisance des protections actuelles. Les droits de l'enfant ont constitué une préoccupation transversale tout au long de la session. Sonia Livingstone du Panel scientifique indépendant a rapporté que, dans 11 pays du Sud global, jusqu'à un enfant par classe a déclaré que l'IA avait été utilisée pour créer des hypertrucages sexuellement explicites le représentant, et que les signalements de contenus pédopornographiques générés par l'IA augmentent de façon exponentielle . Les évaluations des risques menées par Stanford sur les compagnons IA ont révélé que ces produits ne satisfont pas aux normes de sécurité de base . Jhalak M. Kakkar a insisté sur le fait que les systèmes d'IA doivent être conçus avec la sécurité des enfants intégrée dès le départ, et non ajoutée après coup , et a soulevé la question de l'impact d'une surveillance constante sur le développement des enfants en tant que citoyens démocratiques . Isabella Henriques, s'exprimant au nom de plus de 120 signataires d'une déclaration commune de la société civile, a appelé à ce que les droits de l'enfant constituent une considération transversale dans tous les cadres de gouvernance de l'IA .

- Les lacunes en matière de responsabilité dans les systèmes d'IA agentiques et le défi d'attribuer la responsabilité dans des chaînes multi-agents complexes. Plusieurs panélistes ont identifié l'IA agentique - des systèmes dans lesquels des réseaux d'agents autonomes agissent sans instruction humaine directe - comme le défi juridique et de gouvernance émergent le plus significatif. Amal El Fallah Seghrouchni du Maroc a décrit le problème central : lorsqu'un réseau d'agents IA produit un résultat préjudiciable, il est difficile de déterminer quel agent est responsable et comment la responsabilité doit être répartie . Elle a proposé trois obligations : une documentation structurée des chaînes d'agents, la désignation d'un responsable humain identifiable pour les questions relatives aux droits, et un droit à réparation dans un délai raisonnable . Sasha Luccioni a averti que l'IA agentique amplifie les degrés de séparation entre les utilisateurs et la technologie sous-jacente, rendant plus difficile l'identification de l'origine des défaillances . Les résultats du sondage auprès du public, rapportés par Peggy Hicks, ont montré que les répondants considèrent en grande majorité les entreprises développant l'IA comme les principales responsables, suivies des États, puis des organisations qui la déploient .

- Les asymétries de pouvoir mondiales dans la gouvernance de l'IA, notamment l'exclusion du Sud global, de la société civile et des communautés marginalisées des processus décisionnels. Plusieurs intervenants ont remis en question la concentration du pouvoir en matière d'IA dans un petit nombre d'entreprises et de pays. Kakkar a soutenu que sans diffusion de cette concentration de pouvoir, les droits humains - en particulier pour les populations du Sud global - ne peuvent être protégés . Pria Chetty de Research ICT Africa a proposé un cadre « IA juste », soutenant que les modèles éthiques d'IA dominants sont autorégulatoires, calibrés pour les pays à revenus élevés, et approfondissent les inégalités existantes . Linda Bonyo, dans ses remarques de clôture, a souligné que 51 % des discussions sur la gouvernance de l'IA se déroulent à Genève, alors que de nombreuses voix du reste du monde en sont exclues parce qu'elles ne peuvent pas obtenir de visas - un processus lui-même façonné par des algorithmes opaques . L'Association pour les communications progressistes a souligné que les communautés constituent « le premier kilomètre, et non le dernier kilomètre » de l'IA et doivent être placées au centre de la gouvernance .

---

Ton général

Le ton général de la discussion est celui d'une préoccupation urgente tempérée par une détermination prudente. D'emblée, les intervenants ont exprimé un sentiment d'alarme - les remarques d'ouverture de Volker Türk étaient sombres et moralement chargées, invoquant Le Château de Kafka comme métaphore des systèmes d'IA opaques et non responsables , et avertissant que la société risque de répéter les échecs de la gouvernance climatique en permettant à des intérêts puissants de retarder l'action . Cette gravité a persisté tout au long du premier panel, où les panélistes ont présenté des preuves empiriques croissantes des préjudices subis par les femmes, les enfants et les communautés marginalisées.

Cependant, le ton a progressivement évolué vers une résolution constructive des problèmes lors du deuxième panel et des interventions du public. Des ministres des Pays-Bas, d'Espagne, d'Irlande, de Corée et de Pologne ont présenté des mesures nationales concrètes - registres d'algorithmes, lois sur l'IA, lois sur la protection de l'enfance - signalant que la gouvernance n'est pas simplement aspirationnelle . Les représentants de la société civile, bien que critiques, ont formulé des recommandations précises plutôt que de simplement dresser un inventaire des préjudices.

Dans les remarques de clôture, le ton est devenu celui d'une résolution collective, le co-président Oscar López Águeda reconnaissant que « nous sommes en retard », tout en affirmant que la direction prise - vers une IA humaniste et respectueuse des droits - est claire et non négociable . Tout au long de la session, un courant sous-jacent de frustration était perceptible, notamment de la part des représentants du Sud global qui ont noté les exclusions structurelles du dialogue même censé y remédier .

Intervenants

- Ulises Gutiérrez - Représentant spécial pour les technologies émergentes, Mexique

- Jhalak M. Kakkar - Directrice exécutive, Centre for Communication Governance, Université nationale de droit de Delhi

- Clara Chappaz - Ambassadrice pour l'IA et les affaires numériques, France

- Sopio Kiladze - Membre, Comité des droits de l'enfant de l'ONU

- Isabella Hendricks - (s'est exprimée au nom de l'Instituto Alana lors de la session ; voir également Isabella Henriques ci-dessous)

- Felipe Paullier - Secrétaire général adjoint, Bureau de l'ONU pour la jeunesse

- Linda Bonyo - Fondatrice et PDG, Lawyers Hub Africa ; Co-présidente thématique de la session ; expertise en droit, technologie et droits numériques en Afrique

- Sima Bahous - Directrice exécutive, ONU Femmes ; expertise en égalité des genres, droits des femmes et impacts genrés de l'IA

- Zachary Lampel - Conseiller juridique principal et coordinateur, Programme des droits numériques, Centre international pour le droit des organisations à but non lucratif (ICNL)

- João Brant - Secrétaire aux politiques numériques, Secrétariat à la communication sociale, Présidence de la République du Brésil

- Volker Turk - Haut-Commissaire des Nations Unies aux droits de l'homme ; expertise en droit international des droits humains, gouvernance de l'IA et responsabilité

- Shumaila Hussaini Shahani - Responsable des politiques et du plaidoyer, Tech Global Institute ; expertise sur les lacunes en matière d'équité et de responsabilité entre les plateformes technologiques, les gouvernements et les communautés de la majorité mondiale

- Jérôme Bellion-Jourdan - Chercheur associé, Institut pour la négociation mondiale ; Vice-président, Comité directeur de l'initiative AI for Good Impact ; expertise en négociation multilatérale et processus de gouvernance de l'IA

- Esther Eghobamien-Mshelia - Membre, Comité pour l'élimination de la discrimination à l'égard des femmes (CEDAW) ; expertise en discrimination de genre, droits des femmes et gouvernance de l'IA

- Mark Cassayre - Observateur permanent de l'IDLO (Organisation internationale de droit du développement) auprès de l'ONU à Genève ; expertise en état de droit, innovation numérique et accès à la justice

- Franco Giandana Gigena - Analyste des politiques, AccessNow ; expertise en droits numériques, surveillance et gouvernance de l'IA pour la majorité mondiale

- Samuel Arias Arzeno - Juge, Première chambre de la Cour suprême de justice, République dominicaine ; expertise en responsabilité judiciaire, IA dans les systèmes judiciaires et droit des droits humains

- Jeremy Ng - Représentant, Banque mondiale ; impliqué dans la co-direction de la session de dialogue sur la gouvernance de l'IA du système onusien

- Amal El Fallah Seghrouchni - Ministre déléguée chargée de la Transition numérique, Maroc ; expertise en systèmes d'IA, responsabilité algorithmique et droits cognitifs

- Willemijn Aerdts - Ministre de l'Économie numérique et de la Souveraineté, Pays-Bas ; expertise en gouvernance de l'IA, transparence algorithmique et droits numériques des enfants

- David Lametti - Représentant permanent du Canada auprès de l'ONU, Genève ; ancien ministre de la Justice du Canada ; expertise en droit, responsabilité civile, gouvernance de l'IA et droits de l'enfant

- Wanjin Park - Vice-président, KT (entreprise de télécommunications coréenne) ; expertise en évaluation des risques liés à l'IA, diligence raisonnable en matière de droits humains et normes de sécurité de l'IA

- Co-modérateur - A facilité le segment des interventions du public ; rôle et nom non précisés dans la transcription

- Jihoon Cha - Représentant permanent de la République de Corée auprès de l'ONU, Genève ; expertise en législation sur l'IA, droits humains et renforcement mondial des capacités en matière d'IA

- Sonia Livingstone - Membre, Panel scientifique international indépendant sur l'IA ; expertise en droits de l'enfant, sécurité en ligne et évaluation de l'impact de l'IA fondée sur des données probantes

- Oscar Lopez Agueda - Ministre de la Transformation numérique et de la Fonction publique, Espagne ; Co-président thématique de la session ; expertise en réglementation de l'IA, droits numériques et gouvernance de l'IA dans l'UE

- Elizabeth Tan - Directrice de la protection internationale, HCR ; expertise en protection des réfugiés, apatridie et impact de l'IA sur les personnes déplacées et apatrides

- Niamh Smyth - Ministre chargée de l'Intelligence artificielle, Irlande ; expertise en gouvernance de l'IA, violences basées sur le genre et droits numériques des enfants

- Nighat Dad - Fondatrice, Digital Rights Foundation ; expertise en violences basées sur le genre facilitées par la technologie, droits numériques dans le Sud global et responsabilité en matière d'IA

- Peggy Hicks - Directrice, Division des procédures thématiques et spéciales, Haut-Commissariat des Nations Unies aux droits de l'homme (HCDH) ; hôte et modératrice de la session

- Alejandra De Bellis Bonilla - Représentante permanente de l'Uruguay auprès de l'ONU, Genève ; expertise en gouvernance de l'IA, égalité des genres et transformation numérique

- Anna Osterling - Représentante auprès de l'ONU, Forum mondial pour le développement des médias (GFMD) ; expertise en journalisme, intégrité de l'information et impact de l'IA sur les écosystèmes médiatiques

- Porte-parole de Raman Jit Singh Chima - Rebecca Rektimbo, s'exprimant au nom de Raman Jit Singh Chima, Directeur mondial des programmes, Association pour les communications progressistes (APC) ; expertise en connectivité, droits numériques et infrastructures centrées sur les communautés

- Rafał Kownacki - Directeur, Département de la coopération internationale, Ministère des affaires numériques, Pologne ; expertise en gouvernance de l'IA, droits humains et stratégie nationale de numérisation

- Robert Baruch - Affaires publiques, Relations européennes et multilatérales, Universal Music Group ; expertise en droits des créateurs, IA et propriété intellectuelle, et gouvernance de l'industrie musicale

- Pria Chetty - Directrice exécutive, Research ICT Africa ; expertise en politique numérique, cadre d'IA juste et gouvernance de l'IA pour le continent africain et la majorité mondiale

- Sasha Luccioni - Co-fondatrice et directrice scientifique, Sustainable AI Group ; expertise en empreinte environnementale de l'IA, efficacité énergétique et transparence des systèmes d'IA

- Isabella Henriques - Directrice exécutive, Instituto Alana (Brésil) ; expertise en droits de l'enfant dans l'environnement numérique, gouvernance de l'IA et perspectives du Sud global

- Alvitta Ottley - Membre, Panel scientifique international indépendant sur l'IA ; expertise en lacunes probatoires, méthodologie d'évaluation et relation entre science et société dans la gouvernance de l'IA

- Anita Pipan - Conseillère scientifique, Représentation permanente de la Slovénie auprès de l'ONU, Genève ; a présenté la déclaration de la République de Slovénie sur la gouvernance de l'IA et les approches centrées sur l'humain

---

Intervenants supplémentaires :

- Egriselda López - Ambassadrice du Salvador ; Co-présidente du dialogue élargi (mentionnée mais n'a pas pris la parole directement dans la transcription)

- Rein Tammsaar - Ambassadeur d'Estonie ; Co-président du dialogue élargi (mentionné mais n'a pas pris la parole directement dans la transcription)

Intervenants
VT
Volker Turk
121 wpm · 9 min
SB
Sima Bahous
133 wpm · 7 min
OL
Oscar Lopez Agueda
107 wpm · 9 min
ND
Nighat Dad
129 wpm · 5 min
PC
Pria Chetty
123 wpm · 3 min
UG
Ulises Gutiérrez
119 wpm · 4 min
ZL
Zachary Lampel
160 wpm · 3 min
MC
Mark Cassayre
128 wpm · 3 min
SA
Samuel Arias Arzeno
113 wpm · 4 min
AE
Amal El Fallah Seghrouchni
144 wpm · 6 min
SL
Sasha Luccioni
180 wpm · 4 min
WA
Willemijn Aerdts
198 wpm · 3 min
JM
Jhalak M. Kakkar
145 wpm · 7 min
SL
Sonia Livingstone
144 wpm · 6 min
DL
David Lametti
119 wpm · 6 min
IH
Isabella Henriques
136 wpm · 2 min
EE
Esther Eghobamien-Mshelia
127 wpm · 3 min
NS
Niamh Smyth
153 wpm · 3 min
JB
João Brant
142 wpm · 3 min
WP
Wanjin Park
98 wpm · 3 min
AO
Alvitta Ottley
159 wpm · 6 min
FP
Felipe Paullier
135 wpm · 5 min
AO
Anna Osterling
124 wpm · 3 min
ET
Elizabeth Tan
97 wpm · 3 min
JB
Jérôme Bellion-Jourdan
160 wpm · 3 min
RK
Rafał Kownacki
147 wpm · 3 min
SH
Shumaila Hussaini Shahani
150 wpm · 2 min
LB
Linda Bonyo
146 wpm · 8 min
AD
Alejandra De Bellis Bonilla
115 wpm · 2 min
FG
Franco Giandana Gigena
120 wpm · 3 min
JC
Jihoon Cha
107 wpm · 3 min
RJ
Raman Jit Singh Chima spokesman
129 wpm · 3 min
SK
Sopio Kiladze
111 wpm · 3 min
CC
Clara Chappaz
142 wpm · 6 min
IH
Isabella Hendricks
146 wpm · 42 s
RB
Robert Baruch
151 wpm · 3 min
PH
Peggy Hicks
167 wpm · 10 min
AP
Anita Pipan
109 wpm · 3 min
JN
Jeremy Ng
192 wpm · 28 s
C
Co-moderator
108 wpm · 2 min

Résumé élargi : Dialogue mondial de l'ONU sur la gouvernance de l'IA - Groupe thématique sur les droits de l'homme, la transparence, la responsabilité et la supervision humaine

#

Ouverture et contexte

La session a été convoquée dans le cadre du Dialogue mondial de l'ONU sur la gouvernance de l'IA, co-dirigée par le Haut-Commissariat des Nations Unies aux droits de l'homme (HCDH) et la Banque mondiale, sous la direction thématique des co-présidents Linda Bonyo, fondatrice et directrice générale du Lawyers Hub Africa, et Son Excellence Óscar López Águeda, ministre espagnol de la Transformation numérique et de la Fonction publique . Peggy Hicks, directrice de la Division des procédures thématiques et spéciales du HCDH, a ouvert les travaux en accueillant les participants à une session axée sur le respect, la protection et la promotion des droits de l'homme, de la transparence, de la responsabilité et de la supervision humaine dans le contexte de l'intelligence artificielle . Le dialogue s'est déroulé sous la direction de l'Ambassadrice Egriselda López du Salvador et de l'Ambassadeur Rein Tammsaar d'Estonie, et a été décrit comme une réalisation remarquable de la façon dont le système onusien répond aux enjeux actuels .

#

Allocution d'ouverture du Haut-Commissaire

Le Haut-Commissaire des Nations Unies aux droits de l'homme Volker Türk a ouvert son intervention par une analogie frappante : avant qu'un nouveau médicament arrive sur le marché, il fait l'objet de dix à quinze années d'essais cliniques, d'examen réglementaire et d'autorisation de mise sur le marché, et même avec toutes ces garanties, des erreurs peuvent encore survenir . Il a demandé directement si la société prend les mêmes précautions avec l'intelligence artificielle . Sa réponse était préoccupante : l'IA est déjà développée et déployée à une vitesse fulgurante, alimentant la surveillance de masse, la désinformation sur les réseaux sociaux et permettant des discriminations, notamment sous l'angle du genre, tandis que les préjudices causés aux enfants par les algorithmes alimentés par l'IA s'aggravent et que les centres de données hébergeant les serveurs d'IA asphyxient l'environnement . Dans le même temps, il a reconnu le potentiel remarquable de l'IA pour soutenir le diagnostic médical, accélérer la recherche scientifique et renforcer les services publics .

Türk a soutenu que la solution la plus évidente - une réglementation guidée par les droits de l'homme - est là, sous nos yeux, et que pourtant la société continue de l'esquiver . Il a établi un parallèle avec le changement climatique, avertissant que de précieuses années ont été perdues parce que des acteurs puissants ont semé le doute sur la science, retardé l'action et privilégié des intérêts économiques et politiques à court terme, et que la même dynamique risque de se reproduire avec l'IA . Invoquant le roman de Franz Kafka Le Château, il a demandé si la société ressemble au protagoniste K - un géomètre piégé dans un système de pouvoir opaque dans lequel l'agentivité humaine se perd - et a soutenu que l'IA est fondamentalement une question de pouvoir sur les données, les marchés, les ressources et l'information . Sa réponse à ce défi était claire : les droits de l'homme aident à reconquérir l'agentivité et constituent un contrepoids à l'exercice du pouvoir, et des garde-fous sur le développement de l'IA sont nécessaires pour garantir la transparence, l'inclusion et la responsabilité . Il a également invoqué la maxime de Lord Acton selon laquelle « le pouvoir tend à corrompre et le pouvoir absolu corrompt absolument » pour justifier la nécessité de maintenir le pouvoir de l'IA sous contrôle .

Sur la question de la réglementation, Türk a explicitement rejeté l'idée qu'elle constitue un compromis avec l'innovation, faisant valoir que les normes de sécurité applicables aux médicaments, aux voitures et aux avions ne sont pas considérées comme des obstacles au progrès, mais sont précisément la raison pour laquelle les gens font confiance à ces technologies . Il a appelé à intégrer les droits de l'homme dans la conception, le développement, le déploiement et l'utilisation de l'IA, notant que le droit international des droits de l'homme constitue un cadre juridique contraignant pour protéger les données, éviter les discriminations, accéder à la justice et garantir l'égalité . Il a souligné que la diligence raisonnable en matière de droits de l'homme et les évaluations d'impact des systèmes d'IA sont essentielles, tout comme des garanties robustes en matière de protection des données et de respect de la vie privée . Il a également averti que lorsque l'IA collecte des données ou prend une décision sans intervention humaine, la responsabilité peut se dissoudre dans le système, et que la supervision humaine ne peut pas se réduire à un simple tampon, mais exige qu'une personne identifiée soit investie de l'autorité, de la compétence, du temps, de l'indépendance et du pouvoir de modifier, voire d'arrêter un système .

Il a annoncé que le lendemain, son bureau lancerait le Human Rights Data Exchange (HRDX), décrit comme le premier service ouvert et faisant autorité au monde, fournissant des preuves sur les endroits où les droits sont menacés, ce qui s'est passé, pourquoi et ce qu'il convient de faire . Il a également mentionné le service consultatif du HCDH en matière de droits de l'homme, créé dans le cadre du Pacte numérique mondial, visant à aider les États et d'autres acteurs à gouverner l'IA conformément à leurs responsabilités en matière de droits de l'homme . Il a conclu en opposant le slogan officieux de l'industrie technologique - « plus grand, plus rapide, meilleur » - à sa propre alternative préférée : « plus intelligent, plus humain, plus sage » .

