WSIS Forum 2026
Rapport généré par l'IA

The Future of Education and Research in the AI Era: Equipping Young People for Tomorrow

13 intervenants
Résumé

Résumé

Cette table ronde a réuni des universitaires, des étudiants et des praticiens pour explorer la manière dont les universités devraient préparer les étudiants au monde du travail à l'ère de l'intelligence artificielle . Les panélistes représentaient des institutions couvrant les domaines du droit, de l'ingénierie, de l'informatique et de la politique numérique, notamment l'Université de Genève, le MIPT, l'Université des sciences et technologies de Huazhong, et la Royal Holloway University of London . La professeure Christine Kaddous, de l'Université de Genève, a souligné que l'IA devrait jouer le rôle d'assistant plutôt que de substitut à la pensée critique et au raisonnement juridique , notant que l'université promeut activement la formation continue et l'apprentissage tout au long de la vie pour répondre aux besoins des étudiants actuels comme des professionnels déjà en activité . Le professeur Azamat Zhilokov du MIPT a présenté un modèle éducatif reposant sur trois piliers : la sélection d'étudiants talentueux, le renforcement des bases en mathématiques et en physique, et leur intégration dans des recherches de pointe . Les principaux domaines de recherche de son institut comprennent les architectures d'IA économes en énergie, l'interprétabilité, les modèles neuronaux du monde et la robotique . Le professeur Tim Unwin a offert une perspective plus critique, affirmant que la plupart des universités sont devenues des institutions axées sur les diplômes plutôt que sur l'apprentissage véritable , et insistant sur la nécessité urgente d'une formation technique et professionnelle de qualité, notamment au regard du fait qu'un enfant sur dix dans le monde reste non scolarisé . Le professeur Yong Xiao a mis en lumière le fossé croissant en matière d'équité engendré par des outils d'IA coûteux, inaccessibles aux nations les plus pauvres , et a présenté le programme « entreprise individuelle » de l'Université de Huazhong, qui dote chaque étudiant d'outils d'IA et de financements pour mener des projets en autonomie . Les étudiants Alexander Schrier, Joseph Devin et Ava Mitzi ont partagé leurs expériences concrètes issues de projets de fin d'études réalisés avec l'UIT, soulignant l'importance d'un regard critique lors de l'utilisation de l'IA, notamment pour gérer les hallucinations, vérifier les sources et éviter la capitulation cognitive . La professeure Christine Kaddous a également insisté sur la transparence dans l'utilisation de l'IA et sur la centralité de la responsabilité professionnelle, en particulier dans les activités juridiques impliquant des clients . La discussion a convergé vers l'idée que, si l'IA offre des opportunités transformatrices sur les plans éducatif et professionnel, les compétences fondamentales que sont la pensée critique, la vérification rigoureuse et l'utilisation responsable demeurent indispensables , Azamat Zhilokov a cité l'observation du chercheur en IA Andrej Karpathy selon laquelle « on peut déléguer la réflexion, mais on ne peut pas déléguer la compréhension » .

Points clés
  • Points clés

  • Objectif général

  • La discussion est une table ronde tenue lors d'un forum international (vraisemblablement lié à l'UIT et au SMSI), réunissant des professeurs d'université, des chercheurs et des étudiants de plusieurs pays pour explorer la manière dont les établissements d'enseignement supérieur devraient adapter leurs programmes, leurs priorités de recherche et leurs approches pédagogiques afin de préparer les étudiants à un monde en rapide évolution, façonné par l'IA. La session examine également des questions plus larges relatives à l'équité numérique, à l'utilisation responsable de l'IA et au rôle des organisations internationales dans la réduction des écarts de connaissances.
  • --
  • Principaux points de discussion

  • Le rôle de la pensée critique et de l'utilisation responsable de l'IA dans l'éducation : Plusieurs panélistes ont souligné que l'IA devrait fonctionner comme un assistant plutôt que comme un substitut au raisonnement autonome. La professeure Christine Kaddous a insisté sur le fait que les étudiants doivent développer des réflexes éthiques et un raisonnement juridique indépendants des outils d'IA , tandis que le professeur Azamat Zhilokov a fait écho à cette idée en citant la distinction établie par Andrej Karpathy : « on peut déléguer la réflexion, mais on ne peut pas déléguer la compréhension » . Des étudiants de Sciences Po Paris ont renforcé ce point à travers leur expérience de projet de fin d'études, notant que l'IA produisait fréquemment des hallucinations dans les références bibliographiques et attribuait erronément des citations, nécessitant une vérification humaine constante .
  • L'équité numérique et le risque d'élargissement des fractures mondiales : Le professeur Tim Unwin a invité le panel à faire face à la réalité selon laquelle un enfant sur dix dans le monde n'est pas scolarisé, remettant en question la capacité des discussions académiques d'élite à traiter adéquatement ce problème . Le professeur Yong Xiao a développé ce point en soulignant que l'accès aux modèles d'IA de pointe est coûteux, ce qui signifie que les communautés rurales et les plus pauvres sont de plus en plus laissées pour compte, créant un fossé « encore plus grand » que la fracture numérique préexistante . Le professeur Tim Unwin a également soutenu que l'enseignement et la formation techniques et professionnels (EFTP) sont chroniquement sous-financés, malgré leur plus grande pertinence pour la vie professionnelle de la majorité des personnes .
  • Les orientations de la recherche en IA de pointe et leurs implications sociétales : Le professeur Azamat Zhilokov a présenté les principaux domaines de recherche du MIPT, notamment l'efficacité énergétique et le coût de l'inférence en IA, l'interprétabilité et l'explicabilité des grands modèles de langage, les réseaux de neurones à inspiration physique, l'IA pour les sciences naturelles, et les systèmes de contrôle en robotique . Le professeur Yong Xiao a ajouté que l'IA transforme les cycles de normalisation internationale - qui opéraient auparavant sur des décennies - et que les professionnels de l'ingénierie doivent assumer une plus grande responsabilité compte tenu des capacités accrues et des risques de l'IA, notamment son utilisation croissante dans les applications militaires .
  • L'apprentissage tout au long de la vie, la formation continue et le recyclage des actifs : La professeure Christine Kaddous a mis en évidence le double défi consistant à former les étudiants actuels tout en soutenant les professionnels déjà en activité grâce à des programmes de masters exécutifs et des modules de formation continue, tels que le module de gouvernance numérique du programme MEIG de l'Université de Genève . Elle a également soulevé la préoccupation émergente selon laquelle les meilleurs outils d'IA ne sont plus gratuits, créant de nouveaux obstacles à l'accès équitable pour les apprenants tout au long de la vie .
  • Les partenariats étudiants-institutions et la diffusion des connaissances par les organismes internationaux : L'étudiant Alexander Schrier a décrit comment un projet de fin d'études mené avec l'UIT - portant sur l'élaboration d'une norme pour mesurer l'impact environnemental des centres de données d'IA - a conduit à l'intégration d'une recommandation dans l'UIT-T L.1801 et à sa présentation au Forum de la jeunesse de l'ECOSOC en 2025 . Il a présenté l'UIT comme un « organisme de diffusion des connaissances » et a soutenu que l'élargissement de tels partenariats avec les jeunes est essentiel pour des progrès significatifs . Les étudiants de Sciences Po ont également souligné que leur projet de fin d'études lié à l'UIT avait renforcé leur rigueur en matière de discipline des sources et de vérification, notamment parce que leurs conclusions étaient destinées à alimenter de véritables travaux de normalisation .
  • --
  • Ton général

  • La discussion s'est ouverte sur un ton collégial et informatif, les panélistes partageant leurs perspectives institutionnelles de manière structurée et respectueuse. Au fil de la session, le ton est devenu nettement plus direct et parfois provocateur - notamment lors des interventions du professeur Tim Unwin, qui a directement interpellé le panel en affirmant que « la plupart des universités ne sont que des entreprises » et que de nombreux étudiants y assistent uniquement pour obtenir des diplômes . Cela a introduit une tension productive dans la conversation. Le professeur Yong Xiao et les étudiants intervenants ont adopté un ton plus ancré et pratique, s'appuyant sur des exemples concrets tirés de leurs propres recherches et projets de fin d'études. Vers la fin, le ton s'est fait plus réflexif et légèrement philosophique, le professeur Azamat Zhilokov et le professeur Tim Unwin mettant tous deux en garde contre une dépendance excessive à l'IA au détriment du développement cognitif authentique . Les remarques conclusives du professeur Tim Unwin - plaidant pour le « droit à la déconnexion » et mettant en garde contre la « démence numérique » - ont donné à la session une conclusion stimulante et quelque peu préventive .
Intervenants
CK
Christine Kaddous
141 wpm · 11 min
AZ
Azamat Zhilokov
151 wpm · 7 min
YX
Yong Xiao
187 wpm · 12 min
TU
Tim Unwin
171 wpm · 7 min
AS
Alexander Schrier
206 wpm · 3 min
AM
Ava Mitzi
140 wpm · 3 min
JD
Joseph Devin
222 wpm · 2 min
AM
Audience Member 5
148 wpm · 53 s
AM
Audience Member 1
114 wpm · 11 s
AM
Audience Member 2
135 wpm · 29 s
AM
Audience Member 3
168 wpm · 36 s
AM
Audience Member 4
109 wpm · 1 min
M-
Moderator - Regina Valiullina
119 wpm · 10 min