#

Allocutions d'ouverture des co-présidents thématiques

Linda Bonyo a contextualisé la discussion dans une perspective africaine, prenant la parole au nom des travailleurs de plateforme gérés par des algorithmes au Kenya et reconnaissant l'asymétrie du pouvoir ainsi que les voix des jeunes générations absentes de la salle . Elle a mis en avant le lancement de l'Africa AI Governance Index, qui, pour la première fois, suit les stratégies, les lois et les institutions en matière d'IA dans tous les États africains, et a invité les participants à s'engager avec ces travaux sur AIpolicy.africa .

Le ministre López Águeda, s'exprimant en espagnol, a présenté la gouvernance de l'IA comme un choix entre droits numériques et oligarchie, avertissant que sans une gouvernance inclusive, le Sud global ne servira que de batterie ou de fournisseur de données pour les grandes puissances . Il a présenté les réalisations législatives concrètes de l'Espagne, notamment la loi européenne sur l'IA signée sous la présidence espagnole de l'UE, l'interdiction des deepfakes sexuels et de la pornographie infantile dans la législation européenne, la création de la première agence européenne de supervision de l'IA, et une loi sur la protection de l'enfance restreignant l'accès des moins de 16 ans aux réseaux numériques . Il a également mentionné la charte ibéro-américaine des droits numériques comme une initiative portée par l'Espagne .

#

Premier panel : droits de l'homme, responsabilité et justice

Le premier panel était modéré par l'Ambassadeur David Lametti, Représentant permanent du Canada auprès de l'ONU et ancien ministre de la Justice, qui a cadré la discussion autour du rôle des systèmes judiciaires dans la mise en œuvre de la responsabilité et la défense des droits de l'homme pour les personnes affectées par l'IA, y compris les enfants . Il a noté que la responsabilité civile est un principe qui existe dans la plupart des systèmes judiciaires et qui est axé sur la responsabilité humaine, mais que le défi posé par l'IA est d'être soumise au jugement et à l'intelligence humains alors que c'est une machine qui prend la décision . Il a décrit la loi proposée par le Canada qui interdirait aux plateformes sociales d'accueillir des enfants de moins de 16 ans jusqu'à ce que leur sécurité soit prouvée, et exigerait que les chatbots d'IA soient déclarés sûrs avant d'être exposés aux enfants, avec un commissaire au numérique chargé d'assurer la conformité .

Sima Bahous, directrice exécutive d'ONU Femmes, a soutenu que, sans réglementation, l'IA devient non seulement un miroir des inégalités existantes, mais un puissant mécanisme d'amplification de la discrimination, de l'exclusion et de la violence à l'égard des femmes et des filles . Elle a présenté des données probantes étendues : 44 % des systèmes d'IA évalués présentent des biais de genre parce qu'ils sont entraînés sur des données qui reflètent déjà des biais existants ; près d'une femme sur quatre parmi les défenseures des droits de l'homme, les militantes et les journalistes interrogées avait subi des violences en ligne assistées par l'IA ; et jusqu'à 99 % des deepfakes et des images sexuelles manipulées en ligne ciblent des femmes . Elle a noté que 41 % des femmes interrogées s'autocensurent sur les réseaux sociaux pour éviter les abus, ce qui pousse les femmes et les filles hors des espaces en ligne et menace la participation démocratique . S'appuyant sur le rapport du panel scientifique indépendant de la veille, elle a noté que 88 % des principaux chercheurs en IA sont des hommes et que les femmes ne représentent que 30 % de la main-d'œuvre mondiale dans le domaine de l'IA . Elle a également noté que sur les 138 pays évalués, seuls 24 font référence au genre dans leurs stratégies nationales d'IA, et seulement 18 incluent des dispositions substantielles tenant compte du genre . Elle a soutenu que les États demeurent les principaux garants des obligations au titre du droit international des droits de l'homme, notamment la Déclaration universelle, le PIDCP et la CEDAW, et que les entreprises technologiques ne peuvent pas se cacher derrière la complexité de leurs systèmes . Elle a appelé à des évaluations d'impact obligatoires sur les droits de l'homme avant et après le déploiement, avec l'égalité des genres traitée comme non négociable, et à ce que les femmes et les filles, les technologues féministes, les communautés autochtones, les organisations syndicales, les défenseurs des droits des personnes handicapées et la société civile soient véritablement impliqués et habilités . Elle a présenté ses questions de responsabilité - qui construit l'IA, qui en bénéficie, qui en supporte les risques et qui a voix au chapitre - comme des éléments constitutifs de la responsabilité .

Amal El Fallah Seghrouchni, ministre déléguée du Maroc, a introduit une distinction conceptuellement originale entre les préjudices exogènes - qui sont externes aux êtres humains et peuvent être observés et mesurés - et les préjudices endogènes, qui se produisent dans l'esprit humain et sont imperceptibles même pour ceux qui les vivent . Elle a soutenu que lorsque des algorithmes manipulent la psychologie d'un enfant, cela ne peut être ni mesuré ni même perçu, et a demandé comment la société peut auditer ses propres neurones ou comprendre que le cerveau humain a été transformé . Elle a mentionné la tentative législative précoce du Chili d'introduire des droits cognitifs comme une réponse visionnaire à ce défi . Elle a également noté que même les chercheurs en IA ne disposent pas toujours des outils nécessaires pour comprendre la complexité des algorithmes, et que les grands modèles de langage comportant entre 175 et 180 milliards de paramètres rendent la traçabilité de la responsabilité extrêmement difficile .

Dans sa deuxième intervention, El Fallah Seghrouchni a identifié la responsabilité dans les systèmes d'IA agentique comme probablement le défi juridique le plus important de la décennie, notant que dans un réseau d'agents individuellement autonomes, il n'y a pas d'être humain à l'origine de la décision finale, ce qui rend difficile de déterminer quel agent est responsable et comment la responsabilité doit être répartie . Elle a illustré l'ampleur du défi par un exemple concret tiré de la plateforme marocaine Iderati X.0, qui traite 52 millions de transactions administratives par jour, notant qu'à une telle échelle, il devient extrêmement difficile d'identifier la cause exacte lorsqu'un problème survient . Elle a proposé trois obligations concrètes : une documentation structurée des chaînes d'agents précisant quel agent fait quoi et quand ; la désignation d'un être humain identifiable responsable de toutes les questions liées aux droits lorsque des agents sont déployés dans les services publics ; et un droit à un recours en temps utile afin que les êtres humains puissent obtenir réparation à la mesure de la rapidité des systèmes d'IA .

Samuel Arias Arzeno, juge de la Première Chambre de la Cour suprême de justice de la République dominicaine, a soutenu que le véritable défi de la gouvernance ne se pose pas lorsque des lois ou des principes éthiques sont adoptés, mais lorsqu'une personne estime qu'un système d'IA a violé ses droits de l'homme - c'est à ce moment que la gouvernance devient une réalité . Il a soutenu que les systèmes judiciaires doivent être considérés non pas simplement comme des utilisateurs de l'IA, mais comme des institutions essentielles pour convertir les principes des droits de l'homme en véritables droits de l'homme, et qu'aucune institution publique ne peut justifier une décision en disant que c'est ce que le système lui a indiqué de faire . Il a noté que, par le biais de la Justice Action Coalition, des gouvernements, des organisations civiles, des organisations internationales et des systèmes judiciaires travaillent ensemble pour s'assurer que la gouvernance de l'IA ne se limite pas à la conception de politiques publiques, mais peut être rendue effective par des systèmes judiciaires fondés sur les personnes, et que la République dominicaine co-préside avec la Banque mondiale les travaux sur les technologies émergentes et l'IA dans le domaine de la justice . Il a proposé quatre priorités : renforcer les capacités institutionnelles des avocats et des jurés ; assurer la transparence et la traçabilité dans l'utilisation de l'IA ; maintenir une supervision humaine effective ; et garantir que toutes les personnes disposent de mécanismes effectifs pour accéder à la justice et obtenir réparation .

Sonia Livingstone, membre du Panel scientifique international indépendant sur l'IA, a rapporté que le panel a constaté que les preuves de violations des droits de l'homme sont actuellement bien plus convaincantes que les preuves de bénéfices pour les droits de l'homme dans de nombreux contextes où l'IA est utilisée . Le panel a documenté des menaces pour l'écosystème de l'information à travers la persuasion et la tromperie générées par l'IA, la désinformation et la méfiance, avec des impacts négatifs sur la santé collective, la participation démocratique et l'intégrité électorale . Il a également documenté des menaces pour la vie privée et la sécurité, les données personnelles étant collectées, manipulées, utilisées abusivement et exploitées, ainsi que des preuves de préjudices en termes d'inégalité et d'injustice, le pouvoir de l'IA étant concentré dans très peu d'entreprises, de pays et de langues . Le panel a identifié quatre catégories de risques : les risques directs ; les risques découlant d'une capacité mondiale très inégale ; les coûts d'opportunité liés à l'incapacité à réaliser les bénéfices de l'IA ; et les risques liés au déploiement sans garanties fondées sur les droits .

Sur la question des enfants spécifiquement, Livingstone a présenté des données alarmantes : dans 11 pays du Sud global, jusqu'à un enfant par classe a signalé que l'IA avait été utilisée pour créer des deepfakes sexuellement explicites les représentant ; les signalements de matériel d'abus sexuel sur enfants généré par l'IA auprès de la cyber tip line américaine NCMEC augmentent de façon exponentielle ; du matériel d'abus sexuel sur enfants est documenté dans certains ensembles de données d'entraînement ; et les auteurs d'infractions peuvent désormais affiner des modèles ouverts à partir d'images d'exploitation sexuelle d'enfants . Une enquête nationale américaine a révélé qu'un enfant sur trois avait discuté de questions émotionnelles importantes avec un compagnon IA, alors que l'évaluation des risques de Stanford concernant les compagnons IA a conclu que ces produits ne satisfont pas aux normes de sécurité de base . Elle a renvoyé les participants aux appels de M. Kateris invitant les nations à adopter un engagement pour la sécurité des enfants dans le domaine de l'IA . Elle a mis en garde contre le fait que le problème ne sera pas résolu en restreignant les enfants, et a souligné les droits des enfants à l'expression, à la participation, à l'éducation et à l'information à l'ère numérique .

Sasha Luccioni, co-fondatrice et directrice scientifique du Sustainable AI Group, a soutenu que la transparence sur l'empreinte environnementale de l'IA est un droit humain fondamental, permettant aux utilisateurs de faire des choix éclairés entre les outils . Elle a noté que si les bénéfices de l'IA peuvent techniquement être mondiaux, les impacts négatifs en termes d'eau, d'énergie, d'émissions et de santé sont très locaux et affectent de manière disproportionnée les populations les plus marginalisées . Elle a décrit la difficulté d'obtenir de la transparence même pour des organisations internationales telles que l'AIE, qui peinent à obtenir des informations parce que les pays membres eux-mêmes ne les possèdent pas . Elle a présenté son AI Energy Score Project - des étiquettes d'efficacité énergétique pour les modèles d'IA - comme une étape vers une prise de décision éclairée et une réglementation éventuelle . Dans sa deuxième intervention, elle a averti que l'IA agentique amplifie les degrés de séparation entre un utilisateur et la technologie sous-jacente, rendant de plus en plus difficile l'identification de l'origine des défaillances .

Jhalak M. Kakkar, directrice exécutive du Centre for Communication Governance de la National Law University de Delhi, a soutenu que les cadres existants des droits de l'homme - notamment le droit à l'égalité, à la vie privée et à la liberté d'expression - doivent être réinterprétés dans le contexte de l'IA . Elle a soulevé la question de ce que signifie la liberté de pensée lorsque l'IA menace l'autonomie cognitive par le micro-ciblage et la manipulation , et a noté que la concentration des centres de données et l'extraction de minéraux principalement dans des zones déjà marquées par des inégalités dans le Sud global soulèvent des questions fondamentales de justice environnementale et économique . Elle a soutenu que les systèmes judiciaires devront faire face à la complexité de l'établissement de la vérité à une époque d'érosion épistémique, combler les lacunes en matière de responsabilité, naviguer dans les questions de transparence et remédier au fossé de gouvernance transnational et transfrontalier où les lois sont nationales mais les entreprises sont mondiales . Elle a conclu que sans diffusion de la concentration du pouvoir dans une poignée d'entreprises relevant de quelques juridictions, les droits de l'homme - en particulier pour les populations du Sud global - ne peuvent être protégés .

#

Résultats du sondage auprès des participants

Peggy Hicks a présenté les résultats d'un sondage réalisé auprès des participants à la session. Une pluralité de répondants s'est déclarée préoccupée par l'insuffisance des mesures de gouvernance en place pour faire face aux défis et aux opportunités de l'IA . Les deux principales préoccupations en matière de droits de l'homme identifiées étaient la responsabilité et l'état de droit, ainsi que l'impact de l'IA sur les enfants, suivis par la surveillance et l'utilisation de l'IA par les forces de l'ordre, puis la durabilité environnementale . Sur la question de la responsabilité pour les actions de l'IA agentique, les répondants ont massivement désigné les entreprises qui développent le modèle d'IA comme la principale partie responsable, suivies des États, puis des organisations qui déploient le système - un résultat cohérent avec les Principes directeurs des Nations Unies relatifs aux entreprises et aux droits de l'homme . Hicks a également noté que le HCDH et l'UIT présenteraient jeudi une enquête intitulée « Moi et l'IA », menée auprès de 1 000 enfants dans 49 pays, dans laquelle les enfants ont exprimé leur souhait que l'IA soit façonnée avec eux et pas seulement pour eux, qu'elle respecte leurs droits, et qu'ils ne souhaitent pas simplement en être protégés .

#

Deuxième panel : solutions, données probantes et inclusion

Le deuxième panel était modéré par Son Excellence Clara Chappaz, Ambassadrice pour l'IA et les affaires numériques de France, qui a cadré la discussion autour de trois questions : quelles bonnes pratiques existent en matière de diligence raisonnable relative aux droits de l'homme ; quels outils techniques et données probantes sont nécessaires pour comprendre les impacts avant le déploiement ; et comment garantir que les enfants et les autres populations concernées sont véritablement associés à la gouvernance de l'IA .

Willemijn Aerdts, ministre de l'Économie numérique et de la Souveraineté des Pays-Bas, a présenté des outils nationaux concrets, notamment un cadre algorithmique traduisant les exigences légales en lignes directrices pratiques, une évaluation d'impact sur les droits fondamentaux et les algorithmes, et un registre des algorithmes couvrant tous les algorithmes utilisés par le gouvernement néerlandais . Sur la question des enfants, elle a souligné que lors de l'élaboration de politiques pour les enfants et les jeunes, ceux-ci doivent être véritablement inclus dans le processus décisionnel aux côtés des parents et des éducateurs, et a noté que les jeunes sont souvent confrontés à des problèmes trop importants pour les parents mais trop mineurs pour la police .

Felipe Paullier, Secrétaire général adjoint du Bureau de l'ONU pour la jeunesse, a soutenu que la question ne devrait pas seulement porter sur ce que l'IA peut faire, mais sur ce que l'IA peut aider l'humanité à construire comme avenir, et que l'IA doit être évaluée non seulement sur la vitesse et l'efficacité, mais sur sa contribution au bien-être humain, à la dignité et aux droits de l'homme . Il a souligné que les jeunes sont les utilisateurs les plus actifs de l'IA et parmi les innovateurs les plus fréquents au sein des entreprises d'IA, et pourtant ils ne sont presque jamais les décideurs, bien qu'ils vivent le plus longtemps avec les conséquences des décisions prises aujourd'hui . Il a appelé à la création d'espaces significatifs permettant aux jeunes de participer à la gouvernance de l'IA au niveau national, avertissant que si de tels espaces ne sont pas créés, les jeunes en créeront eux-mêmes, comme cela s'est produit dans le domaine climatique .

Ulises Gutiérrez, Représentant spécial pour les technologies émergentes du Mexique, a soulevé la préoccupation selon laquelle les politiques publiques sont toujours en retard sur le développement technologique, et qu'il existe une compréhension politique insuffisante de l'impact du développement technologique, en particulier de l'IA . Il a soutenu que la question n'est pas ce qu'un être humain peut faire avec la technologie, mais ce que la technologie fait aux êtres humains . Il a appelé à un nouveau contrat social aux niveaux national et international, et pas seulement à une réglementation, pour gouverner l'IA dans un paysage en constante évolution .

Wanjin Park, vice-président de KT, a décrit l'approche de son entreprise consistant à définir les risques liés à l'IA en incluant les considérations relatives aux droits de l'homme, à évaluer les modèles et agents d'IA par rapport à ce cadre, et à faire fonctionner un comité de sécurité de déploiement au niveau de la direction avant de lancer des produits d'IA . Il a particulièrement mis en avant la taxonomie des risques liés à l'IA du projet BTEC du HCDH comme un travail important qui doit être développé de manière plus concrète et pratique . Il a noté que le problème se pose lorsque des systèmes multi-agents provenant de différentes entreprises sont interconnectés, chaque entreprise ayant sa propre définition du risque lié à l'IA, et a appelé à des normes communes et bien conçues pour créer une base partagée .

Nighat Dad, fondatrice de la Digital Rights Foundation, a livré l'une des interventions analytiquement les plus percutantes de la session, soutenant que la diligence raisonnable en matière de droits de l'homme telle que la pratiquent la plupart des entreprises d'IA aujourd'hui ne satisfait pas aux normes des Principes directeurs des Nations Unies relatifs aux entreprises et aux droits de l'homme - elle intervient après que les décisions de co-conception ont été prises, là où la réglementation l'impose, et sans la participation des personnes réellement concernées . Elle a qualifié l'inégalité géographique de la diligence raisonnable - des évaluations significatives menées principalement dans l'UE, avec les mêmes systèmes déployés dans le Sud global sans examen équivalent - de choix structurel, et non d'écart de capacité, créant un régime de droits à deux vitesses déterminé par la géographie . Elle a documenté plus de 23 000 cas de violence facilitée par la technologie affectant de manière disproportionnée les femmes, les filles et les jeunes par le biais de la ligne d'assistance téléphonique contre le harcèlement en ligne de son organisation , et a soutenu que tout cadre de diligence raisonnable qui ne traite pas explicitement de la violence de genre facilitée par la technologie a exclu la moitié de la population de sa protection . Elle a proposé trois recommandations concrètes : des évaluations obligatoires d'impact sur les droits de genre et de l'enfant menées avant le déploiement avec la participation des communautés concernées ; des obligations de diligence raisonnable uniformes sur tous les marchés où un système est déployé ; et des évaluations répétées chaque fois que les capacités changent de manière significative, y compris les augmentations de la fonction agentique .

Alvitta Ottley, membre du Panel scientifique indépendant sur l'IA, a identifié deux défis scientifiques connexes : le décalage d'évaluation et le manque de données probantes . Sur le décalage d'évaluation, elle a soutenu que les questions d'ingénierie - peut-on rendre quelque chose plus rapide, plus précis, plus performant ? - ne sont pas les mêmes questions que celles que posent les sociétés, qui concernent la protection des droits de l'homme, la responsabilité des institutions et l'amélioration de la société . Elle a soutenu que si la société décide que le succès consiste à protéger les droits de l'homme, à favoriser une confiance appropriée et à préserver la responsabilité, alors les résultats mesurés doivent correspondre à ce qui est optimisé . Sur le manque de données probantes, elle a noté que si les preuves sur les performances des modèles d'IA sont abondantes, il existe très peu de preuves sur la façon dont l'IA affecte la prise de décision des personnes, l'apprentissage des enfants, les décisions cliniques ou le travail des agents publics, et que répondre à ces questions nécessite des études longitudinales, des collaborations interdisciplinaires et un travail minutieux avec des communautés protégées et mal desservies . Elle a conclu par une formulation qui a résonné tout au long du panel : la société doit décider à quoi ressemble le succès, et la science doit déterminer comment le mesurer .