Résumé élargi : L'éducation et la recherche à l'ère de l'IA - Table ronde

#

Aperçu de la session et participants

Cette table ronde animée par un modérateur, tenue lors d'un forum à Genève vraisemblablement lié aux événements du SMSI et d'AI for Good , a réuni des universitaires, des étudiants et des praticiens de plusieurs pays afin d'explorer comment les établissements d'enseignement supérieur devraient adapter leurs programmes, leurs priorités de recherche et leurs approches pédagogiques pour préparer les étudiants à un monde en rapide évolution, de plus en plus façonné par l'IA . La session s'est déroulée en deux parties : une discussion en table ronde au cours de laquelle des questions ont été posées à chaque panéliste à tour de rôle, suivie d'un échange ouvert avec le public . Les participants représentaient un éventail délibérément large de disciplines et de contextes institutionnels, notamment le droit et la gouvernance, la physique et la recherche en IA, l'ingénierie et les communications sans fil, ainsi que la politique numérique internationale .

Les panélistes comprenaient la professeure Christine Kaddous, directrice du programme de master en gouvernance européenne et internationale à l'Université de Genève ; le professeur Azamat Zhilokov, directeur de l'Institut d'intelligence artificielle à l'Institut de physique et de technologie de Moscou (MIPT) ; le professeur Yong Xiao de l'Université des sciences et technologies de Huazhong en Chine ; et le professeur Tim Unwin de la Royal Holloway University of London . Les perspectives étudiantes ont été apportées par Alexander Schrier, étudiant en master à la Johns Hopkins School of Advanced International Studies, spécialisé dans l'intersection du droit et de la politique numérique , ainsi que par Joseph Devin et Ava Mitzi, étudiants en master à Sciences Po Paris, qui ont participé en ligne .

---

#

L'Université de Genève : gouvernance de l'IA, formation continue et esprit critique

La professeure Christine Kaddous a ouvert la discussion de fond en situant l'Université de Genève dans le paysage plus large de la gouvernance mondiale de l'IA. Elle a souligné que Genève, en tant que ville hôte du troisième Sommet sur l'IA - après Paris et Delhi - a placé l'université au cœur des discussions internationales sur la gouvernance de l'intelligence artificielle, l'université participant activement aux côtés des organisations internationales, du secteur privé, des ONG et d'autres institutions académiques . Ce positionnement institutionnel, a-t-elle soutenu, façonne à la fois l'agenda de recherche de l'université et ses priorités pédagogiques.

Au sein de la faculté de droit en particulier, Christine Kaddous a décrit une approche proactive de l'IA qui va au-delà de son traitement comme un phénomène purement technologique . La faculté s'intéresse à l'ensemble des implications juridiques de l'IA, notamment la propriété intellectuelle, la responsabilité et l'élaboration de politiques de gouvernance . De manière significative, l'université a adopté une position favorable à l'utilisation responsable de l'IA comme outil pour la communauté académique, y compris comme moyen de réduire les tâches répétitives et administratives . En parallèle, Christine Kaddous a identifié deux défis éducatifs distincts : préparer les étudiants actuels et soutenir les anciens étudiants déjà dans le monde du travail par le biais de la formation continue . Le programme MEIG, par exemple, comprend un module dédié à la gouvernance numérique qui s'adresse à la fois aux étudiants en formation initiale et aux étudiants en formation continue .

Le point le plus marquant de Christine Kaddous concernait le développement de l'esprit critique. Elle a observé que les étudiants utilisent désormais l'IA en permanence, y compris pendant les cours, et que le risque d'une dépendance non critique est aigu . Sa réponse consiste à pousser les étudiants à penser de manière indépendante et éthique, et à utiliser l'IA comme un assistant qui améliore leur travail plutôt que comme un substitut à leur propre raisonnement . Elle a donné l'exemple concret de la capacité de l'IA à résumer des centaines de décisions juridiques ou de sources bibliographiques en quelques secondes, soulignant que ce qui est véritablement requis, c'est une évaluation critique de ce que l'IA produit . Sa formulation synthétique « l'IA est un assistant, mais elle ne remplace pas l'analyse, le raisonnement juridique ou la rigueur intellectuelle » est devenue un point de référence récurrent tout au long de la session.

---

#

Le modèle à trois piliers du MIPT et la recherche en IA de pointe

Le professeur Azamat Zhilokov a présenté un compte rendu détaillé de l'approche du MIPT en matière de formation à l'IA, en l'articulant autour de ce qu'il a décrit comme le système FISTECH à trois piliers . Le premier pilier consiste à identifier et sélectionner les étudiants les plus talentueux de toute la Russie, avec seulement environ 1 000 étudiants admis chaque année en licence . Le deuxième pilier est un cycle fondamental rigoureux de trois ans couvrant l'informatique, les mathématiques et la physique, qu'Azamat Zhilokov a décrit comme véritablement exigeant et conçu pour développer l'esprit critique - un objectif qu'il a explicitement aligné sur les remarques antérieures de Christine Kaddous . Le troisième pilier consiste à rattacher les étudiants à des professeurs de premier plan pour une recherche pratique dès leur troisième année, avec un objectif minimum de deux à trois publications dans des revues de premier rang au moment où ils terminent leur licence ou leur master .

Abordant les priorités de recherche spécifiques du MIPT, Azamat Zhilokov a identifié deux grands domaines : les modèles et les agents (y compris l'IA agentique), et la robotique . Dans le premier domaine, il a mis en avant l'efficacité énergétique et la réduction des coûts comme un domaine particulièrement pressant, notant que les dépenses d'investissement et d'exploitation consacrées à l'infrastructure de l'IA sont considérables et que la charge pesant sur les systèmes énergétiques est significative . Il a également identifié l'interprétabilité, l'explicabilité et la fiabilité des systèmes d'IA - y compris le développement de garde-fous pour les grands modèles de langage - comme une frontière de recherche majeure et croissante . Un troisième domaine de concentration est ce qu'il a appelé les modèles du monde neuronaux : des systèmes qui encodent les lois et dépendances physiques réelles, permettant à l'IA d'opérer dans des environnements physiques complexes . Enfin, il a décrit les travaux du MIPT sur l'IA pour les sciences naturelles, soutenant la recherche en chimie, physique et biologie au sein des départements multidisciplinaires de l'université . En robotique, il a présenté des directions de recherche concurrentes - modèles d'action en langage visuel et modèles du monde - visant à améliorer la locomotion, la dextérité dans la manipulation et à réduire la latence .

---

#

Tim Unwin : une remise en question critique des prémisses de la session

La contribution du professeur Unwin a été la plus provocatrice et la plus structurellement déstabilisatrice de la session. Il a ouvert son propos en déclarant sans détour que « la plupart des universités dans le monde ne sont pas à la hauteur de leur mission » et sont devenues des institutions qui délivrent des diplômes sans valeur, une caractérisation qu'il a appliquée à son propre pays autant qu'aux autres . Fort d'une expérience dans 75 pays, il a soutenu que le focus de la session sur la formation universitaire à l'IA s'adressait à une population relativement restreinte et déjà privilégiée .

La priorité alternative de Tim Unwin était l'enseignement et la formation techniques et professionnels (EFTP), qu'il a décrit comme la forme d'apprentissage la plus pertinente pour la majorité des personnes qui occupent de véritables emplois physiques . Il a noté que l'EFTP reste un « parent pauvre » que la plupart des donateurs ne financent pas, malgré son importance . Il a également mis au défi le panel de reconnaître que, selon les derniers chiffres de l'UNESCO, un enfant sur dix dans le monde n'est pas scolarisé, s'interrogeant sur ce que les experts réunis faisaient face à ce problème fondamental . Il a averti que la croissance économique tirée par l'IA accentue les inégalités plutôt qu'elle ne les réduit, et a appelé à un changement fondamental d'orientation vers l'équité .

Au cours d'une intervention en milieu de session, après que le modérateur eut remarqué que les participants « essayaient de s'adapter à ce changement », Tim Unwin a répondu : « nous sommes des humains, nous pouvons encore être aux commandes, nous n'avons pas à nous adapter - reprenons possession de nos vies physiquement dans la nature ; la chose la plus importante que vous puissiez tous faire, c'est de passer trois jours sans aucun de vos appareils numériques, d'aller marcher dans les belles montagnes de Suisse. » Cette remarque illustrait son scepticisme plus large à l'égard de l'hypothèse selon laquelle les êtres humains doivent simplement s'accommoder de l'avancée de l'IA.

Plus tard dans la session, Tim Unwin a renforcé sa critique en déclarant que « la plupart des universités ne sont que des entreprises » et que de nombreux étudiants y assistent uniquement pour obtenir des diplômes, parfois en trichant . Il a remis en question la capacité réaliste des universités à inculquer la responsabilité aux étudiants compte tenu de cette réalité structurelle, et a appelé à une refonte fondamentale . C'est également au cours de cet échange que Tim Unwin a mentionné son livre récemment publié, entièrement consacré aux responsabilités, avant de développer son argument plus large sur les universités comme entreprises.

---

#

Yong Xiao : responsabilité, fractures d'équité et l'entreprise individuelle

Le professeur Yong Xiao a convenu que l'IA transformera fondamentalement la formation et la pratique en ingénierie, mais a articulé sa contribution autour de deux préoccupations qu'il estimait insuffisamment abordées dans les discussions dominantes . La première était la fragmentation de l'IA : plutôt que d'être une technologie unique et unifiée, l'IA est très fragmentée selon les pays et les contextes, différentes nations utilisant différents modèles soumis à des politiques différentes, chacun portant ses propres biais . Cette fragmentation, a-t-il soutenu, crée des risques et approfondit les divisions mondiales plutôt que de les réduire .

Yong Xiao a également souligné que l'IA évolue de jour en jour, contrairement aux cycles décennaux des générations précédentes de technologies sans fil telles que la 3G, la 4G et la 5G, ce qui perturbe les processus de normalisation internationaux . Sur la question de la responsabilité, Yong Xiao a soutenu que les ingénieurs doivent assumer une responsabilité bien plus grande qu'auparavant, précisément parce que l'IA amplifie les capacités individuelles et, par conséquent, amplifie également le potentiel de préjudice individuel .

La deuxième préoccupation de Yong Xiao était la fracture d'équité créée par l'IA. Il a établi une distinction entre la fracture numérique traditionnelle - définie par l'accès à la connectivité internet - et un nouveau fossé plus profond défini par l'accès aux modèles d'IA de pointe . Ces modèles sont coûteux en termes de tokens et d'infrastructure de calcul, ce qui signifie que les habitants des zones rurales et des pays plus pauvres n'ont pas les moyens de permettre aux jeunes de s'engager de manière significative dans l'ère de l'IA . Pendant ce temps, les pays riches consomment d'énormes quantités d'électricité et génèrent d'importantes émissions de carbone pour construire des modèles toujours plus puissants auxquels les pays plus pauvres ne peuvent pas accéder .

Il a décrit une réponse institutionnelle concrète à l'Université de Huazhong : un programme de licence « entreprise individuelle » (OPC) dans lequel des étudiants individuels reçoivent des financements, des allocations de tokens et un accès aux GPU pour réaliser des projets indépendants substantiels sur un seul été . Son argument était qu'une seule personne, équipée d'outils d'IA, peut désormais accomplir ce qui nécessitait auparavant une équipe entière . Cependant, il a précisé que le modèle de l'entreprise individuelle s'applique uniquement à la phase de production et d'ingénierie ; la vente, la compréhension des besoins sociétaux et l'obtention d'un impact dans le monde réel nécessitent toujours la communication et la collaboration humaines . Il a également soutenu que dans un environnement où les idées peuvent être instantanément reproduites par d'autres disposant d'outils d'IA similaires, une pensée véritablement disruptive est plus précieuse que jamais, et que les éducateurs devraient reconsidérer leur tendance à rejeter les propositions non conventionnelles des étudiants .

---

#

Perspectives étudiantes : projets de fin d'études avec l'UIT et pratique de l'utilisation critique de l'IA

Alexander Schrier a décrit comment un projet de fin d'études réalisé à l'Université de Pennsylvanie, en partenariat avec l'UIT, lui a offert une exposition rapide et approfondie à la gouvernance numérique mondiale qu'il n'aurait pas eu autrement . Son équipe avait pour mission d'élaborer une norme pour mesurer l'impact environnemental des centres de données d'IA . La recommandation produite par l'équipe a été soumise au Conseil de l'UIT et intégrée par la suite dans une nouvelle norme de l'UIT, l'UIT-TL1801, qui a également été financée par l'UIT . Alexander Schrier a également décrit la présentation de la recommandation de l'équipe au Forum de la jeunesse de l'ECOSOC 2025 . Il a présenté l'UIT comme un « organisme de diffusion des connaissances » - un lieu où les personnes transfèrent des connaissances et cherchent à améliorer le monde - et a soutenu que l'élargissement de tels partenariats avec les jeunes est essentiel pour des progrès significatifs . Il a identifié trois mécanismes spécifiques de l'UIT comme particulièrement précieux : les groupes d'étude qui réunissent des experts mondiaux pour élaborer des normes numériques, l'Organe consultatif académique qui rassemble des professeurs pour discuter de l'IA, de l'informatique quantique, de la prospective stratégique et de l'espace, et des conférences telles que le SMSI et AI for Good qui réunissent des universitaires, des décideurs politiques et des leaders industriels .

Joseph Devin et Ava Mitzi, participant en ligne depuis Sciences Po Paris, ont fourni un compte rendu complémentaire et empiriquement ancré de leur travail avec l'IA dans le cadre d'un projet de fin d'études d'un an également réalisé en partenariat avec l'UIT . Joseph Devin a décrit les limites pratiques des outils d'IA dans les contextes de recherche, notant que s'ils sont très efficaces pour les tâches de routine telles que la mise en forme, l'édition et la reformulation, ils sont peu fiables lorsqu'il s'agit de travaux conceptuels complexes . Un défi majeur était la gestion des hallucinations de l'IA : l'outil générait parfois des références bibliographiques inexistantes, ou attribuait des idées à des sources qui ne les mentionnaient jamais, nécessitant une vérification constante et un recoupement avec les sources primaires . L'équipe a également utilisé l'IA pour traiter de grandes quantités de données qualitatives issues de neuf entretiens d'experts, chacun durant plusieurs heures, ce qui s'est avéré utile pour organiser le matériel et identifier des thèmes, mais qui nécessitait néanmoins une validation systématique par rapport aux données brutes pour détecter les erreurs d'attribution et les connexions inventées .

Ava Mitzi a prolongé ce récit en se concentrant sur la compétence de niveau supérieur consistant à savoir quand ne pas déléguer à l'IA - ce qu'elle a appelé éviter la « capitulation cognitive » . Elle a soutenu que cela est particulièrement important lorsque le travail requiert un jugement original sur un terrain véritablement contesté, comme les questions de souveraineté des données, de risque géopolitique et des limites de la transparence des chaînes d'approvisionnement, où il n'existe pas de consensus établi . Une dépendance excessive à l'IA dans de tels contextes, a-t-elle soutenu, produirait « une analyse au ton très assuré mais superficielle » . Elle a également décrit le développement de ce qu'elle a appelé une « discipline des sources » - la pratique consistant à distinguer soigneusement ce que les sources primaires disaient réellement de ce que les résumés assistés par IA laissaient entendre qu'elles disaient . Le contexte de l'UIT a encore renforcé cette discipline, car les conclusions de l'équipe étaient destinées à alimenter de véritables travaux de normalisation, rendant concrètes les enjeux d'une représentation erronée . Joseph Devin a conclu que la compétence clé n'est pas simplement d'utiliser des outils d'IA, mais de les gérer de manière critique, en combinant efficacité, validation rigoureuse et jugement solide .

Les deux étudiants ont également réfuté directement la caractérisation par Tim Unwin des étudiants universitaires comme des chasseurs de diplômes. Ava Mitzi a déclaré que la curiosité intellectuelle, et non la certification, était sa motivation pour aller à l'université, et qu'elle n'avait personnellement rencontré aucun étudiant dont l'objectif était simplement de cocher la case d'un diplôme . Joseph Devin a immédiatement approuvé ce point , introduisant dans la discussion une tension générationnelle et expérientielle qui n'avait pas été anticipée dans le cadrage de la session.