Dans ses remarques de clôture, Chappaz a synthétisé les conclusions du panel, notant que la rapidité du développement technologique dans un monde géopolitiquement fragmenté exige de définir ce que le succès signifie pour les êtres humains, sur la base des valeurs qu'ils choisissent de privilégier, et que c'est uniquement la décision de l'humanité, pas celle de la technologie . Elle a repris le cadrage de Dad selon lequel l'inégalité géographique de la diligence raisonnable représente des choix structurels et non des lacunes de capacité, et a averti que si cela n'est pas résolu, les communautés se retireront des espaces numériques ou en créeront de nouveaux sans gouvernance ni valeurs .

#

Interventions du public

Le segment des interventions du public a accueilli des contributions de gouvernements et d'organisations de la société civile de plusieurs régions, modéré par les co-présidents thématiques.

La ministre irlandaise Niamh Smyth a soutenu que la gouvernance mondiale de l'IA doit s'aligner sur les cadres internationaux des droits de l'homme, en promouvant la liberté d'expression, la vie privée, l'accès à l'information et l'égalité des genres, tout en luttant activement contre toutes les formes de violence, y compris la violence sexuelle et de genre . Elle a exprimé une préoccupation particulière quant au fait que, sans les garanties nécessaires en matière de droits de l'homme et de supervision, les systèmes d'IA permettent la violence de genre facilitée par la technologie, et que la modération des contenus en ligne par l'IA sans supervision humaine peut amplifier la désinformation, affectant de manière disproportionnée les femmes et les filles .

Robert Baruch d'Universal Music Group a ancré la position de l'industrie musicale dans le droit international des droits de l'homme, citant l'article 27 de la Déclaration universelle des droits de l'homme et l'article 15 du PIDESC comme protégeant les intérêts moraux et matériels des créateurs à l'ère de l'IA . Il a soutenu que l'avenir de l'IA doit être construit non pas en choisissant entre l'innovation et les droits des créateurs, mais en faisant progresser les deux ensemble, et a proposé trois piliers : le contrôle des créateurs sur la façon dont leurs œuvres sont utilisées pour développer des systèmes d'IA ; la transparence sur le moment et la manière dont les œuvres sont utilisées ; et la supervision humaine garantissant que l'IA amplifie plutôt que remplace la créativité humaine .

La représentante de la Slovénie, Anita Pipan, a soutenu que la confiance dans l'IA n'est pas seulement un objectif éthique mais une condition préalable à une adoption plus large de l'IA, et a décrit la stratégie nationale d'IA 2030 de la Slovénie comme promouvant une approche centrée sur l'être humain, ancrée dans les droits de l'homme, les valeurs démocratiques, l'éthique, la sécurité, la transparence et la responsabilité . Elle a accordé une importance particulière à la protection de la diversité linguistique et culturelle, soutenant qu'une IA inclusive devrait refléter et préserver toutes les langues et cultures .

Zachary Lampel de l'International Center for Not-for-Profit Law a proposé trois actions concrètes : des lignes rouges pour l'IA - des interdictions claires ou des moratoires sur les systèmes et utilisations de l'IA qui présentent des risques inacceptables pour les droits de l'homme, explicitement prévus dans la résolution de l'Assemblée générale des Nations Unies de 2024 intitulée « Saisir les opportunités offertes par des systèmes d'intelligence artificielle sûrs, sécurisés et dignes de confiance pour le développement durable » ; des certifications standardisées liées aux règles nationales de passation de marchés publics, analogues aux certifications de construction écologique, créant des incitations de marché pour concevoir et déployer une IA sûre ; et un droit à réparation, garantissant que les décisions prises via des systèmes d'IA ont un acteur responsable et une voie de recours . Il a également noté que les recommandations de l'UNESCO sur l'éthique de l'intelligence artificielle excluent déjà explicitement l'utilisation de l'IA pour la notation sociale et la surveillance de masse .

La représentante de l'Uruguay, Alejandra De Bellis Bonilla, a mis en avant le Partenariat mondial pour les droits de l'homme comme une plateforme multi-parties prenantes large reliant les efforts sur les impacts transversaux de l'IA, et a appelé à des améliorations en matière de responsabilité, de cadres normatifs, de durabilité et de financement, avec une approche tenant compte du genre dans la conception et le déploiement de l'IA .

Franco Giandana Gigena d'AccessNow a appelé à des normes juridiques contraignantes centrées sur les droits de l'homme et à des mécanismes de gouvernance mondiale pour réguler les technologies de surveillance numérique et l'IA, en s'attaquant aux effets dissuasifs sur les manifestants, les journalistes et les défenseurs des droits de l'homme . Il a appelé à un arrêt du développement et de l'utilisation des outils d'IA facilitant les violations du droit international humanitaire et du droit des droits de l'homme, en particulier dans les contextes de conflit . Il a soutenu que les populations du monde entier n'ont pas seulement besoin d'accéder à l'IA, mais surtout d'agentivité .

Rafał Kownacki du ministère polonais des Affaires numériques a soutenu que lorsque la société décide de ce qu'un algorithme peut faire à un être humain, elle décide de ce qu'elle croit qu'une personne vaut . Il a invoqué l'expérience historique de la Pologne pour soutenir que certains choix ne doivent jamais être faits par une machine seule, et que quelle que soit la capacité des systèmes d'IA, une personne doit garder le dernier mot - pour comprendre la décision, la contester et, là où cela importe le plus, éteindre les systèmes . Il a salué la Convention-cadre du Conseil de l'Europe sur l'IA comme le premier traité contraignant plaçant l'IA dans l'ordre des droits de l'homme, de la démocratie et de l'état de droit .

Isabella Henriques de l'Instituto Alana, s'exprimant au nom de plus de 120 signataires d'une déclaration commune de la société civile sur les droits des enfants lors du dialogue sur l'IA, a soutenu que les enfants sont parmi les premiers et les plus fréquents utilisateurs des nouvelles technologies, mais restent largement absents des stratégies nationales d'IA et des cadres de gouvernance . Elle a souligné que les droits, la sécurité et le bien-être des enfants doivent être intégrés dans les systèmes d'IA dès le départ, et non ajoutés après que le préjudice s'est déjà produit, et que la responsabilité doit incomber à ceux qui construisent et profitent de la technologie .

João Brant du Brésil a décrit le Statut numérique pour les enfants et les adolescents du pays, qui exige des plateformes qu'elles adoptent des mesures de protection de l'enfance dès la phase de conception, limite les fonctionnalités pouvant conduire à une utilisation excessive telles que le défilement infini, et interdit la publicité ciblée basée sur les données personnelles des enfants . Il a également décrit un décret interdisant la création d'images intimes synthétiques basées sur de vraies tierces parties, établissant des responsabilités proactives pour les plateformes et les entreprises d'IA . Il a appelé à ce que l'intégrité de l'information soit comprise dans sa dimension collective ou sociale comme faisant partie du droit d'accès à une information exacte et fiable, et a soulevé la préoccupation concernant l'impact de l'IA sur la viabilité économique du journalisme .

Anna Osterling du Global Forum for Media Development a soutenu, citant Courtney Raj du Centre for Media and Digital Governance d'OpenMarkets, que le journalisme est une espèce clé de voûte de l'écosystème de l'information - stabilisant la confiance, ancrant la vérification, structurant la responsabilité et permettant à d'autres institutions de fonctionner - et que sa disparition entraîne l'effondrement de l'écosystème de l'information plutôt que son adaptation . Elle a noté que les systèmes d'IA sont entraînés sur des contenus journalistiques récupérés sans consentement, compensation ni crédit, et a appelé à une infrastructure numérique publique, à des politiques industrielles soutenant l'IA d'intérêt public, et à une gouvernance de l'IA exigeant que les droits de l'homme soient intégrés tout au long du cycle de vie complet de tous les systèmes d'IA .

Jihoon Cha de la République de Corée a décrit la loi fondamentale sur l'IA du pays, qui exige des opérateurs de systèmes d'IA à fort impact qu'ils évaluent les impacts potentiels sur les droits fondamentaux, et a déclaré la vision de la Corée d'établir le Global AI Hub, avec neuf organisations participantes dans le domaine de l'IA et des domaines connexes, pour renforcer les capacités mondiales en matière d'IA . Il a soutenu que l'accès à l'IA devrait être traité non pas comme un privilège à mériter, mais comme un droit fondamental à exercer .

Pria Chetty de Research ICT Africa a proposé un cadre « Just AI » (IA juste), soutenant que les modèles de gouvernance dominants ancrés dans une IA éthique ou responsable restent structurellement inadéquats parce qu'ils sont largement autorégulés, pilotés par des acteurs dominants, calibrés pour les pays à revenus élevés et ignorent la souveraineté ou l'agentivité . Elle a appelé à une reconnaissance mondiale que la gouvernance de l'IA est inséparable de la gouvernance des données, et à la démocratisation des ressources et des capacités d'IA afin que les nations de la majorité mondiale puissent passer de consommateurs passifs à co-créateurs d'IA .

Jérôme Bellion-Jourdan de l'Institute for Global Negotiation, également vice-président du comité directeur de l'AI for Good Impact Initiative, a soutenu que si un large accord existe sur la nécessité d'une meilleure gouvernance, la question est de savoir comment y parvenir. Il a proposé que les outils et techniques de la négociation multilatérale - notamment la conception de processus, les formats de négociation innovants et le passage du marchandage positionnel et des dynamiques gagnant-perdant à la négociation intégrative et aux résultats gagnant-gagnant - pourraient aider à naviguer dans les jeux de pouvoir et les asymétries de pouvoir qui entravent les progrès vers une gouvernance mondiale de l'IA .

Mark Cassayre de l'IDLO, la seule organisation intergouvernementale mondiale exclusivement consacrée à la promotion de l'état de droit, a recommandé que les droits de l'homme et l'état de droit doivent sous-tendre le développement et le déploiement des technologies numériques dans le secteur de la justice, que des cadres juridiques et réglementaires adaptés doivent accompagner l'innovation, et que l'investissement dans les capacités institutionnelles - notamment la culture de l'IA parmi les professionnels du droit et de la justice - est essentiel .

Rebecca Rektimbo, coordinatrice de projets techniques de connectivité pour LOCNET, s'exprimant au nom de Raman Jit Singh Chima, directeur mondial du programme de l'Association for Progressive Communications, a soutenu qu'avant que les communautés puissent contribuer des données, construire des solutions d'IA locales, influencer les politiques ou bénéficier de l'innovation, elles ont besoin d'une connectivité significative, abordable, fiable et localement pertinente . Elle a soutenu que la connectivité centrée sur les communautés rapproche le pouvoir des communautés et permet aux femmes, aux peuples autochtones, aux communautés rurales et aux locuteurs de langues sous-représentées de devenir des créateurs, des innovateurs, des chercheurs et des décideurs plutôt que de simples consommateurs de technologie . Elle a conclu par une formulation qui a résonné tout au long de la session : les communautés ne sont pas le dernier kilomètre de l'IA - elles en sont le premier kilomètre .

Esther Eghobamien-Mshelia du Comité CEDAW a soutenu qu'il ne peut y avoir de gouvernance de l'IA véritablement fondée sur les droits de l'homme sans la pleine mise en œuvre de la CEDAW, qui exige des États non seulement d'interdire la discrimination, mais d'atteindre l'égalité substantielle en s'attaquant aux obstacles structurels et réglementaires, aux stéréotypes et aux rapports de pouvoir inégaux . Elle a cité la Recommandation générale n° 40 de la CEDAW sur la participation pleine et égale des femmes à la vie politique et publique comme un outil nous rappelant que les femmes doivent participer pleinement et à égalité à la conception, à la gouvernance et à la supervision des systèmes d'IA . Elle a appelé à des indicateurs tenant compte du genre, à un suivi et à une surveillance pour démontrer si les femmes participent et bénéficient à égalité de la transformation numérique et sont protégées contre des préjudices tels que la discrimination algorithmique et la violence de genre facilitée par la technologie .

Elizabeth Tan du HCR a souligné que pour les personnes déplacées de force et les apatrides, les obstacles à la contestation des décisions assistées par l'IA sont particulièrement élevés en raison des lacunes documentaires, des barrières linguistiques, des limitations d'accès numérique et de la crainte d'approcher les autorités . Elle a cité des exemples concrets de Libye, où des contenus générés par l'IA ont incité à l'hostilité et aux abus envers les réfugiés, et de la communauté rohingya, qui continue de faire face à des discours déshumanisants en ligne même en exil . Elle a appelé à une diligence raisonnable en matière de droits de l'homme tout au long du cycle de vie de l'IA, à des tests de biais, à des explications claires sur la façon dont les décisions sont prises, et à des moyens accessibles de se plaindre ou de demander une correction .

Sopio Kiladze du Comité des droits de l'enfant a décrit la déclaration commune des Nations Unies sur l'IA et les droits de l'enfant, co-dirigée par l'UIT, le CRC et l'UNICEF, co-signée par 13 agences des Nations Unies et plus de 60 organisations, et élaborée avec la contribution d'enfants des cinq régions de l'ONU . Elle a soutenu que l'histoire ne jugera pas la société à l'aune de l'intelligence de son IA, mais à celle de sa capacité à avoir utilisé ce moment extraordinaire pour protéger ceux qui ont la plus petite voix et le plus grand enjeu dans l'avenir .

Shumaila Hussaini Shahani du Tech Global Institute a proposé cinq engagements concrets : les affirmations de sécurité doivent être accompagnées des conditions dans lesquelles les systèmes ont été testés et des taux d'échec ventilés par langue, genre, handicap et région ; une décennie d'engagements volontaires n'a pas produit de résultats et des cadres de responsabilité publique assortis de responsabilité des entreprises sont nécessaires ; une capacité financée publiquement et hébergée régionalement pour évaluer les systèmes d'IA avant le déploiement est requise ; la transparence et la supervision doivent s'étendre à la gestion algorithmique et aux travailleurs des données ; et les connaissances autochtones et traditionnelles doivent être protégées par une gestion des données dirigée par les communautés .

#

Remarques de clôture et thèmes transversaux

Dans ses remarques de clôture, Linda Bonyo a soulevé deux points critiques : premièrement, qui finance les droits de l'homme à l'ère de l'IA, notant que les ressources nécessaires pour garantir que les droits de l'homme fonctionnent idéalement dans ce contexte doivent être trouvées ; et deuxièmement, la nécessité d'une ouverture algorithmique autour des processus de visa, notant que 51 % des conversations sur la gouvernance de l'IA se déroulent à Genève et que le reste du monde en est effectivement exclu parce que les gens ne peuvent pas obtenir de visas - un processus lui-même façonné par des algorithmes opaques . Elle a nommé un participant absent spécifique, Freedom Wangi, un travailleur de l'IA incapable d'obtenir un visa, comme illustration concrète de cette exclusion .

Le ministre López Águeda a conclu en reconnaissant que le dialogue ne porte pas seulement sur la technologie, mais sur la démocratie, les droits de l'homme, la vie privée, l'énergie et la paix dans le monde . Il a déclaré clairement que la société est en retard pour relever ces défis, mais a affirmé la direction à suivre : construire une IA humaniste, défendre l'agentivité humaine et utiliser l'IA pour être meilleurs plutôt que pires .

#

Évaluation globale

La session a révélé un degré élevé de consensus normatif parmi un éventail remarquablement diversifié de participants - des ministres gouvernementaux et des responsables de l'ONU aux représentants de la société civile, aux universitaires, aux acteurs du secteur privé et aux représentants d'organisations internationales - sur les principes fondateurs de la gouvernance de l'IA fondée sur les droits de l'homme. Tout au long des discussions de la session, les domaines de convergence les plus forts étaient que la gouvernance de l'IA doit être ancrée dans le droit international contraignant des droits de l'homme plutôt que dans une éthique volontaire ; que la réglementation et l'innovation sont complémentaires plutôt qu'en tension ; que la supervision humaine significative doit être substantielle et non nominale ; que la transparence est une exigence fondamentale dans toutes les dimensions de l'IA ; que les enfants nécessitent une protection spéciale, la responsabilité incombant aux développeurs et aux déployeurs ; que l'IA amplifie les inégalités existantes et nécessite des réponses de gouvernance proactives ; que l'IA agentique crée des défis de responsabilité sans précédent ; et que les communautés concernées doivent être véritablement incluses dans les processus de gouvernance .

Cependant, le consensus sur les principes n'était pas assorti d'un accord équivalent sur les mécanismes d'application, le financement ou l'architecture institutionnelle spécifique nécessaire pour traduire les principes en pratique. Les intervenants du Sud global ont constamment souligné que le fossé entre principe et pratique est lui-même une inégalité structurelle, la diligence raisonnable significative étant appliquée principalement dans les juridictions à revenus élevés . Les résultats du sondage de la session, montrant qu'une pluralité de participants s'inquiétait de l'insuffisance des mesures de gouvernance en place , et que la responsabilité, l'état de droit et l'impact de l'IA sur les enfants étaient les principales préoccupations , ont confirmé que les participants eux-mêmes reconnaissent la distance entre le consensus normatif atteint et la réalité de la mise en œuvre. Le défi de combler cette distance - de manière équitable, urgente et avec la pleine participation de ceux qui sont les plus touchés - est apparu comme la tâche déterminante à venir.