---

#

Discussion avec le public : équité, collaboration, reconversion et formation juridique

L'échange avec le public a soulevé plusieurs thèmes importants qui ont prolongé et complexifié les contributions des panélistes. Un membre du public - un apprenant de français étudiant en Chine - a demandé des conseils pour les apprenants de langues naviguant avec les outils d'IA. La réponse du professeur Yong Xiao a été remarquablement expansive : il a émis l'hypothèse que les grands modèles de langage ont peut-être eu un impact aussi transformateur précisément parce que le langage lui-même est fondamental pour la civilisation et l'intelligence humaines, et il a invité les spécialistes des langues à aider la communauté de recherche à mieux comprendre ce phénomène. Il a suggéré que l'émergence de puissants modèles de langage pourrait refléter quelque chose de profond sur la relation entre la structure linguistique et la pensée humaine.

Sur la question de la reconversion des anciens étudiants déjà dans le monde du travail, la professeure Christine Kaddous a reconnu que ce n'est pas un défi simple . Elle a décrit une approche en deux étapes : d'abord, fournir une forme d'apprentissage et de formation à l'IA qui réduit la peur et développe les compétences de base ; ensuite, permettre aux professionnels d'évoluer au sein de leurs professions existantes ou, si nécessaire, de se reconvertir vers de nouvelles . Elle a souligné que le rythme des changements liés à l'IA signifie que ce qui peut être affirmé avec certitude aujourd'hui pourrait ne plus être valable dans un an, faisant de l'adaptabilité l'objectif central de la formation continue . Elle a également soulevé la préoccupation - pas encore pleinement abordée dans la session - que les meilleurs outils d'IA ne sont plus gratuits, créant un nouvel obstacle à l'accès équitable pour les apprenants tout au long de la vie .

Sur la question spécifique de la formation juridique, soulevée par un étudiant en santé mondiale et sciences politiques qui était également stagiaire à l'UIT, Christine Kaddous a décrit son approche pédagogique comme consistant délibérément à créer une distance entre les étudiants et leurs outils d'IA pendant les cours, en les obligeant à s'engager dans une réflexion indépendante et une interaction avec l'intervenant . Elle a soutenu que les étudiants doivent être maîtres de leurs études, utilisant l'IA pour être plus efficaces tout en développant leur propre raisonnement juridique de manière indépendante . Elle a également insisté sur l'importance de la transparence : les étudiants devraient être encouragés à déclarer leur utilisation de l'IA plutôt que de la dissimuler, en la traitant comme un outil parmi d'autres tout en restant conscients des obligations en matière de protection des données et de responsabilité professionnelle . Dans la pratique juridique professionnelle, elle a souligné que les clients ont des attentes en matière de discrétion et de confidentialité, et que la responsabilité professionnelle - garantir que les concurrents ne reçoivent pas le même service - est une valeur fondamentale que l'utilisation de l'IA ne doit pas compromettre .

Un membre du public venant d'Iran a soulevé la préoccupation que le modèle de l'entreprise individuelle pourrait nuire aux compétences de collaboration, notant que les étudiants montrent déjà une capacité réduite à collaborer parce qu'ils interagissent avec des agents d'IA plutôt qu'avec leurs pairs . Le professeur Yong Xiao a partiellement répondu à cette préoccupation en réitérant que l'entreprise individuelle s'applique uniquement à la phase de production, et que la culture professionnelle chinoise - caractérisée par une forte préférence pour la communication en face à face et le développement des relations - signifie que l'interaction humaine reste centrale pour le succès commercial . Cependant, il n'a pas directement abordé la question de savoir si le programme aggravait le déclin des compétences de collaboration, laissant cette tension sans résolution.

Un fondateur de startup de Mexico dont l'entreprise, Alexandria, se concentre sur la personnalisation de l'éducation à l'aide de l'IA, a également contribué à la discussion avec le public, soulevant des questions sur la manière dont la personnalisation pilotée par l'IA pourrait répondre à certaines des préoccupations d'équité soulevées par Tim Unwin et Yong Xiao.

---

#

Réflexions finales : apprentissage tout au long de la vie, effort cognitif et droit à la déconnexion

Dans ses remarques de clôture, le professeur Azamat Zhilokov a proposé une réflexion de synthèse sur la finalité de l'éducation à l'ère de l'IA. Il a soutenu que ce que les écoles et les universités font fondamentalement, c'est apprendre aux gens à apprendre, et que cette méta-compétence - construite à travers un véritable effort cognitif - est ce qui permet l'adaptation tout au long de la vie . Établissant une analogie avec le développement musculaire physique, il a soutenu que le cerveau ne peut pas se développer sans défi et effort, et que si les étudiants ne développent pas cette capacité pendant leurs études, leur aptitude à continuer à apprendre tout au long de leur vie s'en trouve compromise . Il a ensuite cité la formulation du chercheur en IA Andrej Karpathy : « on peut déléguer la réflexion, mais on ne peut pas déléguer la compréhension » . Sans une véritable compréhension, a-t-il soutenu, les utilisateurs d'outils d'IA perdront de vue et le contrôle des résultats . Il a conclu que l'importance d'une formation technique et scientifique fondamentale restera la même, voire augmentera, à l'ère de l'IA, et que le défi crucial est de savoir comment dispenser cette qualité d'éducation à un public mondial plus large .

La contribution finale du professeur Tim Unwin a renforcé le thème de la santé cognitive tout en y ajoutant une dimension plus radicale. Il a réitéré que le droit à la déconnexion est plus important que le droit à la connexion, car ceux qui ne peuvent pas se connecter conservent néanmoins des droits . Il a averti que le cerveau se détériore plus rapidement sous l'effet d'une utilisation intensive de l'IA que par l'exercice cognitif naturel, faisant référence à l'horreur de la démence - notant que ceux qui ont pris soin de proches décédés avec la démence comprennent viscéralement ce qui est en jeu lorsque le cerveau cesse de fonctionner - comme illustration concrète de ce qui est en péril . Tim Unwin a également noté qu'il avait lui-même animé des sessions au SMSI sous la forme d'une promenade à travers Genève, incarnant sa conviction que reprendre possession d'une expérience physique et déconnectée n'est pas simplement rhétorique. Le modérateur a répondu chaleureusement, suggérant que les participants à la session devraient suivre cet esprit et aller se promener dans les montagnes suisses sans leurs appareils .

---

#

Évaluation générale

La discussion a révélé un consensus fort et largement partagé sur les valeurs pédagogiques fondamentales - notamment la centralité de l'esprit critique, l'importance de traiter l'IA comme un assistant plutôt que comme un substitut, et la nécessité de cultiver la responsabilité chez les étudiants - malgré des différences significatives de contexte institutionnel, de formation disciplinaire et de perspective nationale . Le domaine de divergence le plus significatif opposait la critique structurelle de Tim Unwin, qui considère les universités comme des institutions fondamentalement défaillantes, aux positions plus réformistes des autres panélistes, bien que Tim Unwin lui-même ait convenu des valeurs substantielles de responsabilité et d'effort cognitif . Les panélistes étudiants ont fourni une validation empirique inattendue des positions théoriques des professeurs, parvenant aux mêmes conclusions sur les limites de l'IA à travers une expérience de recherche vécue plutôt que par la théorie pédagogique . Les préoccupations d'équité soulevées par Tim Unwin et Yong Xiao - concernant les enfants non scolarisés, le coût des outils d'IA premium et l'approfondissement du fossé entre nations riches et pauvres en matière d'accès à l'IA - ont ajouté une couche de complexité mondiale qui a remis en question le focus implicite de la session sur l'enseignement universitaire d'élite . Le modèle de partenariats entre universités et organisations internationales, illustré par les projets de fin d'études avec l'UIT, a émergé comme une approche concrète et reproductible du type d'éducation que tous les intervenants ont implicitement approuvé : rigoureuse, ancrée dans le monde réel et orientée vers une véritable compréhension plutôt que vers la simple certification .

Moderator - Regina Valiullina
Thank you. Thank you. Thank you. Thank you. Thank you. countries, the partnerships between developed and developing countries, partnerships with industry. And the whole point of our session today is also how we can best prepare students in this AI era. And we have profiles today from the universities which are focused on computer science, engineering, law, policy, economic development, AI. And we are very happy to have you. So joining us today as a panelist, so we have today Professor Christine Caduz, who is the director of the Master Program of European and International Governance of the Center for European Legal Studies at the University of Geneva, Switzerland. So let us applaud her for her presence. Thank you for your time in this busy schedule. We have also with us today Professor Arne. Azamat Jeloka. Director of the Institute of Artificial Intelligence Moscow Institutes of Physics and Technology from Russia. Thanks for making it today. We have also Professor Yong Xiao, I hope I pronounced correctly, from Huazhong University of Science and Technology from China. Let us applaud. Thank you for joining. We have also Professor Tim Inwin, Royal Holloway University of London, United Kingdom. Let us applaud. Thank you for joining. And then we also have students with us, we have Mr. Alexander Schreier, who did his Bachelor in the University of Pennsylvania and now is pursuing his Master in Johns Hopkins, the US. So, let us speak. And we have also our students online from Sciences Po Paris, Mr. Joseph Devrin and Ava Mitzi. Thank you. And our session today will be structured in two parts. So the first part will be the roundtable discussion, the questions which will be addressed to every panelist. And then the second part will be the exchange with the audience. So we will ask you to prepare your questions. And the first question will go to Professor Christine Caduz. So we imagine that you're witnessing how AI is rapidly shaping the job market nowadays. From your perspective, what are the biggest trends of the University of Geneva current approach to preparing students for the workforce? And where do you see opportunities for the University to evolve its programs in response to the future job market?
Christine Kaddous
Thank you. Thank you. Good morning first to everyone. Thank you very much for the invitation and the participation to this panel. I think it's a very important week that we are having here in Geneva and of course the University of Geneva being in this city is becoming really at the heart of all the discussion about artificial intelligence governance. So after Paris, Delhi and Geneva coming next year in June, we will have the third summit on artificial intelligence and of course the university is part of all these events. So part of the discussion was international organization, part of the discussion was private sectors, NGOs and other academia. And this is really one of the things that we are doing. This is one of the most important, I would say, elements for the Geneva University in terms of teaching and research. We are, of course, at the University of Geneva doing lots of research in the field, not only in more technological things, but also in all the faculties, the law faculty to which I belong. Of course, it's at the heart of the matter with all the impact on intellectual property and all the things, responsibility, and really all what is doing with policies that we have to develop to govern these artificial intelligence. So we are having at the University of Geneva a real proactive role in terms of AI. And we are welcoming. We are welcoming the responsible use of AI, making it as a tool for all the, I would say, academic community. But it is also a way of reducing repetitive and administrative tasks at the university. So we are really working on that issue. What we are also doing more specifically at the law faculty is that we are really not only looking at AI as a tool for the lawyers and the future lawyers and the job market afterwards, but also a way of trying to give all the elements to see all the risks and opportunities in this AI. And it's true that we have to deal with, I would say, two main issues, and these are also opportunities. First, we have to educate our current students, but we also have to look at the former students of the university that are already on the job market and that they need executive education, long life education. And this is really a way we are really promoting. We are promoting executive masters. For example, in the MEIG program that was mentioned before, Master in European and International Governance, we do have a whole module on digital governance. And this is really a way of bringing that not only for the current student, but also for continuing education. But what is also very important, if you may allow me, is that we try really to focus on critical thinking. Because Bachelor and Master students nowadays, they are using artificial intelligence all the time. They are even using it, as you all know, in the classroom. So the speaker or the professor is saying something and they are already interconnected with artificial intelligence. So the idea is really to try to push them not only to use it in a permanent way, in a constant way, but to try also to educate and give some reflex that they can really think. in an ethical way and do things on their own, alone without artificial intelligence, because this is very important for them to make AI as an assistant to improve their work, because we know, for example, in law, artificial intelligence can, in a couple of seconds or minutes, summarize, I don't know, hundreds of cases of jurisprudence. They can summarize bibliography. But what is really needed is a critical look at what the AI can give. So AI, if I may just summarize, AI is an assistant, but it is not a substitute for analysis, legal reasoning or intellectual rigor. This would be my first main message. Thank you.
Moderator - Regina Valiullina
Thank you, Professor. Yes, indeed, AI is not a substitute. Thank you. Thank you so much for your perspective. I think we can further discuss. And the next question will go to Professor Azamat Zhilokov. Azamat, as the director of the Institute of Artificial Intelligence at MIPT, you have insight in both current breakthrough, but also emerging trends in contemporary research. So what frontier technologies or research directions you are generating mostly, and which do you believe will have the greatest impact in the future?
Azamat Zhilokov
Thank you. Thank you for inviting us to participate in this session. I like the name. The education and research is a good combination because one feeds the other. And the research cannot work without the education. So to answer your question, I'll give a quick framework of where we are at MIPT. We have the so -called system of FISTECH of three pillars, and this is how we bring the best talent, educate them, and then make them ready for the economy and the lead in research tasks. So the first pillar is to find the best kids around the country, select them, and filter them because it's only 1 ,000 students that we accept into bachelor every year. The second pillar is they go through the fundamental cycle, as we call it, for computer science, math, and physics. It's very hard, three years, really, really hard, and they study really well. And with or without AI tools, as was mentioned by Professor Cadu, it's important to develop critical thinking. This is when they develop their critical thinking. There's some things you just cannot do. With AI tools, they study hard for three years, and they become ready. For the next step, the third pillar, is when they get involved into real research hand -in -hand with the leading professors. And we try to, for all the talented students, we try to attach them to the leading professors in their area so they can start working immediately with the best in class, with the best possible environment in their third year. And this is what allows them to get to a certain level very quickly. And by the time they complete their bachelor and sometimes master's degrees, we have a target. They have two or three top journal publications, and this is the minimum target that we set for them. So now speaking of the specific research areas that our university is involved in and what excites the young researchers, there's two main sections. One is the modals and agents. agentic AI, and second is robotics. And in the first one, one question specifically very hot right now is economy and efficiency and finding new architectures and architectures not only of the models, but specific layers of the models to introduce optimization in the inference cost and reducing cost of a certain token, a single token. As you know, the CAPEX and OPEX, the investments into infrastructure is enormous right now, and the burden on the energy system is enormous. So energy efficiency and cost efficiency will remain the hot topic for the next few years, and we have a lot of studies and research targeted to that. We actually have a big team at the ICML conference, International Conference on Machine Learning, this week in South Korea delivering the workshops and their papers on the subject. So second is interpretability, explainability, and trust. Also a hot topic I heard yesterday, today, many sessions during the conference, introducing guardrails for the large language models, making the AI results explainable and trustworthy is a big thing. Third, I will mention, is the neural models of real world. Neural world is making the models understand and imagine the real physical laws and roles and dependencies so they can operate in the complex processes and environments, the physically inspired neural networks. That's a big direction in which we're headed. And finally, on the models and agents, I will mention the AI for natural sciences. We, as a multidisciplinary university, we work with different departments with chemistry, physics, and biology. To support their objectives and develop tools for their tasks. That's a big thing. And finally, I will mention that. But last but not least, in robotics, we do research on visual language action models and the world models. These are two competing directions, one of which will probably be the future of robotics control systems and enhanced locomotion and manipulation, dexterous manipulation, and reduced latency. These are the targets that we set in our research for robotics visual language action models and the world models. So, yeah, these are the key things that we are researching in our university. And not only we think it's important, but also it excites the researchers. They think it's important what they do, and it's impactful
Moderator - Regina Valiullina
We will have the next question that would go to Tim, Professor Tim Moonveen. So, Tim, given your background in the digital inclusion, what are the most important things that you think you can do to improve the future of robotics? And what are some of the things that you think you can do to improve the future of robotics? how should universities in the developing world prepare students for the age of AI and how can AI be used to expand educational opportunities and drive social economic development?
Tim Unwin