Peggy Hicks
We are looking at a pretty full room, and I'm sure there will be others coming in, but we have a very tight schedule that I'm in charge of keeping you on. My name is Peggy Hicks. I'm the Director of the Thematic and Special Procedures Division at the Office of the High Commissioner for Human Rights, and it's my pleasure to welcome you to the thematic cluster focusing on respecting, protecting, and promoting human rights, transparency, accountability, and human oversight. You're going to hear today from a diverse range of experts on AI governance from the lens of human rights. OHCHR and the World Bank have been the UN system co -leads in conceptualizing and supporting this session, and I welcome you on behalf of both of our organizations. This dialogue under the leadership of the co -chairs, Ambassador Egriselda López of El Salvador and Ambassador Rein Tammsaar of Estonia, is a remarkable achievement of how the UN system is meeting today's moment, bringing us to the end of the session. Thank you. together to answer fundamental questions on our shared governance. So with that, I'd like to give the floor to Jeremy Ng of the World Bank to introduce himself and the High Commission. Thank you.
Jeremy Ng
Thank you very much, Peggy. Excellencies, colleagues, it's a real pleasure to be here today for this really timely and important dialogue. I want to thank OHCHR, our co -chairs, for really doing a lot of the heavy lifting preparing this session. So thank you so much to Li Zhou in particular, my colleague and counterpart at OHCHR for his really, really massive efforts on this. Thank you so much. Without further ado, I'll pass it directly to Volker Turk, the High Commissioner for Human Rights, for some opening remarks. Thank you.
Volker Turk
Ministers, excellencies, distinguished participants, before a new drug reaches the market, it undergoes years of clinical trials, regulatory changes. We review and market authorization. And this process can last between 10 and 15 years. Many drugs fail due to lack of effectiveness or safety concerns. And even with all these safeguards, mistakes can still happen. Are we taking the same care with artificial intelligence? AI is already transforming societies. It's being developed and deployed at warp speed. It is fueling mass surveillance, disinformation on social media, and enabling bias, including from a gender perspective. Harm to children from AI -powered algorithms is growing. Data centers, housing AI servers are choking our environment. At the same time, AI holds remarkable potential. It is already helping to support medical diagnosis, accelerated scientific research, strengthen public services, and more. So this is precisely why we must ensure people can trust AI. So we need to ask ourselves, why is regulation around the public good even up for debate? We have been here before. We lost precious years in addressing climate change because powerful players cast doubt on science, delayed action, and prioritized short -term economic and political interests. Today, we are asked to accept on faith that AI systems will improve our lives rather than question how they work, who benefits, who loses, and who is accountable. Critical thinking is cast aside with the greatest of ease. It reminds me of the protagonist, Kay. In Franz Kafka's book, The Castle. K arrives in a new village as a land surveyor and spends the rest of the novel trying to understand a system that remains inaccessible and unaccountable are we like K trapped in a system of opaque power in which we are losing our agency and AI is about power over data, markets, resources and information but to what end? if you ask people around the world what they want from power the answers are the same peace, safety, a decent standard of living in short, they want their human rights which are a check on the exercise of power human rights help us reclaim our agency and have a say in decisions that affect us to take back our rights to take back our agency we need guardrails on the development of AI to ensure transparency, inclusion and accountability. Political leaders who exercise power on our behalf can ensure this happens by regulating in the public interest. The most obvious solution, regulation guided by human rights, is right in front of our eyes. But somehow we keep skirting around it. But there is an opening. Concerns about AI's impact on jobs and the economy are becoming more credible. Momentum is building for oversight of the growing security risks posed by powerful AI systems, including their use in cyber operations and armed conflict. Around the world, especially young people, are demanding actions. AI is not just about technology. It is about equality and accountability. It touches all areas of our lives, from jobs to crossing a border to accessing information. That is why human rights needs to be embedded into the design, development, deployment and use of AI. International human rights law, and let's never forget about this, international human rights law constitutes a binding legal framework to protect data, to avoid discrimination, to access justice, to ensure equality and so much more. Human rights due diligence and impact assessments of AI systems are essential, and so are robust data protection and privacy safeguards. Some areas need specific attention, for example, protecting children online. And I encourage you to draw on my office's guidelines that we issued last month. Every country is involved in the AI value chain. whether through data, critical minerals, labor, markets, compute, or cloud capacity. And that means that all should have a meaningful stake in shaping and benefiting from AI. Otherwise, AI will deepen inequalities rather than help close them. It is also about directing AI investment towards the building blocks of resilient societies, including health, education, social protection, and climate action. Decisions around AI need to be transparent about how data is being used and where people can seek justice. When AI gathers data or carries out a decision without human agency, responsibility can disappear into the system. And this is increasingly the case as AI agents are deployed in healthcare, finance, recruitment, and development. And more. We all know the frustrations of dealing with automated systems when we want to change a doctor's appointment. We get caught in a human -free vortex with nowhere to turn. So now imagine that level of unaccountability around life -altering issues, whether you are hired, whether you can get credit, whether you are returned to a country where you could face torture. The stakes do not end there. As AI is increasingly integrated into military systems, automated decisions can mean life or death. Human oversight of AI systems cannot be just a rubber stamp. It requires that an identified person be granted authority, competence, time, independence and power to alter and even stop a system. Distinguished participants, I asked a question, earlier, why regulation of AI systems? Why regulation of AI systems? is even up for debate. One argument that is sometimes advanced is that regulation will stifle innovation. But it is not a trade -off. We don't consider safety standards for medicines, cars or aircraft as obstacles to progress. They are the reason why people trust those technologies in the first place. It is possible to design technologies that advance rights, economic opportunity and safety simultaneously. And I encourage you to share examples during this session. My office is ready to support efforts that translate principles into practical action. Tomorrow we will launch the Human Rights Data Exchange, HRDX. This is the world's first open, authoritative service with evidence of where rights are under threat, what happened, why and what to do about it. My office's human rights advisory service that was set up under the Global Digital Compact is also aimed at helping states and others to govern AI in line with human rights responsibilities. The unofficial slogan of the tech and AI industry is bigger, faster, better. I would rather advocate for smarter, kinder, wiser. The historian and politician Lord Acton famously wrote, and I quote, power tends to corrupt and absolute power corrupts absolutely. History has shown time and again that power needs to be kept in check. So our task is clear, to ensure that AI serves people, not the other way around. And human rights provide the compass, so let us use them to the fullest. Thank you very much.
Peggy Hicks
Thank you. Hi, Commissioner, for those thoughtful words that I'm sure we'll come back to in the course of this discussion. At this point, I'm delighted to introduce the thematic co -chairs for this session, Linda Bonyo, founder and CEO of the Lawyers Hub of Africa, and His Excellency Óscar López Águeda, the Minister for Digital Transformation and Civil Service of Spain. As I said, they're the thematic co -chairs, and I will give the floor next to them for their opening remarks. First to Linda Bonyo.
Linda Bonyo
Thank you so much. Your Excellencies, ladies and gentlemen, my name is Linda Bagno. I am the founder of Lawyers Hub. We are an African organization working at the intersection of law, technology, and justice since 2018. For close to a decade, we have been training lawyers and governments and technologists on how they can engage on digital rights across Africa. We run the Africa AI Policy Lab. We also run the Africa Startup Law Accelerator because we believe in a multi -stakeholder approach to digital governance. This week, lawyers have launched the Africa AI Governance Index. It tracks AI strategies, laws, and institutions across all African states. For the first time, we have an assessment that covers everyone across the African continent. I invite you to go to AIpolicy .africa and interact with the work that we do at the AI Lab, courtesy and supported by various partners, including the Patrick McGovern Foundation, that I want to thank. But why does this matter in this conversation? I think that human rights is individual rights. I want to speak for the gig workers who are managed by algorithms in Kenya, who really hope that this conversation will take them to the next level. But I also want to acknowledge the asymmetry of power and to acknowledge the voices of the Gen Zs and Gen Alphas that may not be in this room today. I also want to acknowledge the civil society organizations that are here, yet constrained by, you know, least resources. But also the Africans and the Global South who are in this room and have different perspectives. It's a difficult time getting here using visas and spending so much time. We see you and we will speak for you. Thank you so much.
Peggy Hicks
Thank you very much and I'll now turn the floor to Minister López Aguedo
Oscar Lopez Agueda
Good morning everybody, I will go in Spanish Dear colleagues it's an absolute honor it is a true honor for Spain to co -chair this session together with a number of governments and such as organizations as Linda mentioned one of the most challenging parts of the Declaration of Human Rights which was established many years ago there are many of us here today but we still feel that we need more voices, so in Spain we'd like to open a discussion a democratic discussion that is also inclusive where all of these voices lie, especially voices from Africa. How can I put this? Either we do something or the global south will just serve as a battery for the greater powers because it is also just going to be someone who provides data. So we have to be inclusive and global when we consider AI. In Spain, we would like to open up this global dialogue and to move towards AI that is trustworthy, innovative, sustainable, ethical and human -centered and not the other way around. AI for science, transparency, common good and cooperation. The geopolitical times show us that AI is not a consumer good. It is a political tool. therefore the world needs AI that finds cures for cancer and which can detect earthquakes but is not a tool to exploit our data. Today AI is powerful and only has two paths, that of digital rights or that towards an oligarchy. From working with UN we can govern AI and this is what we're going to look at in Spain since 1948. The UN has served to craft human rights and the UN has to promote these digital rights given the menace of algorithms. In Spain we have AI during our presidency of the EU we have signed the EU AI Act which is the first ambitious legislation in this area we have moved towards the prohibition of sexual deep fakes deep fakes and child pornography within the EU legislation we have put in place a Spanish agency for the supervision of AI which is the first in Europe we've been pioneers in driving forward digital charter which is also opening up for the Iber American digital rights charter also in parliament we have a child protection law in the digital era which which is going to restrict under 16s to digital networks Spain sees it very clearly we all have to be very clear no digital technology no is only going
Peggy Hicks
Thank you very much, Minister. We are now going to move. I'm going to allow our distinguished co -chairs and high commissioner to move off the podium, and we're going to change over to the first of our panels within this session, which will focus on human rights, accountability, and justice, including for children. So I will now ask the panelists and moderator for that session to come forward as we do this changeover and announce their names to save us a little time along the way. So we're very fortunate to have as the moderator for this session Ambassador David Lametti, the permanent representative of Canada to the UN and former Minister of Justice of Canada. Welcome, Ambassador Lamedi. And we have a very distinguished panel, including Sima Bahous, the executive director, of UN Women. Pleasure to welcome you, Madam. His Excellency Amal El Fallah Seghrouchni. The minister delegate from Morocco, welcome. Samuel Arias Arzeno, the judge of the first chamber of the Supreme Court of Justice of the Dominican Republic. We are also very fortunate to have with us Sonia Livingston, who's a member of the Independent International Scientific Panel on AI and a noted expert in this area on child rights. And as well, Sasha Luccioni, co -founder and chief scientific officer of the Sustainable AI Group. And finally, Jhalak M. Kakkar, the executive director, Center for Communication Governance of the National Law University of Delhi. Welcome to the panel and moderator. Over to you, David.
David Lametti
Thank you, Peggy. Thank you, Peggy, for that kind introduction and for the efforts to organize. This discussion and thanks to the thematic co -leads and co -chairs as well. Pleased to be joining you today as the moderator on this panel on human rights, accountability and justice in the age of AI. close to my heart as a former Justice Minister and as a former law professor. Our discussion will explore the role of the justice system in delivering accountability and upholding human rights for people affected by AI, including children. The justice system, I think it's obvious to a jurist, has a clinical role in ensuring that the AI revolution is just and lawful, and it's grounded in the rule of law, trying to ensure, as Dr. Turk has pointed out, transparency, inclusion, accountability through human oversight. Civil liability is a principle that exists in most justice systems, and it's focused on human responsibility, but what happens when it's a machine making the decision. Human beings bring not only intelligence, but also judgment and wisdom to choices that have to be made. So the challenge for AI is being subject to that human judgment and human intelligence. In Canada, we have a new policy, AI for All, which is trying to ensure privacy rights and safety in the use of AI, and in particular, a new proposed law on children, which would restrict any social platform to children under 16 until it was proven to be safe. And also true for AI, that AI would have to be proven to be safe, a chatbot, before it could be exposed to children, and with a digital commissioner to ensure that. So there are measures being taken, and the co -chair has mentioned a number of European measures. So with that, I won't reintroduce the panelists, because Peggy has already done that. We'll move right to the questions. I will allow all the panelists to weigh in on questions. I would ask them to moderate their answers to a couple of minutes, perhaps going longer on one of the three questions, if they feel a particular need on that. But in order to get through this, I have to ask you to all be economical. So the first question, I invite the panel to share their perspectives on what's already been done or what is needed and how justice systems or other accountability mechanisms can be strengthened to respond to harms related to AI. For those of you engaged in issues of AI and gender, I would welcome your reflections on how AI governance can strengthen accountability and ensure that the rights of women and girls are protected online in light of the gendered impacts of AI highlighted in the preliminary report of the scientific panel. So I shall begin with Sima Bahous
Sima Bahous
Please go ahead. Thank you. Thank you very much, Ambassador Lametti. It's great to see you here also across the ocean in Geneva. And let me start by saying that I thank you all for being here and also for being a keen audience on this. divides of women and girls in this age of technology and digital divides also. So if we can start, if I can start by saying, left unregulated, AI becomes not only a mirror of the inequalities that already exist in our societies, but it becomes also a powerful mechanism for amplifying discrimination, amplifying exclusion, and violence against women and girls. At a time when we are witnessing a growing global backlash against women's rights, AI governance really cannot be treated solely as a technical issue, and it is fundamentally a human rights issue, as we have heard from our colleague Turk a little bit earlier this morning. The gendered impact of AI are already well documented. AI systems reproduce and amplify harmful stereotypes. Because they are trained on data that already reflect existing biases. And evidence is showing us that 44 % of assessed AI systems demonstrate gender bias. So UN Women Research also found that almost one in four women, human rights defenders, activists, and journalists whom we surveyed had experienced AI -assisted online violence. And 6 % of them said that they have been victims of deepfakes or digitally manipulated imagery. And so much of this goes undocumented, goes unnoticed, goes unreported. But also let me tell you that research is showing us that up to 99 % of online deepfakes and are manipulated sexual imagery target women. So also women and girls. And so we really need to be very careful and cognizant about this and see what are the solutions for all this. This also reveals that impact goes beyond. 41 % of all women responders tell us that they self -censor on social media to avoid abuse. These harms are pushing women and girls out of online spaces, limiting their participation in public life, threatening democratic participation, and reinforcing existing inequalities. So the challenge today is not the absence of principles of justice, because as we said, the principles are well established, but the absence of implementation and accountability. So first and foremost, states remain the primary duty bearers. International human rights law already provides a comprehensive framework to protect rights, including through the Universal Declaration. Human rights. the International Covenant on Civil and Political Rights, the Convention on the Elimination of All Forms of Discrimination Against Women, or CEDAW, we call it, and also all member states have an obligation to respect and protect and fulfill human rights, including by ensuring accountability. At the same time, technology companies and platforms cannot hide behind the complexity of their systems. So responsibility must extend across the entire AI lifecycle, from those who design foundation models to those who deploy them to those who integrate AI into products and services. So just as we would never release food or medicine or aircraft or vehicle or electrical products for use without rigorous oversight, AI should not be deployed. Without that, we cannot deploy AI. We cannot deploy AI without mandatory safeguards to ensure it is safe. accountable and does not cause harm for anyone, of of course, including women and girls. I thank you, Ambassador.
Amal El Fallah Seghrouchni
Bonjour. Good morning. I'll answer in French, if I may. I think the three questions you've raised this morning, which bring us together, have something in common, which is the existence of rights, human rights, for example.These are already there, but they seem insufficient in the face of this technology, which is spreading at breakneck speed. There are two points I'd like to raise before I answer the questions. The first point is... the concept of something exogenous compared to something endogenous. Now, what do I mean by this? I listened to what was being said at the beginning. People talk about facing up to climate change. Climate change is external to human beings. It's part of the environment. And a lot of what we regulate is external. Now, when you want to regulate psychological manipulation, that's something that can't be done easily. It's something that will take time. When you use algorithms that manipulate a child's psychology, that's not something you can measure. That's not something that you can even perceive. Therefore, the internal nature of the child's psychology and the internal nature of the effects of AI create significant issues at the level of the rights and the legal mechanisms that we have at our disposal. so the question we could ask is how to audit our neurons how can we understand that our brains have been transformed now you might remember that in the early 2020s Chile had proposed a law against cognitive manipulation they'd introduced the concept of cognitive rights and that was quite visionary at the international level today these cognitive rights must be defended but that opens up a whole host of questions and the real issue is when there's a decoupling between technology and the technology and laws to regulate AI and AI systems, we need to be able to open up the black boxes of AI. And there are very few lawyers and there are very few sociologists who are in a position to understand exactly what is going on. And even AI researchers do not have the tools sometimes to understand the complexity of these algorithms. If you take LLMs that exist today, that exist on the market, I'm not going to advertise for anyone, but you have between 175 billion and 180 billion parameters involved. These are systems that... are working full -time, we make 10, 15, 20 requests a day, say, and if we put aside environmental issues, if we put aside that issue of economical use, how can we make sure that there's accountability for what appears in these systems and affects the entire planet? Today, when we talk about accountability, and that's the topic this morning, we need to understand that accountability means that we need to be able to unwrap all these causal effect dynamics to understand AI. And I've looked into the technical side of AI, and I think we're very far from where we need to be. I'll stop there, and I can come back on the detail afterwards.
David Lametti
Yes, thank you very much. Transparency is also at stake in this issue of accountability. .
Samuel Arias Arzeno
Good morning. Thank you I'll go in Spanish. Allow me to begin with a very simple reflection. When we talk about governance of AI, we need to look at regulation, innovation, and technological development. However, the true challenge of governance doesn't happen when we adopt laws or ethical principles, but it happens when a person considers an AI system has violated their human rights and their data privacy. It is at this moment when governance is no longer just a concept, but it then becomes a reality. Justice systems. should be seen not women shouldn't be seen just as users of AI these are essential institutions to convert human rights principles into true human rights to guarantee the existence of accountability when AI causes harm international human rights law has a very solid foundation human dignity equality before the law non -discrimination due process privacy and transparency and right to effective remedy and these need to be fully fleshed out in the area of AI in this vein the international community or international corporation should reflect this as well Through the Justice Action Coalition governments, civil organisations and international organisations and justice systems, we're all working together to ensure that AI governance is not only restricted to the designing of public policies, but rather it can be effective through justice systems which are based upon people. For the Dominican Republic, it is an honour to be part of this initiative as a member of its executive board and to co -chair with the World Bank the work on emerging technologies and AI in the area of justice. If we want to strengthen our justice systems, I am of the view that we have to focus on four priorities to strengthen institutional capacities of lawyers. And jurors to look at transparency and traceability in the use of AI. to have human oversight that is effective and to ensure that all people have effective mechanisms to access justice and to repair harm. Thank you. Gracias.
David Lametti
Dr Livingstone.
Sonia Livingstone
Excellencies, Chair and colleagues, thank you. So I'm speaking from the Independent Scientific Panel and our report treated human rights as a cross -cutting issue. So we recognise mounting evidence of human rights violations and I have to say the evidence for the violations is currently much more compelling than the evidence for human rights benefits in many settings where AI is being used, in the home, in education, work, communities, even though the promise of AI remains substantial in specialist fields. So the science is clearly documenting harms to individuals and vulnerable individuals. and disadvantaged groups, as we've heard, and to society as a whole. And the report documents the threats to our information ecosystem as AI is being used to create and amplify persuasion and deceit, to spread disinformation, distrust and dissensus, and there are adverse impacts already being documented for individuals, also for collective health, for democratic participation, election integrity and all of the harms, especially focus on women, on children, on a range of disadvantaged groups. We documented threats to privacy and safety, and there is really much to be said about safety, as personal data is taken, manipulated, abused, exploited by AI systems, exacerbated by bad actors. And we see those adverse impacts also clearly documented across a range of groups. It remains including the right to freedom of expression. in more private life, in emotional attachment, in mental health, even in the right to life. And thirdly, the panel has documented evidence of harms in terms of inequality and injustice. As AI power is concentrated in a very few companies, a very few countries, a very few languages, we can see adverse impacts on cultures, languages, inclusion, education, discrimination of a range of kinds, as has already been mentioned in relation to AI decision -making. And all of these land unequally. So I think looking across the evidence that we found from the panel, we highlighted four kinds of risk, which I would like to just note. So there are the direct risks, which have been referred to and which is easiest to get the evidence. We also see the risks because of a hugely uneven capacity globally. So AI deployment in anyone... region, effective as it might be, beneficial as it might be, exacerbates risks of inequalities in many other parts of the world and to other populations. We identified also the opportunity costs, the failure to realise the AI benefits is also a risk to human rights. And then we documented a range of ways in which the development and deployment of AI without rights -based safeguards introduces its own kinds of risks. And I think those kind of unintended but really consequences we can anticipate are very evident. And I do want to say something about children, but will we have another round or shall I