I already see your eyes. Okay. I've got two questions. They're huge questions. Firstly, thank you for the honour of being here. I'm usually rolled out to sort of try and be a bit more critical and get some debate going, so let's go for it. Right, well, the first question. Most universities in the world are not fit for purpose. I think we can all agree on that. Most universities have just become institutions for handing out worthless certificates. And I'd say that about my own country in Britain, let alone the countries I've worked in. I've worked in 75 countries across the world. Yeah, I'm the oldest here, I think, so I've been around the block. But it obviously depends how far in the future we see tomorrow. And I think we need not to be thinking of it literally as next year or even the 20th. We're not even thinking about what's going to replace the SDGs. Now, that's the most important thing we need to be doing. They're just around the corner, 2030. But if we think 10 years down the road, it gets really easy. So I think we need very, very much better technical education rather than university education. Most people don't go to university. Most people don't have the skills to work in institutions like yours, Asmat, and yours, which are the top echelon. That's not where I work. I work with the dirty stuff underground at the bottom. But actually, technical education is going to be the learning. It's the learning that people actually who have real physical jobs do. and we need skilled people to do that plumbers in my country and much more than academics and that's probably right because most academics sit in their ivory towers i never thought that was true but it really is with no connection with the real world at all just doing great wonderful science so good technical education for the roles that human will still do i i was once in an african country talking minister of employment and she said you know t vet is the most important thing but it's a cinderella subject um most donors are not funding t vet because it's unproven um so so we need that technical education and and universities well you know universities should be places where we have brilliant people as you were describing have great thoughts that are really really eager students who don't get somebody else to write their phd or don't get ai to write their phd but who think who are creative and they work with good professors like like you who want to create a good university and they're not going to get a good university that's a tiny tiny majority of people in the world and it should happen but it's not the dominant one I agree about critical thinking but most people define it in different ways that is the skill that people need to have the second part of the question how can AI be used to expand education opportunities and drive economic and social development solve the world's problems do you know how many people, how many children in the world are not in school one in thirty not in school don't know if I bet it's one in ten one in ten children in the world according to UNESCO's latest aren't in school, what the H -E -L -L are we sitting in a room like this brilliant brains doing about that that's the education we need to get right that is the education that we need we need our young people because in Africa when they're talking about the youth dividend if you have that number of uneducated people without any jobs it's not a dividend it's a millstone around the neck of africa i don't know if my african brother or sisters here there aren't many but we have to be true and speak about those things um and of course digital tech and ai i mean i just see ai as an extension of what's been happening over the last 50 years it's not really something profoundly new um of course that's going to drive economic growth of course that's it but economic growth drives inequality and we're doing very very little work on how ai um and other technologies can can do something about these one in ten kids out of school across the world and and we have to shift away from a a global world view that economic growth is going to solve the world problems to focusing on equity on equity
Moderator - Regina Valiullina
okay thank you very inspiring yeah we that's true that's why we are here i think that we know that education also starts from early childhood also from the school but then we go to the university and we see with the eye and other emerging tech that uh it's changing so fast even in one two years we see that the old jobs are evolving and also the careers are evolving that's why i think uh what the eye is doing usually it's learning all the models they're learning and they're adapting and i think it's what also from our side what we are trying to do we're trying to adapt to this change which is happening which
Tim Unwin
okay we're humans we can still be in charge we don't have to adapt let's reclaim our lives physically in nature the most important thing you can all do is take three days without any of your digital devices go walking in the beautiful mountains of switzerland it'll really charge your batteries you'll come back a different person
Moderator - Regina Valiullina
thank you thank you professor so we will have the next questions which will go to professor xiao uh with the increasingly automating technical stats we see that many industry experts also predict that engineering careers or computer science careers will be fundamentally transformed do you agree and how should universities rethink engineering and computer science education to prepare students for the new reality thank you
Yong Xiao
yeah all right uh first of all thank you uh very much for having the opportunity to present our idea because uh to be honest uh i think this uh definitely i totally agree there is a definitely the ai will transform the entire engineering part especially in the education and the thing is that uh right now people always talking about ai right so one thing i thought i saw that it is quite of course people think ai as one part or one part of one specific technology, but it is not. Actually, the AI part is highly fragmented nowadays. And every country, every people, they have different way of using the AI. And even the China and US, they're using different AI models, right? And also the different models. Actually, the people trust AI models. However, every single AI model, they have the bias. So in other words, it's actually create more risks and also more fragmentations throughout the entire world. So, and also not alone, especially talking about for us, right? Our people, especially me, I'm working on the 6G and wireless technology era. And also I'm attending the ITU standard during the past few days, a few years. And I do believe that the international standardization has already been transformed, right? Because in the past, people always, right? Especially for the wireless technology, we have 10 years period, right? People talk about, well, 3G, 4G, 5G. Normally we think 10 years, the technology revolution, right? But AI, changing on a daily basis. And also different countries, they have different policies to constrain the AI capabilities because right now, even the China and the US government, they think about a few years ago, I remember there was a big claim about the government to restrain the AI in the military or some very specific area. However, nowadays you can see, even if a few years ago, people were talking about, no, we should not use fully autonomous in the military. But right now in China and US, they're actually doing a lot of things on automizing the battlefield. So it's actually, in other words, what I really think that for the technological person, especially for the engineering people, we must take much more responsibility than ever because the AI has definitely much higher capability. However, it also creates much higher risks. And also, this is just one part. Another thing is that the AI also creates a lot of larger gaps. equity gap, right? In the past, people always talked digital gap, right? Like ITUT, we talk about digital gap means that you can talk to people, you can internet connections. But right now, people should be able to assess the state of the art models. However, the state of the model are expensive, the token expensive and the computational facility expensive. So a lot of people, a lot of countryside or rural area or poor country, they couldn't afford to let the young people to really understand what happens in the AI era. So eventually we'll call it even higher gap. So in the rich country put a lot of efforts, electricity and create much waste and also create a lot of carbon emissions to building much stronger AI models. But the poor country couldn't even assess it. So it's actually created a lot of issues. So that's the thing I do believe as engineering people in the education part, of course, I do agree with with Tim, a lot of previous intervenants, they talk about, okay, we students need to talk about creative thinking. However, I do believe the university they have even higher big role to make students to be responsible, to let them understand what is responsible. Because you have the AI tool. You have even single person they have higher capability to do a lot of even good thing or bad thing. The students must take the role of teaching people to use the technology responsibly. And also, for example, in our university, the Huazhong University of Technology, we actually recently have an undergraduate program we call one person company. I assume a lot of you guys think about this idea, right? One OPC, right? One person company, just one people. They have a lot of AI tools. They can do a lot of big things. And actually, the undergraduate students, we have a program that each undergraduate student can apply for the funding. So they can give the token, a certain amount of token expense, and also the GPU card for you. And you can do a project as long as you have idea on just one single summer. In other words, one student, even one people, they can do a lot of huge things right now, not in the past. In the past, they have to consult and build up a team and do a lot of things. But right now, any single person, they must, because they have a higher capability, they must take more responsibility. That's my thing, that we as an engineering person,
Tim Unwin
Can I just respond, because I completely agree about responsibilities. My new book just published is all about responsibilities, which I didn't mention. But I'm going to be really naughty now. Most universities are just businesses. Most students who go to universities just want certificates and they cheat to get it. You may laugh, it is true, even in good universities. So how on earth can you expect universities to be responsible? That's actually a very serious point, because I think we need a fundamental overhaul along the lines you're talking about, because most universities are not responsible.
Yong Xiao
so we are very looking forward for your book so we are looking forward to the
Tim Unwin
it's here I can give you a couple
Yong Xiao
all right yeah I'll sell you a couple you see we have to take the responsibility so you should give us for free I wish oh okay okay that's really good thank you yeah that's my point thank you
Moderator - Regina Valiullina
thank you very much it was yes very inspiring I think for all of you and what we've learned yes indeed responsibility and of course we have public universities private universities in different countries it's structured differently but I mean the whole purpose of universities is to educate right that's why they were created in the first place thank you so much I think responsibility and critical thinking is what will stay with us so now we go to Alexander to provide his experience with the capstone project Alexander so prior to working actually I I'm Alexander is working he's doing internship now with us and I'm And he also completed his capstone project with the University of Pennsylvania. And we think it could be a good idea to see more this type of partnerships and how it evolves, what actually he's learned from that.
Alexander Schrier
comes in. Because when I studied at Penn, as a suburban kid from America, I didn't know much about the ITU, which is, you know, the UN's digital standardization organization. And I think through the capstone project, which I was tasked with, and my team was tasked with coming up with a standard to measure the environmental impact of AI identity centers, we learned a lot very fast. And I think in order to make an impact in this field, you must understand the UN's, one of the most important digital agencies. And I think we learned a couple of things. We learned about the ITU's mandate, where it is in digital standardization. We learned about the ITU's commitments to its member states, not just committing to one country, but to all countries. And we also learned about the flexibility that the ITU offers to young people. We were able to speak at the 2025 ECOSOC Youth Forum on our recommendation. And our recommendation was actually submitted to ITU Council, and then integrated into a new one called ITU -TL1801, which is a long name, but was integrated and funded by the ITU. And we learned a lot about the flexibility that the ITU offers to young people. And so we're has that actually led me? How has ITU led me? So now I'm a master's student at Johns Hopkins School of Advanced International Studies, focusing on the intersection of law and digital policy. And I think that wouldn't have happened without, I think, my experience learning from ITU, taking classes on digital transformation, on machine learning, et cetera. And I think this third point of where is ITU at the forefront of research education, because we are here, I think there's a couple of things to learn from there that have been pioneered by our Secretary General. One is the study groups, which bring together the top experts around the world to come up with standards on digital technologies. Two is some of the new initiatives that have been created by ITU, such as Academic Advisory Body, which is a group of professors that discuss AI, quantum, strategic foresight, and space. These are exciting, new in the past. And then three is conferences such as this, such as WISIS, such as AI for Good, such as AI Global Dialogue, which bring together people. like this and professors and academics, CEOs, whatever, to discuss AI and the frontier of research and education. So at its core, for me, ITU is a knowledge dissemination body. It's a place where people get together and transfer knowledge with each other and try to make the world a better place. And I think for the ITU, the more it transfers that knowledge to young people and students like myself through the Capstone Project, through other sorts of partnerships, then we're getting someplace better. And so thank you very much for that question and I appreciate
Moderator - Regina Valiullina
Thank you, Alex. Maybe you can clarify regarding your recommendations because you included the name of standards. I don't know if everybody was able to get to the topic that you've been working on.
Alexander Schrier
Yeah, our topic was on creating a standard for the environmental impact of AI data centers. And essentially what we did was this had already been studied previously. And we were able to get to the topic that we've been working on. And we were able to get to the topic that we've been working on. And we were able to get to the topic that we've been get to the topic that we've been working on. And we were able to get to
Moderator - Regina Valiullina
Thank you, Alex. And now we go to Sciences Po Paris. So Ava, Joseph, are you ready?
Joseph Devin
Yes, we are. Thank you very much. Thank you, everybody, here for your nice insights. I think, Ava, you wanted to say something first. Can you speak louder, please?
Ava Mitzi
Yeah. Can you hear us okay now? Can you hear us? Can you hear me? I can hear you, Ava. You can't hear us? Can they hear us? Go ahead. Yes. Go ahead.
Moderator - Regina Valiullina
Okay.
Ava Mitzi
Thank you so much for having us. We're really excited to be part of this discussion regarding education and research in the AI era. We have a statement that we'd like to say regarding our capstone project, regarding the AI that was used therein. But also I'd like to respond. Thank you. Thank you. quickly to say to the professor, as a student and representing Sciences Po, I didn't meet, or I am not a student that went to a university to get a certificate. That certainly wasn't part of my agenda. It was more about intellectual curiosity. And I personally didn't meet a single student that just wanted to tick the box for their certificate. But that's for a longer conversation that we can take on and, you know, continue to have after this conversation. But thank you so much. Now we'll go to our statement. Go for it, Joseph.
Joseph Devin
Yeah, I must say that I totally back Eva on this. I didn't win for just a certification or a ticket. But anyway, regarding the use of AI throughout our studies, and specifically regarding the CAPSUN project that we did in partnership with ITU, as a master's student, one of the most valuable skills that we have developed is maintaining a strong critical oversight while using AI tools. And this has been repeated over the conference, but they are highly effective at speeding up routine tasks such as formatting, sparing, rephrasing, and editing. But they are not always reliable when it comes to handling concepts and complex ideas, which creates new challenges, rather than eliminating them. A key issue we faced throughout our research was dealing with hallucinations. And for example, when managing our bibliography, the tool would sometimes generate references that did not exist because it was trying to satisfy our search for a relevant source or attribute ideas to sources that never mentioned them. And over the course of a year -long project with an extensive literature review, this required constant verification and cross -checking. We also used AI to process large amounts of qualitative data, particularly the transcripts for our nine interviews, each of which lasted several hours, which was very helpful in organizing the material, identifying themes, and tracking who said what. But even then, it could occasionally misattribute statements or invent connections that were not actually present, which meant we had to systematically return to the raw data to validate everything. Over to you.
Ava Mitzi
Thank you. Another skill that became really essential was knowing when not to delegate to AI and not to do cognitive surrender. So this is particularly clear when the work requires an original judgment just versus a synthesis. Our research involved genuinely contested new terrain, which is questions about data sovereignty and geopolitical risk and the limits of supply chain transparency, where there's no single established consensus yet. In these moments, over -reliance on AI would have produced a very confident -sounding but shallow analysis, and the capstone taught us to reserve AI for what it does well, which is organizing, structuring, and accelerating, and to protect the interpretive and the argumentative work that needs to require also a human involvement. We also developed what I would call a source discipline. AI makes it very easy to accumulate information quickly, but that speed can create a false sense of coverage. And therefore, the nine expert interviews and a full literature review spanning mining certification, regulatory policy, and international governance framework. were essential to deliberate about distinguishing between what is primary sources, what they actually said, and what AI -assisted summaries implied they said. The ITU context sharpened this further because our findings were intended to inform real standard -setting work, which meant that the stakes of misrepresentation were concrete and pushed us to treat verification as a continuous part of the process rather than a final check.
Joseph Devin
So the CAPTCHA project with ITU reinforced this skill set. Working at that scale forced us to be disciplined in our methodology and to put clear verification processes in place and to work as a self -checking pair while remaining fully accountable for the accuracy of our work. In the end, the key skill is not just using these tools but managing them critically by combining efficiency with rigorous validation and sound judgment. At the same time, AI remains a tremendous time saver. Thank you very much.
Ava Mitzi
Thank you for having us.
Moderator - Regina Valiullina
Thank you Joseph, thank you and I think now we go to the very interesting part the second part, it's questions and exchange with the audience so we have already we have already people raising their hand so we would like to ask everyone to keep questions short and focused rather than make a statement so we can maximize our time for the discussion and we will take a few questions at once and then we will distribute to the panelists to respond, okay, thank you please introduce yourself before asking
Audience Member 1
I'm a French learner in China and my question is do you have any advice for language learners in China as students
Audience Member 2
okay language learners language learners Professor Kadu, thanks a lot for your comment on the continued education, right, executive education I'm actually very interested in this aspect of lifelong learning I'm from the startup stage in quantum computing so integrating yet another frontier of new technologies in there but I'm very curious about how do we upskill, reskill the generations of students that are already out of
Audience Member 3
Hi Professor Kadu, this is for you as well I'm a student at UC San Diego studying global health and political science I'm also an intern at the ITU and it's kind of going off of your previous statement but I've witnessed the exponential use of AI in the classroom and many would say an over -reliance on AI and for students interested in pursuing law school how do you think AI will alter the dynamics within the studies specifically and In your assessment, how will this field change, and how should students be better prepared for these changes within law?
Audience Member 4
Okay. Yes, please. Hello. Can you hear me? Yes. I'm Eugenio. I'm a startup owner from Mexico City. I have a startup called Alexandria that is, the focus is to personalize education using artificial intelligence. And in using AI governance practices. My question is, what are the skills, for example, in engineering that now we have by coding tools, and we have other agents automating all the very different skills that now students need to have? And also for the one company person concept that Professor Yang mentioned, what is the... the concern that we need to have in the future and maybe what are the policies or best practices that we need to do as a startup owners that want to have a platform
Moderator - Regina Valiullina
I think we can just stop for a few seconds, respond to questions, and then we go with the next batch. You wanted to ask? Okay, please go ahead.
Audience Member 5
Thank you very much. I have a small question for the fellow professor from China. So in countries like Iran, they are pushing for people to be entrepreneurs and universities are also pushing towards entrepreneurship. Are you in favor of what you said, the one person company? Because students learn to interact and to share together when they learn to mutualize. But when it's a one person company, then you don't mutualize and you don't collaborate. My experience in classrooms are that students have much harder time collaborating today because they're in exchange with AI agents. So I would like to know what's China's experience. Do you feel that collaboration and sharing skills are getting higher in China due to what you're doing or what would be your recommendation for Iran? Thank you.
Moderator - Regina Valiullina
Okay, I think we can start responding. Maybe we will start from Professor Xiao because there were a few questions regarding the language learners from China, then the questions from Mexico regarding skills, and also one person, one company. If you can address it.
Yong Xiao
Thank you. First of all, I'm not really understanding the language study because this is not my major. I'm really in engineering and doing the wireless communication stuff. But my personal thing, the . And because I don't have an understanding about the language study, so I could be, my answer could not be a thoughtfully thing, right? Because it could be, because my understanding that the language is mainly focused on understanding why the human language works. The thing is that, to be honest, we are always thinking, because right now, all the AI models, they are actually called the large language model. Of course, there are some other type of models, but the real innovation is happening. Really, most of the time is large language models. So actually, I have heard a lot of people saying that, oh, this model is just purely based on human languages. So that is really, there is an illusion. There is a lot of issues simply because human languages, even us, sometimes what we talk may not be what we think, right? Because that's how language works. However, I'm actually always thinking from another angle, right? The main reason that the large language models can really take a huge impact on the human society in the moment, it is because, could be, because of the language model. Language is really shaping the human civilization. right because the main issue because that's that's my own opinion because i don't have the i don't really read a lot of literature but the thing is that i could be because uh for example the dinosaur could be a live longer but human they have emerged some really intelligence could be because of the language right and also some other people talking about collaborations language is a way for people to communicate however language have issues language is slow and people people communicate using languages they could be causing bias causing illusions a lot of issues however is if we can really because right now uh most of the model is based on language however in the engineering part we're also talking about like visual large models or wireless communication we call it wireless models these data set are fundamentally different from the languages However, we are keen to understand what are the fundamental similarities between, for example, the large language models of fruits or the results, the research results could be extended to some other area like the AI for science, right? AI for science, if you build on the large language models, then you must have some similarity from the languages, right? Because the people, the large language model use the language to communicate, to reason, to infer some of the meaning from what you say, right? However, that's the thing. I think the language study could play a huge role in the future because you guys understand how the language evolved because even the new term comes up every year, every single year, even now the influencers, the