David Lametti
You will get one more shot at the children.
Sonia Livingstone
Because I think the children has been highlighted both by the High Commissioner and the Secretary General and we have really abundant evidence. So I will come back to it. Thank you.
David Lametti
Thank you. Dr. Luccioni.
Sasha Luccioni
Thank you. I'm very honored to be here as a founder of a recent organization called Sustainable AI Group, and we're dedicated to measuring and reducing the environmental footprint of AI. And essentially I do see it as a very key point in terms of human rights, transparency, accountability, and human oversight. And I think that often, I read an article about this recently, often AI is such a vague and hard -to -define term and hard to pin down, and I think that often debates are on different levels, from the technological to human rights. And one way to anchor this debate, for example, are data centers. And this is where I spend a lot of my time and energy, pun intended, studying. But essentially it really brings concrete that while the benefits of AI can technically be global, right, anyone with a cell phone, with mobile, service. A laptop can connect to this cloud and profit from AI to some extent. But the impact, the negative impacts that we see are very, very local. And so in terms of water and energy and emissions, in terms of the health impacts that are getting more and more dire, and we're seeing it really impacts the most marginalized populations already. And what's really difficult is that it's so hard to get transparency. Even, for example, the organizations like the IEA, which I've worked with, struggle to get information because even the country, the member countries don't have the information. And so transparency has been something that I've been fighting for several years now because it does seem to be a very fundamental human right that we should be getting, right? When we use AI, what is this cost? What is the data coming from? Where is the energy coming from? How can I make a choice? How can I, as a user, make a choice between tool A or tool B if I don't know what the cost is? And I think that, you know, maybe it's hard to... operationalized regulation right now because of the pushback, because maybe as a technology it's hard to start restricting it while potentially we haven't reached its full potential. But I think that transparency is really something that a lot of people, a lot of organizations, a lot of governments can get behind. Okay, before we start, for example, I created the AI Energy Score Project, essentially energy efficiency labels for AI models. And, you know, before we start saying, you know, you have an F, so you can't deploy your system, how about we start quantifying and measuring these impacts so that people can make informed decisions when choosing one tool over another. And I think that this is really the stepping stone to so many other things that we can be doing and we should be doing when it comes to AI. Thank you.
David Lametti
Thank you. And thank you for underlining the human rights dimensions of climate and the climate impact of AI. Finally, Executive Director Kakkar.
Jhalak M. Kakkar
Thank you so much. And it's beautiful. I mean, it's an absolute honor to be speaking here today. when we talk about the intersection of AI and human rights and international human rights law, we have recognized the right to equality, privacy, freedom of speech and expression, economic, social, and cultural rights. And we've seen guiding principles, the UN guiding principles being developed on how to operationalize these in various contexts. We've seen national laws imbibing these human rights. But all of these have to be sort of reinterpreted and evolved in the context of AI. If you take the freedom of speech and expression, it's always encompassed the freedom of thought. But what does freedom of thought mean in the context of AI, which is threatening cognitive autonomy, which is raising fundamental questions of information reliability, which has aspects of micro -targeting and manipulation. Similarly, think about the right to privacy. We live in a panopticon. Which is, you know, where we are constantly surveilled by companies. by governments, but even by standards. There are cameras, there are wearables, there are AI glasses, the location of our phones. We also see a concentration of data centers and extraction of minerals, predominantly in sites where there's already a lot of inequality in the global south. And it raises fundamental questions of how do we measure the environmental impact of AI? How do we deal with the implications that economic and other value is being accrued in a location that's very, very far away from the communities that are most impacted? We have fundamental questions of labor justice of the people around the world, but particularly the global south. And another fundamental human rights question is, how do we move away from a focus on individual rights to a broader focus on group and community rights and impact? justice systems to grapple with all of this will really have to be strengthened to not only identify harms, it's not always going to be obvious when a harm is being caused, there will be some harms that will only emerge after we have societal level monitoring they will have to grapple with the complexity of arriving at truth in a time of epistemic erosion and a true fragmentation of our reality justice systems will have to bridge accountability gaps of who is truly liable, they'll have to grapple with the transparency questions, how does the person who is affected prove harm and gather evidence without access to models and information around how they are being impacted justice systems will have to fundamentally grapple with enforcement gaps, they will need technical capacity and expertise, we will have to think about shifting of burden and on perhaps companies to demonstrate that harm is not present in a given situation. They will need to rely on independent sociotechnical experts, auditors. They will have to grapple with the biggest question, which is the transnational cross -border governance gap. Laws are national, companies are global, supply chains run across various countries. How do you access the information and hold people and companies liable in this sort of increasingly global context? So whether it's identifying harms, bridging accountability, enabling transparency, navigating transnational governance, all of this fails if we do not acknowledge the fundamental concentration of power in a handful of companies in a few jurisdictions. These companies frame themselves. And the safeguards control their operationalization, the information they share. with various countries, with courts, with judicial systems. We need to enable the development of national AI ecosystems. We need public interest alternatives to our current technological models. We need to diffuse the power. Because without this diffusion of power, we will not be able to safeguard human rights of the people of the world, but especially those of the global south. Thank you.
David Lametti
I have been told we're running on time. So I'm going to give each panelist one more short intervention. I would ask you to limit yourself to one of the two following topics. One is how do we ensure accountability, especially in the age of agentic AI, or focus on the impact of AI on children and their rights? And if you wish, on gender, age, or disability, or other intersectional identities. again I have to ask you to keep it very very short I've been told 30 seconds thank you Executive Director Bahous
Sima Bahous
I don't remember I was ever able to say anything useful in 30 seconds but I'm going to try I think I wasted my 30 seconds rightly I just want to say that accountability and risk analysis should not be seen as reactionary but as influencing every system that we are building when it comes to AI I also think that the accountability of AI governance begins with four questions who builds AI who benefits from AI who bears the risks and who has a voice for you and women we see this as the building blocks that when connected then we become we have a basis for a good foundation for accountability and when we look at these building blocks allow me to say the data speaks for itself when we look at who builds AI women are largely underrepresented When we look, women make up only 30 % of AI workforce globally. From yesterday's independent scientific panel's report, we learned that 88 % of leading AI researchers are male. So when we look at who benefits and who is at risk, we know that while AI offers a great promise of economic growth, women continue to face much higher job automation risks compared to their male counterparts. And when we look at who has a voice, gender equality remains largely invisible within AI governance. And I will tell you, of the 138 countries assessed, only 24 reference gender in their national AI strategies, and just 18 include substantive gender -responsive provisions. So to address these gaps, AI governance frameworks must embed accountability from the outset. This begins with mandatory, of course, we've spoken, human rights impacts assessments before and after deployment with gender equality treated as. non -negotiable and not as optional. I think I will say finally, Ambassador, accountability requires participation also. Women and girls, feminist technologists, indigenous communities, labor organizations, disability advocates and civil society must be meaningfully involved and empowered. And I will stop there. Thank you very much.
David Lametti
Madame la Ministre.
Amal El Fallah Seghrouchni
Well, I speak French, that will be quicker. I'm going to go for the agentic AI question and the network of agents question and the challenges in terms of accountability. I'd say that the issue of accountability in agentic systems is probably the most significant legal challenge this decade because we have a network of AI agents that autonomous each individually and there's no human that is at the origin of the final decision so when we want to make an agentic network accountable we don't know which of the agents is accountable and we don't know how to distribute this accountability amongst the different agents when you have three people A, B and C and each person does a little bit, the result can be bad but you need to be able to find accountability at the level of each person, at the level of a particular moment so there's an issue with detecting the moment where the mistake was made in these agentic systems and we don't know how to respond to these systems which have billions of parameters and it's extremely difficult to move back up the chain of accountability so very briefly I'd say I take Iderati X .0. It's a platform in Morocco. We process millions of administrative transactions every day, 52 million. When we have 52 million, it's very difficult to find the exact cause when there's a problem. So what we need today is to find a new accountability paradigm because we need to find new methods to face up to these decision -making factories, if you will. So I believe that unfolding would be the key approach. Unfolding would allow us to develop a causal graph, a non -deterministic causal graph of agentic systems to identify each, each, node which rights apply and understand where there is a lack of coverage because that's where harm can be caused. so to sum up I think that we need to impose three obligations today in terms of accountability first structured documentation of the agent chains, which does what when, when these agents are deployed in public services we need to appoint an identifiable human being in charge for all issues that are tied to rights and there needs to be a right to redress in good time in line with the speed of AI systems so that humans can receive redress if necessary therefore the functioning of the algorithm must never be autonomous in regards to rights and the law.
David Lametti
thank you very much for having explained the challenge and thank you for those suggestions Arias Arzeno
Samuel Arias Arzeno
Thank you very much. I'll just be quite brief. So the challenges of the judiciary. The true debate is not really asking us whether AI should be present in our courtrooms. The reality is it's already there. The true challenge is to ensure that its use is transparent, responsible, can be audited and completely compatible with ethical, judicial priorities, human rights and the rule of law. No public institution can justify a decision by saying that that is what the system told us to do. The responsibility should lie with people and with institutions. Technology should help the judge. It should never replace him or her. Thank you.
David Lametti
Gracias and thank you for your brevity. Very much appreciated. Dr. Livingstone.
Sonia Livingstone
Thank you. Thinking also of the panel's evidence on harms and specifically around children, just very few facts. Across 11 global south countries, up to one child per classroom reported that AI was used to make sexually explicit deep fakes of them. Reports of AI -generated child sexual abuse material to the US cyber tip line NECMEC are rising exponentially. Child sexual abuse material is documented in some training sets and offenders can now fine -tune open models on child sexual exploitation. A national US survey found one in three children had discussed important and serious emotional matters of an AI companion. Stanford's risk assessment of AI companions found these products fail to meet basic safety standards and they deliver sexual, violent and harmful language. tests of 32 AI models for child safety found consistent failures even when tested under child -facing conditions. So the problem is very real and urgent and we are a long way from implementation. Child rights violations facilitated by AI are often not recognised in law and law is transparency, accountability and remedy or address available to children. So we're not going to solve this problem by restricting children if that is on the agenda. Think of the hopes being poured globally into AI tutors, AI mental health support and think also of children's rights to expression, participation, education and information in a digital age all of which they are calling for actively and their voices must also be heard even if not quite in this room today. So it must be about building rights respecting safeguards into AI data sets, models, AI procurement and deployment and I refer people back to... Mr. Kateris' call for nations to adopt an AI child safety pledge.
David Lametti
Dr. Luccioni.
Sasha Luccioni
Yes, thank you. I want to talk about agents and how essentially we're already in a little bit of a dire place when it comes to transparency, accountability, and human oversight, and I think agentic AI is making it really worse because it amplifies the degrees of separation between a user and the underlying technology. Already as a scientist, I used to study AI models, which you could probe and look at to some extent. Then these models became AI systems with filters, with all sorts of other parts to the machine, and now with AI agents, you have an even larger degree of, yeah, essentially more and more parts in the machine, and they become harder and harder. When something fails, you don't know where it's coming from, essentially. So I think that it's really important to keep imposing safeguards and standards as well. Actually. I have to run in like a few minutes. I'm giving a. talk in a session about AI standards. So I think we should be having standards for safety, sustainability at each part of this machine, no matter how many there are, and no matter how easy it is to say, well, the agent made the decision. Thank you.
David Lametti
Thank you. Executive Director Kakkar.
Jhalak M. Kakkar
Thank you. I want to sort of build on some of the points that Sonia has already made and sort of add to that to say that given that AI is getting so integrated in various fundamental aspects of our lives, you know, it has various implications for children's safety, well -being, development, which means that from the get -go, we need to be thinking about the design of systems that are being deployed, and we should not be reacting to the occurrence of harm to children. So every system, you know, pre -deployment through the design stage needs to be thought, through, we need to have certification and post -deployment monitoring of harms and impacts. But we also need to fundamentally think about the business models of these AI platforms and the challenges of current approaches, which are very dependent on attention maximizing design and advertising based engagement and what that means for the well -being of children. So I think we should, you know, very often the onus is on parents and children to, you know, figure out the load of identifying, assessing and managing risks and harms that are arising. And I think we need to sort of replace the onus back on platforms, companies, also governments, parliamentarians and regulators to set up systems that really, truly work safely while allowing for, you know, children's autonomy and growth and development. But the last point I want to sort of make is AI and child also raises a fundamental point of surveillance of children. How does a child who's concerned? How does a child who's constantly watched and monitored grow up? as a sort of a contributing member of a democratic society. I think that's something for us to sort of fundamentally think about what this generation of children is going to grow up as and what we are setting ourselves up for a society
David Lametti
Thank you. Thank you to all our panelists for a very thought -provoking series of comments. Gracias. Merci. I turn it back over to Peggy.
Peggy Hicks
Thank you for being here. Thank you very much, Ambassador Lametti and the whole panel. I think we've all been challenged on a number of fronts to think deeply about some of these issues of transparency, accountability, and oversight and what they mean. And I personally found the last segment, I'm very glad we gave them some time for that second round. I'm sorry we had to cut things short. I'm sure we could have listened to much, much more. But we're now going to move on to, to the section of the program where we will hear from audience interventions, which will be moderated by our co -chairs. So I'll welcome Linda Bonyo. Minister Lopez Agueda forward again. And these are the intervenants who have pre -inscribed on the list. Unfortunately, of course, we won't be able to hear from everyone, but we're very grateful to all of you that are here for this segment. So I'll hand over to the two of you. I think you're going to have some assistance in getting everything set. And we will move forward with this segment. Thank Thank you.
Co-moderator
Thank you. Thank you. We will now facilitate interventions for this segment. Statements will be delivered based on the speaker's list established through inscription that was made available on the UN Global Dialogue on AI website. Please note that in the interest of promoting broad participation from stakeholders, we will proceed with one representative from member states and one representative from other stakeholders, and we will alternate. The intervenants will be invited to intervene from the lectern on the stage. Intervenants, if you can come forward and sit in the first two rows, it might be easier for you to reach the lectern quickly and speak. The e -delegate list is closed, and any changes that you wish to make would need to be communicated to the secretariat, who is sitting just to the right of the podium. And my favorite part, a timer has been set. You will be able to see it on the screen, and you will also be able to see it on the lectern. We kindly encourage all intervenants to respect the allotted time so that the largest possible number of intervenants may be accommodated during this segment. As some of you may have learned yesterday, microphones will be automatically switched off once the allocated speaking time has expired. And so I would ask you to try to be aware of that. With that, I will turn it over to our distinguished co -chairs who will begin reading from the list. Thank you very much.
Oscar Lopez Agueda
Okay, so now we move to the Minister for Artificial Intelligence from Ireland, Ms. Niamh Smyth. And then we move to Mr. Arash Ajikani, Research Professor from the Technical Research Centre of Finland.
Niamh Smyth
Distinguished colleagues, Minister, lovely to see you here, and guests, ladies and gentlemen, it's a privilege to be here with you today. And given the transformation... ...of effects AI has on societies, it is clear that we face a future of unprecedented opportunities and challenges. But we must not lose sight that it is the responsibility of all of us to ensure that technologies are developed to the benefit of humanity and not to its detriment. We firmly believe that the way we do this is to ensure that human rights are at the core of AI governance. This is reflected in the human -centric and rights -based approach emphasised in our own national digital and AI strategy. Global AI governance must align with international human rights frameworks, promoting freedom of expression, privacy, access to information and gender equality, while actively countering all forms of violence, including sexual and gender -based violence. Of course, AI and digital technologies have a significant impact on human rights in different ways, but I want to highlight today the necessity to protect the rights of women and children. While AI has the potential to help people with disabilities, it also has the potential to help people with disabilities. While AI has the potential to advance women's and girls' rights, we see how the current lack of implementation of regulations has exacerbated gender -based harm. Ireland is concerned that without the necessary human rights safeguards and oversight, AI systems enable technology -facilitated gender -based violence. The deployment of AI online content moderation without human oversight can amplify disinformation, undermining the integrity of the information landscape, and this disproportionately affects women and girls. While we are particularly concerned about how AI can exacerbate discrimination against women and girls, discrimination through algorithmic bias and lack of representation in training data impacts the accuracy and fairness of AI systems. Furthermore, AI tools lack gender response lens. They can hinder women and girls' participation in public life. Whether journalists, politicians, human rights defenders, or an advocate for gender equality, women and girls are particularly targeted by online disinformation campaigns. AI -amplified disinformation campaigns and hate speech undermine the rights of women and girls to exercise freedom of expression by forcing them to self -censor, close their digital accounts or otherwise withdraw from participating in online spaces. This contributes to a broader chilling effect on their full, equal and meaningful participation in society. Their digital participation is also hindered by significant disparities in the access of women and girls to have technology, contributing to the gender digital divide. The full, equal, safe and meaningful participation of women and girls in all realms of society, including online space, is critical to achieving gender equality, sustainable development, peace and democracy. This is everyone's responsibility. Thank you very much.
Oscar Lopez Agueda
Now it's... Mr. Afra Ajikani is a research professor from the Technical Research Center of Finland he's not so we move to this one so we move to Mr. Robert Baruch Public Affairs, Europe and Multilateral Relations from Universal Music Group okay there we go, thank you okay, rock and roll
Robert Baruch
since the invention of of the of the record and radio artists like the first global superstar Enrico Caruso all the way to Taylor Swift, Blackpink, Ladipo and John Legend who are here by the way have always stood at the intersection of culture and technology at Universal Music Group our approach to music and thus our approach to the intersection of music and AI starts with one simple principle, put artists first, protect their rights, advance their interest, and from that foundation create opportunities for creativity, innovation, and growth. This approach is firmly rooted in human rights. Article 27 of the Universal Declaration of Human Rights and Article 15 of the International Covenant on Economic, Social, and Cultural Rights recognize that creators are entitled to protection of the moral and material interests from their work. Those principles are just as relevant in the age of AI. Over the past year, we have put this into practice. We have established partnerships with companies including NVIDIA, Clay Vision, Udeo Stability AI, and Splice, announced a broad -based AI collaboration with Spotify, and advanced important AI protections through our agreement with TikTok. These partnerships demonstrate that the future of AI is big. Built not by choosing between innovation and innovation. and creators' rights, but by advancing both together. From the perspective of those who create, invest, and bring music to audiences around the world, responsible AI governance rests on three pillars. First, creators and artists must remain in control of how their work is used to develop AI systems. Accords reached freely and voluntarily between creators, rights holders, and technology developers create trust, legal certainty, and sustainable innovation. Second, transparency. Transparency. Creators should know when and how their works have been used to develop AI models and how they will be used for outputs. Without transparency, there can be no accountability, no trust, and no meaningful collaboration. With transparency, artistic integrity can be protected and, where necessary, enforced. Third, human oversight. AI is an extraordinary creative tool, but it should amplify human creativity, not replace it. Music is profoundly human. Long before we wrote stories, we used to think that music was a tool. As Bono said, we sang before we spoke. Music is more than data or patterns. It carries memory, identity, emotion, and lived experience. As Bjorn Olvea said, music is a testimony. testimony, AI offers tremendous opportunities, it can help more creators in more languages and more cultures participate in the global creative economy, it can preserve cultural heritage unlock creativity and deepen the relationship between artists and their fans if applied responsibly AI can ignite a new renaissance of human creativity preserving our cultural heritage shaping culture through the power of artistry and inspiring future generations
Linda Bonyo
Thank you, thank you so much Robert from Universal Music Group you say put artists first and we sang before we spoke thank you for your sentiments at this point I would like to invite from South Africa His Excellency Monty Gungubele Deputy Minister of Communications and Digital Technologies you have the floor if he's not here okay we will go to Slovenia Her Excellency Anita Pipan forgive my African accent we pronounce everything Anita the Permanent Representative to the United Nations Office and other international organizations in Geneva excellent thank you so much the pleasure
Anita Pipan
so Excellencies ladies and gentlemen dear colleagues I'm Scientific Counselor at our Permanent Representation and I'll be delivering the address on behalf of the Republic of Slovenia thank you for understanding hi Commissioner for Human Rights Mr. Turk has emphasized that earlier that upholding and advancing human rights is fundamental to building trustworthy AI. In our view, in Slovenia's view, trust is not only an ethical objective, but also a prerequisite for the broader adoption of AI. And this approach is reflected in our national AI strategy 2030, adopted in March this year. The strategy clearly promotes a human -centric approach to AI, grounded in human rights, democratic values, ethics, safety, transparency, and accountability. Its objective is not only to advance AI capabilities, but to ensure that AI is developed and used responsibly in the public interest. The strategy also recognizes that building trust requires effective and practical implementation, and this entails ensuring responsible governance, legal certainty, appropriate support mechanisms, and sustained investment in AI literacy and technology. skills. And as a small linguistic community, Slovenia places particular importance on protecting linguistic and cultural diversity. We believe that inclusive AI should reflect and preserve all languages and cultures, ensuring that no community is left behind. Looking ahead, we should first continue promoting a human rights -based and human -centric approach to AI, firmly grounded in democratic values and the rule of law. Second, promote inclusive AI by supporting linguistic and cultural diversity and reducing digital inequalities. Third, complement regulatory frameworks with practical implementation support, including capacity building, guidance, and AI literacy. and, of course, strengthen international cooperation through the exchange of good practices, interoperable approaches, and capacity building to advance trustworthy AI globally. Thank you very
Linda Bonyo
Thank you, Slovenia, especially on legal certainty. Thank you very much. At this juncture, we will go to International Center for Not -for -Profit Law. We have Zachary Lampel, Senior Legal Advisor and Coordinator, Digital Rights Program.
Zachary Lampel
Thank you, Excellencies, ladies and gentlemen. On behalf of ICNL, the International Center for Not -for -Profit Law, it is an honor to speak to you today. Yesterday, we heard from our global leaders, including the Secretary General and Heads of State, about the importance of digital rights in the international community. We heard about the AI we want and the AI we envision. A running theme throughout their interventions was that AI must be human -centered. AI must create value for humanity. AI must protect democratic principles. And AI must be trustworthy and anchored in the rule of law. Sadly, the envisioned AI is not the AI that we have today. The AI we have today restricts the freedom of expression and the freedom of association and violates the right to privacy by flooding and manipulating the information ecosystem, producing child sexual abuse materials at an unprecedented rate, and fueling mass surveillance amongst other harms. So the question becomes, how do we achieve the AI we want? ICNL believes that the AI envisioned and articulated throughout this dialogue can be achieved via the following three actions. First, AI red lines. Clear prohibitions. Or moratoria on AI systems and uses that pose unacceptable risks to human rights. This is not a radical proposition. Rather, it is explicitly stated in the 2024 UN General Assembly resolution, seizing the opportunities of safe, secure, and trustworthy artificial intelligence systems for sustainable development. Furthermore, UNESCO's recommendations on ethics on artificial intelligence already explicitly rule out the use of AI for social scoring and mass surveillance. We can and should implement red lines where AI cannot comply with human rights. Second, standardized or uniform certifications with national procurement rules linked to those certifications. Think of a process like green building certifications. Multilateral and multistakeholder bodies create a certification system based on embedded safety constraints at the design stage. National procurement laws and regulations then are linked to that certification process, which ensures basic standards of safety and creates a market incentive to design and deploy safe and secure artificial intelligence systems for sustainable development. And we can't afford to give up on a system that is so important to the future of artificial intelligence. We can't afford to give up on a system that is so important to the future of artificial intelligence. Third and finally is the right to remedy. Simply put, too often decisions are made via AI systems, and those wrongs have no avenue to challenge that decision. Laws and regulations must contain an accountable actor with decision -making workflows. These three actions are rooted in international law and the rule of law. The rule of law requires implementing binding legal frameworks equally without regard for wealth, power, or influence. We must start to enforce international law, including international human rights law, today. ICNL is working and will continue to work to implement these three action points through multilateral bodies like the Freedom Online Coalition and its Task Force and AI and human rights, OHCHR, UNESCO, UNGA, and other UN agencies to create and enforce rules for safe, trustworthy AI. I see now we'll also continue to support... Thank you.
Oscar Lopez Agueda
Mr. Lampert for your three ideas and now we move to Alejandra De Bellis Bonilla permanent representative from Uruguay.
Alejandra De Bellis Bonilla
Very good morning one and all thank you very much co -chairs and facilitators we'd like to especially thank the panelists for this morning for Uruguay undoubtedly the impact of AI in the protection of human rights is absolutely critical the example given to us this morning by the High Commissioner what do we do when there's a new drug that comes into the market I think this really makes us truly think and it was certainly a powerful message. The report from the scientific panel also addressed the concern of algorithmic bias. In the following minutes, Uruguay would like to focus on gender equality for women and young girls. Uruguay agrees that AI offers an opportunity to bridge gender bias, but in the design and deployment of AI systems should focus upon gender approaches. To bridge the existing gender divide, we need improvements in accountability, normative frameworks, sustainability and funding, and systemic inclusion. Uruguay is fully committed to drive forward a digital transformation that includes the development and use of AI. in an inclusive and agenda -based approach. In this vein, it adheres to the Madrid Declaration, His Excellency the Minister referred to this on the Fifth Ministerial Conference on the Feminist Foreign Policy. And looking at this, we'd like to highlight the recent Global Partnership for Human Rights. This initiative from the Office of the High Commissioner of Human Rights is a broad, multi -stakeholder platform and network which connects all of us, all our efforts, which has a cross -cutting focus on the impact of AI, and which, in our view, is a very natural way in which we can address all of these topics with everyone involved. With that, I thank you.
Oscar Lopez Agueda
Muchas gracias, Alejandra. Thank you very much, Alejandra. We move to Ms. Merth Hickok. from the Center for AI and Digital Policy. So we move to Miss Paulina Ibarra. She's the gerenta general from Fundación Multitudes. Good luck. Good luck. I've been better in football than world champions. Now we move to Mr. Franco Giandana Gigena, policy analyst from AccessNow. Okay.
Franco Giandana Gigena
Thank you very much. Thank you. happy to be here Argentina is playing later I come from Argentina so no problem with that excellencies, colleagues co -chairs, thank you for this opportunity to contribute to this thematic breakout cluster 4 Access Now is a grassroots to global civil society organization dedicated to defending and extending the digital rights of people and communities at risk I will focus my intervention on 4 main points first and foremost the objective of this global dialogue is to foster international cooperation on AI governance while ensuring that the perspectives of the global majority are reflected alongside those of leading AI powers human rights centered AI required the right respecting collection and processing of data alongside robust governance and oversight of automated systems and Model development and sharing practices must balance necessary safeguards with meaningful openness so that global majority countries can build their own technical capacity and overcome a historic dependence on externally held expertise Second, AI tools are supercharging surveillance and suppression of protesters, journalists, human rights defenders and other communities at risk We therefore echo calls for the adoption of human rights -centered, binding legal standards and global governance mechanisms to regulate digital surveillance technologies and AI, also addressing chilling effects Third, we call for a halt to the development and use of AI tools that are facilitating violations of international humanitarian law and international human rights law, especially in context of conflict or war Finally, the current geopolitical landscape presents an opportunity for global majority regions as well as middle powers to seek and foster collaboration and joint resources in an unprecedented fashion There is nothing magical about AI development It is a governable technology, but we need alternatives to the dominant paradigm being pushed by a handful of companies who are prioritizing AI acceleration over all other concerns Fundamentally, people around the world do not only need access, but most importantly, they need agency We trust all those gathered here understand the urgency and the unique opportunity and responsibility we share in
Linda Bonyo
Excellent Thank you very much to Ix de Bestra Argentina The day Kenya plays football is over for all of you because of our athletes. Anyway, I'd like to invite at this point Poland, His Excellency Rafał Kownacki , Director of Department for International Cooperation, Ministry of Digital Affairs. The floor is yours.
Rafał Kownacki
Poland didn't qualify to the World Cup, so I wouldn't mention. But good luck to any who qualified. Thank you, co -chairs, excellencies, distinguished delegates. It's a really privilege to address the first global dialogue on AI governance on behalf of my country. For Poland, governing artificial intelligence is not just a technical challenge. It is a question of values. When we decide what an algorithm may do to a human being, we decide what we believe a person is worth. And no doubt, from the perspective of every government, each and every individual should be the key priority. But let us be honest about both sides of AI technology. Artificial intelligence is a great opportunity for health, for education, for science and Poland intends to use it well. But opportunity and risk travel together and they must be held in balance. We already see the warning signs. Deep fakes that pass of synthetic lies as truth, eroding the trust on which democracies depend. Weapons that edge closer to choosing by themselves, with no human hand who lives and who dies. Systems that reduce a person to a prediction that they can neither understand nor contest. Some choices a machine must never make on its own. Poland knows. It knows from its own history what it costs when people lose control over the systems that decide they fake. so our answer is not to halt innovation but to keep one factor constant at its core the human being, the human factor the age factor however capable of our systems become a person must keep the final word, to understand the decision to contest it and where it matters most to switch the systems off this is non -negotiable above all in high risk systems where a single decision can cost a life, a livelihood or a fundamental right it means writing human control into our values, our laws our standards and our institutions not leaving it as a declaration. Poland therefore welcomes the Councils of Europe Framework Convention on AI the first binding treaty to place AI within the order of human rights, democracy and the rule of law. We are turning these principles into practice at home Poland's first comprehensive national digitalization strategy Poland 2035 places the human being as its center and our national AI policy commits us to artificial intelligence that is human -centric and trustworthy. The human factor belongs to no single nation. It is the common ground on which a truly global governance of AI can be built. Let us hold ourselves to one measure of success, that after all our progress, the human being still stands at
Linda Bonyo
Excellent. Thank you so much. Thank you, Poland, for talking about human -centric in a world that is full of AI agents. And so at this point I would like to invite Ms. Isabella Henriques from Instituto Alana, who she is the CEO. You have the floor.
Isabella Henriques
Thank you, Excellencies, distinguished delegates and colleagues. I am Executive Director of Instituto Alana, a Brazilian-based global organization that brings a Global South perspective and highlights that children are not a homogeneous group. We must recognize their intersectionalities and how indigenous children, girls, children with disabilities, and children of African descent, for example, have their rights disproportionately impacted as a result of the gaps in AI governance. Speaking from Brazil, I am proud to note the recently approved milestone, the ECA Digital, a landmark framework for the protection of children's rights in the digital environment, which innovates by stabilizing the digital environment. Establishing a logic of prevention and productivity, accountability, including administration... achieve accountability and enforcement through the justice system. I also speak today on behalf of over 120 signatories of the Joint Statement on Children's Rights at the AI Dialogue, a global campaign of civil society organizations, academics, and child rights experts coordinated by Five Rights Foundation and united in one message. Children's rights must be central to this dialogue. Children are among the earliest and most frequent users of new technology, yet they remain largely missing from national AI strategies and governance frameworks. As discussions continue, it will be important to ensure that children's rights are reflected as a cross-cutting consideration across AI governance. Children's rights, safety, and well-being must be built into these systems from the start, not added after harm has already occurred. Children's data, images, and data are all important. Children's voices and biometric information are being collected commercially. exploited and used at scale. These harms fall hardest on children who are already in situations of inequality without access to other rights such as education and health.
Isabella Hendricks
Innovation should not come at the cost of these risks. If there is potential harm, we need to prioritize culture. Children should interact with AI systems only if they are proven safe and designed to respect children's rights and considering their perspectives in decisions that affect them. Responsibility must sit with those who build and profit from this technology. AI holds real promise for children, but without strong governance, it can cause real harm. We are not asking you to choose between innovation and protection. We ask you to build both together an AI future worth of every child's trust and safety. Thank you.
Linda Bonyo
Thank you very much. Thank you, Isabella. And thank you for highlighting that children's rights are not homogeneous. We have different children, immigrant children, disabled children, and we appreciate that. At this point, I think we'll take a break so that the Minister of Spain can continue to celebrate.
Peggy Hicks
Thank you, co -chairs, for moderating that segment. And despite some of moderating that segment. And despite some of the absences, I think we were given a very broad perspective from both ministers, governments, and stakeholders, showing the breadth of the issues that we have to grapple with on the human rights side. We also wanted to be able to report back to you on the survey that we asked you to take on the way in. And theoretically, I'm going to have a slide up soon that will allow us to do that. Here we go. Looks promising. So the first question was really asking everybody to think about how optimistic or pessimistic you are about how we're doing on the governance side, the topic of the dialogue today. And we don't actually have it there yet, but I'm going to tell you even before seeing. Oh, here we go. So in reality, what we found is that a lot of people are uncertain about whether or not we're where we need to be. But there is a plurality of those of you who are quite concerned that we don't have the measures in place that will really help us move forward to address all the challenges and opportunities that we've been hearing about in this session. So that makes sense. But I guess it's important to then think, what are we worried about? And that was the next question. And what we found in terms of the results of the numbers of what concerns raised to the top for most people, it's not surprising that the top two human rights issues that we came across are those that we've been talking a lot about in this panel. And they go to accountability and rule of law and the impact of AI on children. And as you'll see, the ot significantly in the survey were the issues around surveillance and the use of AI by law enforcement and then also environmental sustainability. And we're grateful for the panelists' contribution on that earlier. Drilling down into that issue of accountability in the context specifically of agentic AI, we also asked who should be responsible for the actions of agentic AI. And overwhelmingly, the top three results are, first, that the companies that develop the AI model, and then, secondly, states aren't off the hook, that there is also a responsibility there, and that the public authority that allowed the system has a responsibility. That, of course, is consistent with the UN Guiding Principles on Business and Human Rights, that the first pillar is the states are responsible for ensuring that the regulatory environment exists to ensure that companies don't do harm. So that makes sense. And then the third. The third is the uses. The organizations that deploy it should be held accountable for agentic systems. So we'll go back to the poll, the second half of the poll later. But the next segment of our discussion today will bring up our next panel, which I'm very happy to introduce. So for this panel, we have a wonderful moderator of Her Excellency Clara Chappaz, the Ambassador for AI and Digital Affairs from France. So I welcome her to the stage now. And I welcome all of the panelists as well. And as before, I'll read off your names while you're helping to get us moving here on the second panel. So one of the first intervenants on that panel will be Her Excellency Willemijn Aerdts, the Minister for Digital Economy and Sovereignty of the Netherlands. Welcome, Minister. We also have Felipe Paullier, Assistant Secretary General of the U .N. Human Rights, U .N. Office, Youth Office. Sorry, Felipe. We, of course, partner with the Youth Office. frequently, so I should get the title right. Good to see you, Felipe. And Ulises Gutiérrez, the Special Representative for Emerging Technologies of Mexico. We also have Wanjin Park, Vice President of KT. Nighat Dad, the founder of the Digital Rights Foundation. Nice to see you, Nighat. And Alvitta Ottley, another member of the Independent International Scientific Panel on AI, who will bring us some of their findings as well. So welcome to this panel and to the moderator, and I turn the floor over to you. Thank you.
Clara Chappaz
Thank you, dear Peggy. Excellencies, dear co -chairs, thanks for all the work done in bringing us here. I'm really thrilled to be on stage alongside this panel and all the amazing guests we have to share about those topics, which are so important to France. As some of you might know, we hosted the AI Action Summit just last year, and we are also hosting G7 Presidency this year, and we've made all the discussion here a priority for the G7 Presidency, in particular when it comes to the intersection between AI and human rights, the responsibility that we have to draw for ourselves, in particular towards children. Building on the latest conversation, we'll try and dig deeper into the topic, focusing on the solutions that we see collectively could help us move forward. Good practices, as AI becomes more powerful with the agentic AI coming up, what are the existing practices that are effective in terms of human rights due diligence? How to ensure that such diligence takes into account the impact on populations most at risk of harm and least likely to be included in AI governance? In the AI governance discussions, I mentioned children, but we have a lot of other challenges ahead of us. Second, the need of tools and evidence. What technical tools, evidence are needed to understand the impacts before they are actually released? What mechanisms do we need to ensure compliance in practice? And three, the specific topic of children. What are proven approaches that can enable affected populations, right holders, especially children, to improve their understanding of those tools because the potential of AI for kids is humongous, but definitely the arm that can happen to their rights of decision -making structures that can effectively participate into the AI governance discussion needs to be addressed. How can we ensure that children are meaningfully engaged in the development of AI governance is something that I think we'll discuss together. To answer those questions, We have the honor to have a distinguished panel with me on stage. Our Excellency Willemijn Aerdts, Minister for the Digital Economy and Sovereignty of New Zealand. Felipe Paullier, Assistant Secretary General for UN Youth Office. Ulises Gutiérrez, Special Representative for Emerging Technologies in Mexico. Wanjin Park, Vice President of KT. Nighat Dad, Founder of Digital Rights Foundation. And Alvitta Ottley from the AI panel. We are quite a few people on stage. So I'm going to have to ask you to keep your intervention as sharp as possible. You have three minutes each. And I'll have to watch for the time. So I'll ask you first, Your Excellency, if you could give us
Willemijn Aerdts
your perspective on those topics. Thank you. Thank you, Ambassador. And thank you for the opportunity to speak here to you today. I think we tried to make it as concrete as possible. And I think that we have some examples in the Netherlands that I would like to share with you. For example, if we make it more tangible, and you won't have to do it again, have some oversight and some control of what's going to happen. We have, in the Netherlands, the algorithm framework that translates legal requirements into practical guidelines also for public organizations. It provides an overview of the most relevant laws and regulations, including both the mandatory and the recommended measures to identify and mitigate bias and discrimination, including a fairness handbook, bias testing instruments, and risk profiling assessments in the framework. Other one of the instruments, the fundamental rights and algorithm impact assessment that helps to identify human rights risk in the use of AI systems and support organizations in taking appropriate mitigation measures. And we have the algorithm register that includes all the algorithms used by the Dutch government in light of transparency. And then maybe on the other topic, Anne, how can we have some effective approaches when we think also about protecting children? Transparency is key in that light, and we speak a lot about AI literacy. Not only do we have to be very careful about how we're using AI, but we also have to only for people working in the government, but also for young people and children. Ambassador, you were mentioning it already, and I think that one of the key messages that I would like to convene here today is that when we take young people into account, we make policy for children and young people, we really need to include them in the decision -making process. We do that by speaking to the UNICEF youth panels, for example, and we also want to make sure that we include parents as well. Without giving them the full responsibility for what their children do, do we feel that platforms play a very important role and should take their responsibility, but we really want to include young people as well as their parents and educators in these issues. And one of the things that struck with me is when I spoke to a lot of young people, that they actually mentioned that some of the issues that they are facing today are too big for their educators in school or their parents, but feel too small for the police. So sometimes they're really looking for a tangible way to make a
Clara Chappaz
Thank you Your Excellency I'm going to turn to you Felipe Paullier Assistant Secretary General of UN Youth Office and in particular I think bringing back on what was just said by the Minister how do we bring the youth in this conversation
Felipe Paullier