genders, they have some languages that I have no idea because I took my daughters here. My daughter, sometimes they say some of the terms, I have no idea what they're talking about. But these are terms they created by the young people. They are easy to communicate by themselves. So I think could be some reason that why language models could be really shining right now, because AI is not a new term, right? But AI becomes so revolutionary simply because of the brook suit in the large language model. So I do hope that we can have some clues from the language studies, right? That's my first question, right? The second one is about the one -person company. I think that both of them are about the one -person company, right? Because one thing I have to clarify is that one -person company doesn't mean the whole business is one person, right? It's simply because the one person, through communication, through understanding the need or requirement of the society, of the people, they can build up something by themselves. That's it done. Only in the engineering part. It's not in the society part because you have to sell the product made by one -person company. So you have to understand the need. You have to communicate with a lot of people. Especially for the Chinese, right? Chinese people like to go to restaurants to talk business. Simply because we like to communicate, right? We like some good food and then people talk. So that's probably a difference. But again, right, because I think there is a huge company and a huge risk for the one -person company, right, because especially for the startup, right, because in the past, if you really want to create a big company, you have to go to a very good university. You have a lot of – go to a university that is not for the certificate. Of course, Professor Tim talked about the certificate, but build up your human resource, right? You understand a lot of people because a good university means your classmates could eventually become president or become a lot of resource you can use. You build up a connection from the university, also the alumni from that university give you a lot of resource, so make you easier to success. That is the past. But right now, simply, you have a one -person company, so you have to think about something that normal people couldn't think of, right, because if you have the idea that is similar to others, then you can do this. The other people can do the similar things. So it means that. If you have the idea, well, we'll just – It's not like in the past. In the past, probably if you bring some new idea, especially for 40 years or 50 years in China, we have some people go to US. They learn some of the new technology, like eBay, a lot of sense, PayPal. They learn this new model and go back to China, simply because in China, they don't have a well -connected, there's a digital gap between China and US. So they build out the model, so they thrive in China. But right now, this is a wholly connected world. So people can understand everything, what happens in other parts of the world. So if you really have the idea that's similar to others, that will not shine. However, it also means that if the people have a good idea, even if you are not the major, for example, I'm doing the wireless, for example, I can write a paper about international studies. But I have no idea about this, but I can ask larger language models and I can do some research quickly than ever. I can get a lot of information, a lot of news. So this means that there's a big risk for the studies. It's a startup because it means that if you don't have a really disruptive new ideas. That's the reason I'm actually talking about some other professor in China we talk about. In the past, if the students come over, say some really ridiculous question, you can say no. If you continue to do this, then let's stop talking. I won't sponsor you. But right now, we have to rethink. Because the students, if they have some really disruptive or even sounds like a not good idea, they could be because they have a lot of tools, advanced solutions. Probably something not available in the past. Could be tomorrow or because of AI evolutions. So if you have a really disruptive idea, even a lot of people surrounding you say no, it's not right. Don't do this. Then probably that's the opportunity for the startup. That's my own opinion. Could be wrong, but yeah.
Moderator - Regina Valiullina
Thank you very much. And we go to other questions. So I think there were a few questions for Professor Saboos regarding life learning and also regarding loss of knowledge.
Christine Kaddous
were a few questions for Professor Saboos regarding life learning and also regarding loss of knowledge. Thank you. Thank you so much for the questions. Maybe about rescue former generations. So this is not an easy question, of course, because it's important first maybe to give to these generation kind of AI apprentissage. So they have really to know what it is about and not to only think that this AI is a risk. We should try to avoid all the fears that this AI brings to them. And that they are capable afterward to use it as a tool. So there is a first step of apprentissage. And then I think another step where we have to give to the generation the way to evolve with the new tools. To evolve and to allow them either to evolve in their own professions or if it happens, they may also change professions. So, of course, I think the evolution is very important. So what we are capable of saying today. is not sure that it will still be valid in one year or even less than one year. So this is very important to get this evolution in the continuing education. And this is really what we try to do at the University of Geneva in all what we put into place in digital, in a very broad sense. So there are different levels, but really with the same objective, evolution. And maybe one element we didn't really discuss yet, but I think it's very important to what has been said to who is capable of using AI, who is not, etc. We see today that the best AI tools are not given for free anymore, and that we have to pay for this. So this is also an element we haven't discussed yet, but I think we can put it. But I'm not... I'm not going to start the topic now, but I think this is really a very important... key element in the discussion. So the other question about more specifically the law studies, and I think this is very important because we do have, I would say, different type as a professor. Of course, there are the lectures. So the lectures taking place at the university building first. There are, after that, what the students have as assignments. They can do assignments at home or wherever they want to be. And there is the third step, the professional activity. So what I am really trying to put into place in my lectures is to oblige in a way that the student is not always taking into consideration what AI says in a constant and immediate way. It should not be an immediate tool. So what I'm trying to do is just to take them a bit away of these AI tools and to make them think and try to make them think. And try to put the right question just to have an interaction with the speaker. And I think this is the most important thing. So they have to free their mind. of any, if I may say it like this, I see all computers here, of any computer or technological instruments in order to develop these critical thinking. They have to be their own master in their studies and use, of course, artificial intelligence. It's a tool. It's to be more efficient, better. It's a wonderful tool. I'm using it as well. But you have to really build your own way of thinking in the legal reasoning. And they have to be transparent. Students are still afraid of saying that they have used artificial intelligence. So we really encourage them to say transparency is the most important thing. So it can be used, of course. It's a tool like any other tool. But it has to be mentioned because there are some data protection. There are many, many rights that have to be protected. And maybe at last point not to be too long. to give the opportunity to everyone to participate in the discussion. The professional activity, this is very important, because here you have, if you are a lawyer, you have a client. So you have people, you have companies, even if it is a one -person company, but you still have a client. And what is very, very important for the client is that there are expectations. And when there are expectations, there must also be a kind of discretion. You don't want that your competitor has exactly the same service that you as a lawyer or whatever, as a technological engineer or whatever, gave already to another client. So there is competition, data protection, discretion, and professional responsibility. Because at the end, this person sitting opposite to you or on Zoom or in Sciences Po Paris or whatever, they have to be responsible. A professional responsibility. Responsibility is really key. And this is one important, I would say, value that we have to, well, we didn't discuss much of values. There are many values that have to be safeguarded in the use of AI, of course.
Moderator - Regina Valiullina
thank you very much and we are running out of time so we have the last comments concluding remarks i think from uh azamat and then team so azamat if you can also give your view on life learning skills and the same for team
Azamat Zhilokov
yeah i just have a very brief comments on these two things one is uh i'll say this maybe a little bit disruptive but what people really do in schools and universities is they learn how to learn and then they apply that skill throughout life and like building physical muscles same thing with the brain you cannot build it without struggle right you have to feel the pain you have to have the challenge so what's important is whether you use ai tools or you don't use the ai tools the education in the university and the school as well has to be associated with a certain challenge right it needs to you need to you need to be able to do it right and so i think that's a very important point put the power and time and effort it should challenge your brain so So that will allow you to build a skill for lifelong learning, right? And that's very important. If you don't learn during school or university, then you don't build that skill, and then your life is questionable later on. That's one thing. Second is one of the best things I heard most recently on AI from Andrej Karpathy, who is considered to be one of the best researchers in the world in this area. He said this, you can outsource thinking, but you cannot outsource understanding. So if you don't build the skills, you don't understand how things work. You can use the tools all day long, but without understanding, you will lose the sight and control of things, and that will get you in trouble. So the importance of the technical and natural education will succeed. It will remain the same, if not more, the importance of that education. the question is how do we deliver that type of education how we transform education for a broader audience as it was said earlier today by Tim how do we deliver that same level of education and make people excited about being educated that's a whole different roundtable discussion I think thank you
Moderator - Regina Valiullina
thank you very much
Tim Unwin
I'll be incredibly brief to the sort of previous question to Christine and we cannot doubt we don't have to use AI you do not need you know for most of my 70 plus years I haven't used AI so it's terrible isn't it no I actually know recently I have but no I mean for most of it as you do not have to use AI and remember too that the right to be unconnected is more important than the right to be connected just take that in the right to be unconnected is more important than the right to be connected because if you are unable to be connected you still have rights and and i i love the point you just made using your brain struggles it really does and most people don't want to struggle so they use ai and that leads to digital dementia i don't know how many of you have cared for relatives who've died with dementia it is horrible when your brain stops working and your brain will stop working far faster if you use a lot of ai than it will if you use your brain for what it was intended
Moderator - Regina Valiullina
wow thank you very much so we've learned a lot today we have disruption struggles but i think we will go through that and we will all go for a hike disconnected in mountains in switzerland i've run sessions at wissis that have been a walk around geneva we should do more of that yeah the recording has stopped you give a big applause to our panelists and the audience

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.