Great Thank you Ambassador and thank you Minister for making my job easier I think it's great that we are here delegates, colleagues, private sector, civil society and a few young people in the room which I'm happy that you are also here I think artificial intelligence it's definitely one of those technologies that is advancing at a speed which is extraordinary some of the achievements that this technology has been able to do in just a few years have no precedent so I maybe my perspective here is that the question should not be only about what AI can do, but it's actually what AI can help us to build as a future. And the importance of not only benchmarking AI in terms of how faster our systems, how larger or more efficient are the model, but actually about how AI is contributing to our human well -being, to human dignity, and definitely to grant and protect human rights. Because technology is not... It's about serving people and not the other way around. And AI definitely is helping and should be helping us to learn more, to connect more, to innovate, to solve problems, and expand, at the end, what is human agency. And as Ambassador was saying, when we're talking about youth, this is not an abstract topic. Young people are the most active users of artificial intelligence. They're actually most of the innovators and the creators within some of these huge AI technology firms. But almost always, they are not the decision makers. And they will live the longest of the consequences of the decisions that we make today. So I think maybe my key final message in this panel is really about the inclusion of young people in AI governance, not only within global spaces, but especially at the national level. Because young people are those that first experience the impact of technology. in their education, in their employment cycles, in their everyday digital life, because young people forge also their identities in the digital space. So young people are actually the ones that understand the most about which are the opportunities that they can bring, but also they are the ones that will help us the most to identify which are those bias, those discriminations, those exclusions that are happening because of the use of technology. And just to be clear that if we don't create those meaningful spaces for young people to engage, young people will find their way and they will create their own spaces. This has happened within the climate space, for example. So let's, I think, when we are talking about AI governance, let's bring the topic of young and youth participation at the center, in terms of how they can contribute to the governance frameworks, not only globally, but especially within the national. realities and the national strategies. Thank
Clara Chappaz
you, thank you so much I'm turning to Felipe, sorry to Ulises Gutiérrez, social representative for emerging technologies in Mexico maybe like adding to the conversation we talked about global and national how do you think about merging both Thank
Ulises Gutiérrez
you very much Chair I'm a victim of AI that's not my photograph, my name is Ulises Gutiérrez but I would just like to say something quite disruptive perhaps given what we were discussing this morning I'm always aware that we address the issue of AI as if we were in a context that was nothing was changing but unfortunately this is not what happens not only do we have to address the challenge of the momentum of technological development but also public policy is also always lagging behind technological development but always something that I think is very important there isn't alignment with technology we don't fully understand politically what is happening on the impact of technological development particularly when we are looking at AI I'd just like to underscore three points firstly, given this rapid technological development in the international context we see some relativism, that is to say in terms of our values and this is evident when we discuss issues of governance of such a powerful technology such as AI AI AI AI I obviously spoke about this previously it isn't what a human being can do with technology the question is rather what is a technology doing to human beings and I think this is still something that we have to look at even more closely secondly, we are working against an international backdrop where we see a separation between legality and what is legitimate here we're discussing preserving the rule of law uphold democracy and inclusion it doesn't really help us to be honest but also we have to be aware that the dynamic of technological development is working from the ground up because the ground is constantly moving as we know things are changing as Sigmund Freud once said we have a very changing landscape before us and we have a very changing landscape before us and we have a very changing landscape before us let me close by saying not only do we have to be convinced about the regulation of AI this is a constantly changing landscape but rather we should talk about governance and rather the adoption of a new social contract not only in the international context but also in national context but that is what I'd like to say for now, thank you
Wanjin Park
First of all, thank you it's my honor to be here and participate in such a valuable discussion Thank you While AI adoption accelerates, we are seeing vulnerable people exposed to harm at an equally rapid pace. So in terms of HRDD and KT, we defined AI risk, including human rights, especially for children and the elderly. We evaluate our AI models and agents based on that. And we run an executive deployment safety board before we release AI product. We publish the result in a technical report. But the problem comes when we interconnect multi -agents from different companies. Companies have their own definition of AI risk. Some agents may consider human rights more, but some may not. So... So I think to ensure that HRDB considers the impact on the most valuable, we need to have common and well -designed standards. So the work like AI risk taxonomy from the BTEC project at OHCHR is very important. I think we need to develop this work in a more concrete and practical way so that we can find a common baseline together with many stakeholders. Thank you.
Nighat Dad
Thank you, Your Excellency, to the co -chairs and to OHCHR for convening this cluster. I'll start with an honest assessment and contextualizing first panel and also intervenants who have spoken so far. Your Excellency, human rights due diligence, as most AI companies practice it today, does not really meet the standard of UN guiding principles on business and human rights. It happens after co -design decisions are made. It happens where regulations compels it. And it happens without the participation of the people actually affected. What we have in most cases is not due diligence. It's really a documentation. Three things are important to acknowledge here. First, human rights due diligence is applied unevenly across markets. Companies conduct... meaningful assessments where the law requires them to do, primarily in the EU, I would say. Then they deploy the same system in the majority world. A lot of us call it Global South, with really no equivalent process. And I would say this is a structural choice, not a capacity gap. The result is a two -tier rights regime determined by geography, where the depth of your risk assessment depends not on the risk you face, but on the jurisdiction you happen to live in. Second, the evidence of harm already exists. It is simply not consulted. There are so many digital rights organizations around the world, so many feminist digital rights organizations. The one that I run, Digital Rights Foundation, we helped a cyber harassment helpline for the last 10 years, and we documented more than, 23 ,000 cases of technology -facilitated abuse. disproportionately against women, girls, and young people. Non -consensual synthetic imagery, automated moderation that fails in Urdu, Hindi, Bengali, recommendation system that amplify gendered harassment. And let me tell you, when women withdraw from digital spaces, that is not only an individual harm, it is a collective harm to the public discourse and democratic participation. This is why technology -facilitated gender -based violence cannot be treated as a niche safety issue. Any humorized due diligence framework that does not explicitly address tech -facilitated gender -based violence has written half the population out of its protection. Yet, organizations holding this evidence are engaged after deployment for remediation, not before for prevention. That sequencing is core failure. And third, as AI becomes... more identic, a one -time assessment at launch is structurally ineducated. Harms emerge in deployment, in interaction, in iteration. Due diligence has to be continuous across the life cycle and capable of learning. So really three concrete recommendations. One, gender and child rights impact assessment must be mandatory. Conducted before deployment with affected communities participating, not merely documented. Second, human rights due diligence obligations must apply uniformly across all markets where a system is deployed. A risk assessment that covers Brussels, but not Pakistan, Kenya, or Brazil is not due diligence. And three, assessments must be repeated whenever capabilities materially change, including increases in agentic function. Thank you.
Clara Chappaz
Thank you. Thank you for those great solutions and for asking us to move in that direction. I'm now moving to Alvita Oatley from the AI panel.
Alvitta Ottley
All right, Your Excellencies, panelists, ladies and gentlemen, it's an honor to be here. And I'd like to offer a scientific perspective that I think addresses all three guiding questions today. So as I reflect on yesterday's discussion and the guiding questions, I realize that there are two related but distinct problems, scientific challenges, that we also brought up in our report. And that is evidence gap and evaluation mismatch. And I'm going to talk about evaluation mismatch first. And what I mean by that is are we asking the right questions? And similarly, are we asking the same questions? In engineering, when we naturally ask, can we make something faster? Can we make it more accurate? Can we make it more capable? And we've achieved remarkable things because of that. But these are not the same questions that societies, asking these are not the same questions that we are discussing today. Thank you. A parent doesn't wake up and ask if the model scored 94 % on the recent benchmark. No, they ask, can my child use this and can it decide when not to trust it? A physician, a doctor doesn't ask when they're using the model if it's the latest model. They ask if they can rely on it, especially when their patient's lives are at stake. The conversations in this room are not asking whether or not the AI is more capable than it was last year. We're asking whether human rights are protected, whether institutions remain accountable, whether society is better off because of it. And none of these questions are wrong. They're just simply different questions. The challenge is that when we're evaluating success, we evaluate these things differently based on the questions. And so what science has really taught us is if we can improve. If we can improve it, then we can improve what we can measure. And we can measure speed easily. We can measure accuracy easily. We can measure whether or not a system is more capable than it was. And so if society is deciding that success is about protecting human rights, fostering appropriate calibrated trust, preserving accountability or supporting better human decisions, then the outcomes must match what we're evaluating. And so I guess one of the questions and one of the things that we should be thinking about in this discussion is, do we know what we should be optimizing for? Do we know the evidence? Do we know what questions we should be asking? And can we coordinate with scientists so that we're asking the same questions? The second thing that I want to talk about is evidence gap. And we mentioned this in the report as well. Do we? Do we? We have the evidence to know whether or not we are achieving what we are optimizing for. And so I guess one of the things that we should be thinking about is evidence gap. So today, we have a lot of evidence about how well AI models perform, but we have very little evidence about how AI is affecting people's decision making. We know very little about how AI, whether or not and how well AI has helped teachers teach. We do highlight some of these anecdotal evidence in our report, but evidence is very limited because AI is very new. We know very little about how and how well AI affects clinical decisions, and we know little about how AI helps children learn. We know little about how AI affects public servants and how it affects their work. And so I guess what? What I'm trying to say here is that. we don't have enough evidence, and that is almost why we're here, and that's okay, but answering these questions require longitudinal studies, it requires partnership with organizations, it requires interdisciplinary collaborations, it requires careful work with protected and underserved communities. This evidence is growing, but it is still evolving. And so essentially what I want to leave you here with is just a different way of thinking about the relationship between society and science. Society has to decide what success looks like, and science has to determine how to measure it. And together, we have to build the evidence that tells us whether we're actually achieving it. So have we agreed on what success looks like? And so I think it's important to think about the relationship between society and science. Do we have the evidence to know whether we're achieving it? Because good decisions on the uncertainty doesn't come from eliminating uncertainty. They come from making our values explicit, our evidence rigorous, and our uncertainty understood. Thanks.
Clara Chappaz
Thank you. Thank you for this meaningful conversation. I think what we agree on collectively is that the speed at which the technology is developing in such a fast-moving environment fragmented geopolitical world, we actually need to define what success means for us humans. And depending on the values, we want to push forward, because that is only our decision, not the technology's decision. And based on that, we'll be able to evaluate better what are the frameworks that we want to see before the deployment of the models. that can guarantee human rights. But to do that, we do need participation from all the communities, in particular children, that they cannot be taken out of those discussions. And we also need to carefully address the local and global intersection so that everything we define works for everyone, not just a few markets. And I'll finish and wrap up with what you said. I think this is structural choices, not capacity. And if we don't do that, it would probably mean that either some communities will withdraw from digital spaces or they will create their own without control, without values, which is probably worst. We have only a few minutes left. So based on that, I'm going to ask each of you to say the one thing that you think we should do collectively after this gathering here in Geneva, starting with you, Your Excellency, Minister.
Willemijn Aerdts
Thank you so much, Ambassador. I think that we should share our good experiences and also try to learn from each other in the room. The Netherlands aims for the optimal use of AI and to safeguard the public values, and we do this, among others, by using the EU framework of the AI Act. And that also includes, and I'm very happy that we could have done that, includes a prohibition of sexual deepfakes that was actually added later to the process. So we conduct a risk -based approach based on human rights, transparency, risk managers. I love the addition of the evidence and scientific research. I think that's something that we should include, and we're very open to share our experience, but also to learn from you here in the room. Thank you.
Felipe Paullier
Thank you, Ambassador. And maybe as a final point, I would like to bring one of the key issues that constantly across regions, young people raise when we are discussing about AI and technologies, which is... the human dimension and the human impact. Young people definitely value technology. They use it. They innovate with it. They drive change in their communities through it. But what young people and citizens in general value most is belonging, is trust, is meaningful human relationships. So I would say that one of those critical areas where we need to emphasize, where we bring the scientific capacity, the value, but also the responsibility of those driving the innovations behind, the private sector, is what is the impact of these technologies in the well -being and the mental health of our communities. And for that, again, I would emphasize that it's not about writing or defining these solutions without the communities, especially. It's how we work not only for youth but with youth. So that's my message. Thank you.
Clara Chappaz
Thank you so much. Thank you very much, Your Excellency.
Ulises Gutiérrez
Just a few comments coming back to your question. I think we have to really underscore the question of values during this morning's discussion. We said that AI is not only a question of technology, but it's absolutely a question of values, of which I completely agree with as well. Well, secondly, I also think it is paramount that in the multilateral system we continue to bolster education, raising awareness, particularly of vulnerable groups, children, persons with disabilities, and the impact this technology has on human beings. Currently, I don't feel that we have a standard. Upon which we can educate people. Thirdly, also to underscore. technology as a tool, as an enabling tool for rights, but also technology as a right itself. I acknowledge that there is a right involved with this, and I think this is what we really need to address when we address this topic. Thank
Clara Chappaz
Thank you. Thank you so much, Mr. Park.
Wanjin Park
Thank you, Ambassador. I want to mention about children. Today's children are AI native, so they use AI naturally in search, recommendation, and learning in their daily lives. So I think it's important for them to accept it with critical thinking. That's why I believe the education for young people is really important. It's really needed. When school programs for education and discussion are connected with NGOs and relative committees, children can voice how AI affects them in a balanced way. So those voices can flow into governance discussions. Thank you.
Clara Chappaz
Thank you. Thank you so much, Mrs. Dad.
Nighat Dad
Ambassador, I believe two panels this morning, including ours, have given us a clear architecture due to diligence before deployment, independent oversight, and participation by those most affected. I just want to mention very concretely one issue, which is the evidence problem. The evidence problem is a resourcing problem. Civil society organizations or digitalized organizations in the majority world are the earlier sensors of AI -related harm. We have documented synthetic intimate imagery targeting women years before it entered global policy conversations. But this documentation is unfunded, precarious, and treated as anarchism. This is an anecdote rather than evidence. If this dialogue wants an early warning system for AI harms, This is an anecdote rather than evidence. This is an anecdote rather than evidence. This is an anecdote rather than evidence. This is an anecdote rather than evidence. This is an anecdote rather than evidence. I would like to tell the the folks who are here, that it already exists. It needs to be resourced and connected to decision-making, including the scientific panel, because scientific consensus takes time that affected communities, and scientific consensus takes time that affected communities really do not have. Thank you so much.
Clara Chappaz
And a final word from Ottley.
Alvitta Ottley
Thank you. So immediately two things come to mind. Oftentimes when we talk about AI, we talk about it like it's an adversary relationship, and I feel like that could be like a flawed perspective, and we need to realize that there's a lot of cooperation that could happen between the scientific world and society to make sure that we shape AI the way we want to shape it. But when we also talk about children, one of the things that also comes to mind is that there is also a methodological challenge related to studying children. So children is considered a part of a protective category of the population. So children, pregnant women, people in prison, there are limitations on what we can study and how we can study it. And I think that's one of the reasons why we see lack of evidence on the impact of AI on children. So that is a perspective that we don't really talk about a lot. But it also means that a lot of times we are developing things without having children involved. So I guess what I'm saying is that, yes, we have ethical safeguards, and we don't need weaker ones. But that is something that we should also bring into the picture. The consideration that we're talking about protected categories and protected categories in regards to also research. And we need to think about how do we navigate this space when we're talking about it. Thank you.
Clara Chappaz
Thank you so much to each and one of you. Thank you for the listening. Thank you so much, everyone. Have a good rest of the day.
Peggy Hicks
Thank you, Ambassador. Thank you, panel. I think we're getting a real taste of some of the issues here. But what I really appreciated here is that we've also got some very concrete ideas about what needs to come out of the dialogue. So I hope you're all taking that back. In particular, for those of you that want to look more deeply at the children's rights area. Oh, I'm sorry. I'm supposed to say, please, can we have the co -chairs back on stage for the next segment of the audience intervention? So welcoming co -chairs Banyo and Lopez back on stage. So I wanted to mention that we do have with ITU a discussion on Thursday at 3 .30, in which we'll be presenting a survey that we did called Me and AI of 1 ,000 children in 49 countries, sort of asking them what they think about this debate. That's one of the things we were told we needed more of. And one of the things they were saying to us is that they want AI to be shaped with us, not just for us, and they want it to respect their rights, and they don't just want to be shielded from it. So that's another place where we can continue the conversation. But with that, I'll turn it back over to the co -chairs for the next session of the audience interventions.
Oscar Lopez Agueda
Thank you. Okay, so after this brilliant panel, now we move to the ambassador to the United Nations, Angel Banjo. Angel Banjo from Bulgaria. Okay, so you move to the next government, this one? No this one. Okay, so the Secretary for Digital Policy, Social Communication, Secretariat, Presidency of the Republic of Brazil Joao Brand from Brazil
João Brant
Okay, thank you Hello, good morning everyone Thank you very much, Chair Some quick notes on behalf of the Brazilian government. Firstly, we've been prioritizing children and adolescents protection with the approval of the statute, the digital statute for protection of children and adolescents online where platforms must adopt child protection measures from the design phase The architecture of the platforms must limit features that can lead to excessive usage like infinite scrolling and companies are now required to implement accurate age verification systems, parental supervision tools, transparency obligations, and dedicated reporting channels for violating involving minors. Target advertising based on children's personal data is also prohibited. Those obligations closely relate to the responsible management of AI, since resources, features, and systems used by platforms employ artificial intelligence in their design, and content mode available in platforms by third parties might also employ AI, such as deepfakes. In addition to the Digital Statute for Children and Adolescents, Brazil released last month a decree to address violence against girls and women in online spaces, establishing proactive responsibilities for platforms and AI companies, including the prohibition of creating synthetic intimate images based on real third parties. This leads to my second point, my second note, which is related to liability. We think Brazil considers that private companies should have more reliability over third -party content, and this was a decision of our Supreme Court and now implemented and supervised by the executive branch with shared liability based on the guiding principles on business and human rights. And I would highlight two issues. Firstly, the due diligence process that has to prevent and mitigate risks related to serious offenses and rights violations, and the idea that they are co -responsible, they have shared responsibility for paid -for content. So in this case, they have to remediate harms they contribute to. Third, quick note, we should take a broad perception of human rights when discussing AI, and I would highlight information integrity as one of the main issues we should be looking for. Thank you. taking information integrity as part of the Article 19 of the government in a broad perception of access to information in its collective or social dimension. So the idea that we need an information ecosystem that can provide consistent, accurate and reliable information and we should be looking for that. And finally a quick note on the idea that the civil and political rights are also underpinned by economic arrangements and the way AI is now supporting and affecting the economic arrangement and sustainability of journalism is something that should be on our concern. Thank you.
Oscar Lopez Agueda
Thank you and thank you for introducing this point of view on journalism. It's really important misinformation also. And now we move to Ms. Anna Osterling, UN representative from the Global Forum for Media Development.
Anna Osterling
Thank you. Thank you. Thank you. My name is Anna Osterling, and I represent the Global Forum for Media Development, a network supporting over 200 media organizations worldwide. Excellencies, ladies and gentlemen, the million -dollar question today is, who controls our reality? In our everyday life, do we choose to make decisions based on fabricated or synthetic content, boosting profits and serving the political interests of a handful of mighty corporations based in a few countries? Or do we seek out independent reporting and evidence from public interest journalism? I've been engaging with the global dialogue since it was a mere idea on paper. I've heard much talk about information integrity and saving our democracies. Journalism, the institution designed to hold power to account with facts, seems to be largely missing in this dialogue. As my colleague Courtney Raj from the Center for Media and Digital Governance at OpenMarkets has said, in an information ecosystem, journalism is a keystone species. It stabilizes trust, anchors verification, structures accountability, and enables other institutions to function. Remove it, and the information ecosystem doesn't adapt. It collapses. This is about the concentration of power that has allowed big tech to lay claim to all of humanity's information and creativity, what we typically call content. AI systems are trained on journalistic content, which is scraped, without consent, without compensation, without credit. This is about AI infrastructure, which is increasingly concentrated with a few hyperscalers and digital platforms, reinforcing the power of AI. This is about reducing existing inequalities and undermining journalism's business models and public trust. The solution, journalism is critical infrastructure. This is not just about reading your Sunday paper. Every democracy, every market, every security architecture relies on trustworthy information. This requires a multi -layered governance approach. We need, one, public digital infrastructure, including cloud services that are accessible, affordable, non -discriminatory, and human rights compliant by design, driven by communities, including media, in the public interest. GFMD is providing space for the development of the journalistic stack. Two, industrial policies that ensure public interest AI, not just for the developers, but most importantly, for those using AI in the public interest, including media. And finally, AI governance must require all human rights are embedded across a full life cycle of all AI systems. AI technologies. and AI governance must also support economic sustainability of
Linda Bonyo
Excellent. Thank you so much, Anna, from Global Forum for Media Development, on highlighting on who's missing in this room. I'd like to mention Freedom Wangi is an AI worker who was supposed to be here but was unable to get a visa. And it's useful to say who's missing and how to involve them the next time. Thank you also for highlighting on DPI, on public DPI. We need digital public infrastructure so that we do not have dependencies on big tech. And at this point, I would like to invite the Republic of Korea, His Excellency Jihoon Cha, Permanent Representative. Forgive me if I butchered your name. Excellent. A round of applause. He has a long walk to do.
Jihoon Cha
Thank you. Thank you Madam Co -Chair Distinguished Excellencies and distinguished participants the Republic of Korea thanks the Co -Chairs for convening this conversation on AI and human rights No one can truly benefit from AI while living in pure harms such as defects, disinformation discrimination and surveillance and these dangerous all hardest on the most vulnerable those least able to push back including children Without our proper and timely attention this problem will only deepen as AI advances The Republic of Korea enacted AI Basic Act this year to reduce the risk of a comprehensive legal framework on AI. Under this AI Basic Act, the operators of high -impact AI systems have an obligation to assess the potential impact of AI systems on Fundamental Human Rights Act. At the same time, given the complexity and nobility of this issue, these efforts would be well complemented by a balanced mix of regulation and voluntary measures, and by including various stakeholders in the discussion. Protection of human rights in AI must also be universal. In this vein, the Republic of Korea has endeavored to lead the adoption of Human Rights Council Regulation 59 -11 on new and emerging digital technologies and human rights. This resolution highlights the importance of promoting everyone's right to enjoy the benefits of new and emerging digital technologies. The Republic of Korea considers access to AI not as a privilege to be earned, but as a basic right to be enjoyed. As the scientific panel report suggests, the distance between our respective endeavors is noted. As countries deepen their readiness for AI, so does their approach to how to govern it. And globally, these efforts remain pigmented. That is why the Republic of Korea declared our vision for establishing the Global AI Hub, together with nine participating AI and related organizations, to contribute to strengthening of global AI capacity. In conclusion, we must keep sight of what all of this is for. The promise that humanity sees in AI is not in how parallel technology advances. It is in how...
Linda Bonyo
Thank you so much, Your Excellency. At this point, it is my honor to have a voice from Africa, Research ICT Africa, Ms. Pria Chetty, the Executive Director of RIA. Priya is in the room.
Pria Chetty
Excellencies, distinguished colleagues, Research ICT Africa, a digital policy think tank working across the African continent and in solidarity with global majority partners, welcomes the opportunity to make this submission. We engaged... Today in a spirit that is both constructive and urgent. because the stakes for the majority world could not be higher. We submit this input grounded in the Research ICT Africa Just AI framework, which reframes global AI governance by centering justice as the core quality that AI systems must embody in the public interest. A growing global community of Just AI scholars and practitioners hold that prevailing governance models anchored in notions of ethical or responsible AI remain structurally inadequate. As endorsed in various conversations this week, such frameworks are largely self -regulatory, driven by dominant actors and calibrated to high -income countries, disregards sovereignty or agency, and reinforces vastly different socio -economic and institutional realities. Such frameworks deepen existing inequalities, extracting value from individuals, communities and nations without reciprocal benefit, input or control. For children, these effects are both current and with long -term implications for the future realities. Just AI offers the global dialogue a normative framework to correct this trajectory. The focus, active intervention and specific allocation of resources to guide societal level shifts, materially bringing the development and prosperity of local populations into view and the planetary and sustainability impacts of AI. Building on the framework and the commitment in paragraph 55 of the Global Digital Compact to inclusive multi -stakeholder participation and drawing on the foundational language of Resolution 71, we propose the following priorities. Establish a common normative foundation through Just AI that moves beyond aspirational ethics to economic policy and legislative normative frameworks rooted in international human rights law and actively informing economic and social justice. Establish a globally recognized understanding that the governance of AI is inseparable from the governance of data. The separation of data governance and AI governance in international forums obscures the foundational data justice that must inform all AI systems. Data justice is crucially informed by a right of access to data in substantive and procedural forms. For equitable AI that accords with first and second generation rights, we must democratize AI resources and capabilities so that all nations, particularly those in the global majority, and specifically small and medium enterprises, shift from being passive consumers of AI systems to becoming co -creators of AI that reflect their developmental priorities, democratic democratic values. Thank you very much.
Oscar Lopez Agueda
Now we move on. We move to Mark Cassayre . He's a permanent observer of IDLO to the UN in Geneva.
Mark Cassayre
As the only global intergovernmental organization exclusively devoted to promoting the rule of law the international development law organization excuse me has extensive experience supporting countries to harness digital innovation to expand access to justice strengthen governance and improve public service delivery digital technologies can make justice services more accessible streamline court processes and bring institutions closer to the people they serve yet rapid adoption of AI and digital systems is often outpacing the legal regulatory and institutional frameworks needed to ensure that these very technologies protect and promote people's rights without adequate safeguards the use of AI systems and decision making improve the quality of life is a key to the development of intergovernmental intergovernmental intergovernmental may reinforce existing biases, produce discriminatory outcomes, and fail to account for essential contextual nuances. These are risks that can make justice unattainable. The challenge is therefore not technological. It is one of governance and, as we heard this morning, values. In this regard, IDLO offers three recommendations. First, human rights and the rule of law must underpin the development and deployment of digital technologies, notably in the justice sector. AI systems must be designed in a way that expands access, serves people's needs, and prevents new forms of exclusion and inequality. Second, fit -for -purpose legal and regulatory frameworks must accompany innovation. Effective regulations. Legal, regulation, transparency, accountability, and human oversight are not obstacles to innovation. They are the foundations of trustworthy and sustainable innovation. As countries modernize their justice systems and public institutions, clear rules and responsibilities are essential to ensure that AI strengthens rather than undermines public confidence. Third, we must invest in institutional capacity. This includes strengthening public digital infrastructure for justice delivery, enhancing AI literacy among legal and justice professionals, and supporting judicial services in a way that manages the risks of AI. These efforts must aim to enhance efficiency and access to justice, particularly for those who are most vulnerable or marginalized. We collectively have a responsibility to ensure that human rights, accountability, transparency, and meaningful human. Oversight remain at the heart of AI governance. Ideola looks forward to supporting governments and other stakeholders to anchor these important values in AI governance. Thank you.
Oscar Lopez Agueda
Thank you very much and now we move to Mr. Raman Jit Singh Chima, Global Programme Director from the Association for Progressive Communications
Raman Jit Singh Chima spokesman
Hello, distinguished guests. I'm speaking on behalf of Mr. Raman, the program manager for the Association for Progressive Communication. My name is Rebecca Rektimbo, and I'm the Connectivity Technical Projects Coordinator for LOCNET, a collaboration initiative of the Association for Progressive Communication and Rhizomatica. APC is a membership organization working mostly within the global south. When we talk about inclusive AI, we often focus on biases, governance, data sets, and representation. These conversations are essential, but I would like us to begin with a different question. Who has the connectivity to participate in AI in the first place? Before communities can contribute data, build local AI solutions, influence policy, or benefit innovation, they need meaningful, affordable, trusted, and locally relevant connectivity. Community-centered connectivity is more than providing internet access It is about communities owning, managing and governing the infrastructure that connects them It creates a space for local knowledge, local languages and community priorities to thrive while ensuring that people have real urgency over their digital lives This matters because connectivity is not just a technical issue It is a question of equity, power and justice As the gender in digital coalition feminist guiding principles of global AI governance remind us meaningful AI governance requires community participation, data sovereignty, diverse expertise and investment in community-led digital infrastructure Not simply mere powerful AI systems It calls for the recognition of the importance of digital technology lived experiences and community knowledge as expertise and ensuring those most affected have influence over AI governance. Community -centered connectivity makes these principles possible. It shifts power closer to communities and enable women, indigenous people, rural communities, and intervenants of underrepresented languages to become creators, innovators, researchers, and decision -makers, not simply consumers of technology. As government researchers and international organizations and the private sector invest billions in AI, we must invest intentionally in community infrastructures that make equitable AI possible. I would like to leave you with one thought. Communities are not the last mile of connectivity, or they're not the last mile of AI. They're actually the first mile. So let's think of communities when we think and restructure governance of AI.
Linda Bonyo
Thank you. Thank you so much, Rebecca, for highlighting communities. Zimbabwe's national AI strategy is focused on Ubuntu. It's not solitary. We are together. I am because you are. At this point, I'd like to invite Tech Global Institute, Ms. Shumaila Hussain Shahani , Policy and Advocacy Lead. You have the floor.
Shumaila Hussaini Shahani
Hi, and thank you for the opportunity to speak here. I speak on behalf of Tech Global Institute. We are a policy lab with a mission to reduce equity and accountability gaps between technology platforms, governments, and the communities in the global majority. The principles that we have so extensively spoken about today here only protect human rights if they run the full length of AI value chain, from the data workers who label training data to the person denied welfare or wrongly flagged by a system. They cannot contest. Today, the harms that concentrate, the harms concentrate where the power to challenge them is the weakest. We therefore propose five concrete commitments. First, safety claims must be accompanied by the conditions under which systems were tested, known limitations and failure rates disaggregated by language, gender, disability, region, skin tone, and more. Disclosure of training data provenance must be standard. We cannot hold a system accountable if we cannot see where it fails. Second, a decade of voluntary commitments has shown that it has not delivered. We need public accountability frameworks and corporate liability that follow the multi -actor chains through which AI is built and deployed with accessible remedy for affected people wherever they live. Third, global majority countries largely import models trained elsewhere and deploy them in high -stakes settings without the means to evaluate them first. We call for publicly financed, regionally hosted capacity to evaluate AI systems before they are deployed and often open benchmarks built with and for under -resourced languages so oversight is not outsourced to the same few actors it is meant to oversee. Fourth, transparency and oversight must extend to algorithmic management, data and platform workers, a workforce in which women carry disproportionate harms, need existing labor protections extended to them, including across borders. Fifth, indigenous and traditional knowledge is being absorbed into training data without consent, recognition or return. Community -led data stewardship must therefore be recognized as part of human oversight. Lastly, compatibility between our approaches globally do not require uniformity. We can have a common floor of rights, safeguards which includes disclosure, contestability and remedy for every data. We can have every person affected, but with genuine pluralism above that floor. Thank you.
Linda Bonyo
Excellent, thank you Thank you so much and I think your sentiments resonated and made me think about the diaspora especially asylum seekers who left their countries for human rights violations and how is AI pacifying those issues so thank you At this point I'd like to invite Ms. Esther Eghobamien-Mshelia from the Committee on the Elimination of Discrimination Against Women Excellent, a round of applause as she comes on stage A round of applause It's been established that hungry people are angry people so let's give it up
Esther Eghobamien-Mshelia
Excellencies and distinguished colleagues in the context of this global dialogue one message is clear there can be no truly human rights based AI governance without the full implementation of the convention on the elimination of all forms of discrimination against women CEDAW AI as we hear and know is already influencing decisions about every sector and every sphere of life including public administration and access to information technologies used are not neutral when developed using biased and non -representative data or deployed without adequate safeguards they often reproduce and amplify the structural inequalities that women and girls continue to face CEDAW provides an essential framework for addressing these realities CEDAW is the world's leading AI development and development organization CEDAW is the world's leading AI development and development organization CEDAW is the world's leading AI development and development organization It requires states not only to prohibit discrimination, but to achieve substantive equality by tackling structural and regulatory barriers, stereotypes, and unequal power relations. It also recognizes that discrimination is often compounded by factors such as race, disability, age, migration, status, and poverty, and that tailored and proactive action is necessary to curb this intersectional discrimination. Today's discussion themes are at the heart of the CEDAW Convention. Transparency is necessary to identify discriminatory impacts, address gender wash, and provide the right ecosystem for effective redress. Accountability ensures that states and private actors can be held responsible when AI undermines women's rights and better prevent automated decision -making from entrenching or reinventing discrimination. CEDAW tools, including its general recommendation number 40, also remind us that women must participate fully and equally in the design, governance, and oversight of AI systems. AI cannot be truly inclusive if half the world's population, their perspectives and data about them is underrepresented or outrightly excluded. We must also ensure that progress is measured. Gender responsive indicators, monitoring, and surveillance are critical to demonstrating whether women are participating and benefiting equally and are also constructively benefiting from digital transformation, are protected from harms such as algorithmic discrimination, technology -facilitated gender gender-based violence, and manipulative consumer exploitation. Measuring these realities is an important part of ensuring accountability for the commitments and obligations states have already undertaken under CEDAW. CEDAW and its optional protocol already provides a clear...
Oscar Lopez Agueda
sorry sorry sorry sorry now we have an important voice it's Elizabeth Tan from the UN High Commission for Refugees she's the Director General
Elizabeth Tan
thank you Chair Excellencies ladies and gentlemen UNHCR speaks from its international protection and statelessness mandate and from the experience of refugees asylum seekers internally displaced and stateless people international human rights law already gives us a framework to assess AI related harms the challenge is how to put it into practice Accountability remains uneven and regulation has not kept pace. For forcibly displaced and stateless people, barriers are especially high. Documentation, language, digital access, insecure status and fear of approaching authorities. AI governance must therefore be practical, inclusive and accessible to those most affected. This is particularly important in high stakes areas such as asylum and border management. If AI supported tools assess risk, verify identity, translate interviews or support case triage, an error can affect access to territory, documentation or access to asylum procedures. Yet the person concerned may not know how. The decision was reached. or how to challenge it. Human oversight must be real, informed, and able to change outcomes. We're also seeing how AI -generated disinformation and online hate can quickly become protection risks. In Libya, violent hate speech and dangerous misinformation, including AI -generated content, incited hostility and abuse of refugees, and affected the safety of UNHCR and humanitarian workers supporting displaced people and Libyan host communities. The Rohingya, they too continue to face dehumanizing narratives online, even in exile. This kind of hate can fuel abuse, exclusion, violence, and fear, and can deepen divisions between displaced and host communities. These examples point to the need for practical safeguards that work in real humanitarian contexts. For UNHCR, three safeguards are essential. First, human rights due diligence across the AI lifecycle, from design to deployment and monitoring. Second, bias testing, clear explanations of how decisions are reached, and accessible ways to complain or seek correction.
Oscar Lopez Agueda
Sorry, you're out of time. Thank you. And now we move to Ms Sopio Kiladze, from the Committee on the Human Rights. The rights of the child.
Sopio Kiladze
Sopio, miss, yeah. Doesn't say anything, didn't know. Thank you, Chair. Excellencies, distinguished delegates, dear colleagues, let me begin today with children. Children benefit from AI without doubt, but they are really very much exposed to all sort of violations of children's rights, maybe the most affected from all vulnerable groups. Numbers of child sexual abuse, mental health issues, all sorts of violence, privacy violations, and many, many other violations of their rights are the numbers are skyrocketing. And children are not simply future citizens. First of all, they are rights holders of today. and their rights do not disappear when they enter the digital world. That's why the United Nations entities came together to develop the joint statement On AI and the Rights of the Child, launched in 2026, recently, co -led by ITU, the CRC, and UNICEF, and which unified 13 UN agencies as co -signatories and over 60 different organizations across the globe. And most importantly, the children from all five regions of the United Nations themselves helped to shape it. They told us what gives them hope, what worries them, and what they expect from us. We listened and we reflected. I also would like to... warmly welcome the joint statement ahead of the global dialogue, which was recently issued, supported by over 120 experts and organizations across the globe. Due to the fact that the children are not developed fully yet, depending on their age, maturity, and their evolving capacities, they need special care, and we are those who are responsible for them. The question is whether there is a will, a will from states, the companies, and international organizations to design, to develop, and to govern AI that respects children's rights. History will not judge us by how intelligent our AI or machines are, but it will judge us by something far more important, namely, if we use this extraordinary moment. To protect those who had the smallest voice, but the greatest stake in the future. I'm sure if there is the will, then we can shape AI in line with the Convention on the Rights of the Child. the future on artificial intelligence is a choice
Linda Bonyo
thank you so much for speaking on behalf of the children and now to a final speaker we do have from the institute for global negotiation we do have Jerome who is a senior fellow and joining us today the floor is yours
Jérôme Bellion-Jourdan
thank you very much co -chairs and to the joint secretariat for this opportunity I'm speaking today on behalf of the institute for global negotiation I'm also one of the vice chairs of the AI for good impact initiative steering committee I would like to offer some reflections on the work that we're doing at global negotiation support an initiative to support chairs co -chairs of negotiation projects processes in the UN and you may have seen a piece that we published in past blue for recommendations to the next secretary general if I may, this is not a statement I was a diplomat myself I'm used to delivering statements but just a few reflections from what we heard since yesterday from UNSC Guterres and his call for more governance and I think everybody in the room would say yes we need more governance but the question is how to get there and that's where if we look at negotiation, I mean artificial intelligence by the way is an amazing tool in support of multi -level negotiations so let's not forget that but if we want to use it properly we need also to see how negotiation can be used and leveraged in support of the steps towards global governance of AI we've heard repeatedly since yesterday and in many other fora that geopolitics competing interests are an impediment to work towards global governance of AI and I would argue that if we leverage the power of negotiation, it would help us with all the negotiation tools and techniques to navigate the power politics, to navigate the power asymmetry, and this is what multilateral negotiation is about. And that would allow us to get to formats that allow us to get to forge common ground. But this has to be engineered, and that's where, you know, for the work that we're doing, we see the need to have, to invest in process design, process management, innovative formats of negotiations. It will not be enough to lock in 193 states in a room and the stakeholders not far from a room or at the table and get to agreement. What is needed is to shift from positional bargaining, red lines, to integrative negotiation, to from win -lose that we see too often on the global stage to win -win. And that's why we're working in formats of negotiations where we can reconcile many aspects related to AI. We see AI for innovation and opportunities, but how do we reconcile this with the impact on the environment, with the risks when it comes to human rights and other safety risks? So that's where negotiation can be leveraged, and we stand ready to support all states and stakeholders moving
Linda Bonyo
Excellent. Thank you very much. Appreciate it. I think my final comments will be one is who finances human rights? We are at a juncture where we need the money to ensure that human rights ideally works in the age of AI. And then also number two is to point you again to the Africa AI Governance Index by the Lawyers Hub, which I represent, to see what AI workers across the African continent are talking about and are impacted by training these models, the mental health toll on them, and how do we continue to speak for them even when they're not in the room. And therefore, we must look at algorithmic openness around visa processes. It is time that we talked about transparency. for the algorithms that already determine whether people travel or not, because we are seeing a lot of human rights violations. And I want to ask the global north to reconsider the kind of algorithms that we outsource to determine whether people travel here or not. There's a great report by an organization based in Geneva, I forget their name, but they said this, that 51 % of AI governance conversations this year happen in Geneva. And that means that the rest of the world were ideally locked out because they could not access visas to this place. So I want to ask us to reconsider what that means. Thank you very much.
Oscar Lopez Agueda
Okay, so I would also like to thank the United Nations and our coaches and all the participants for this debate. I think we all in the room are fully aware of the importance of this debate. Last. General Assembly of the United Nations last September. we decided to we decided to create this debate and to create also a scientific panel and we've seen the first things of this scientific panel and this is like the first world meeting discussing on something that it is not the last technology. It is not a question only of engineers it is a question also of philosophers because as you all know in this hall we are discussing about democracy we are discussing about human rights, our privacy our energy our peace in the world about everything so as we know that this changes everything I know that I think that we all know that it is not about doing something or not it is that we are late we are late so we will keep on moving to build a humanistic AI because it is about defending ourselves, about getting the best cases and using AI for being better, not for being worse. Thank you very much.

Avertissement : Il ne s'agit pas d'un compte rendu officiel de la session. DiploAI génère ces ressources à partir d'enregistrements audiovisuels ; elles sont présentées telles quelles, y compris d'éventuelles erreurs. En raison de contraintes logistiques (audio/vidéo ou transcriptions), les noms peuvent être mal orthographiés. Nous nous efforçons d'être aussi précis que possible.