AI opportunities and implications: social, economic, cultural, linguistic and technical dimensions

The speaker argues that AI is advancing too quickly for any single government, company or institution to manage alone, so multilateral cooperation is indispensable. The goal is not only to agree on principles, but to bridge divides in access and capability and convert AI into practical, inclusive industrial development outcomes.

He states that AI is transforming economies and societies at unprecedented speed, with new models and applications emerging in months rather than years, creating a challenge no actor can meet alone . He says multilateralism is essential to guide progress by shared principles and common benefits, warns against a widening AI divide in technology, skills, data and computing power, and presents UNIDO’s industrial AI work, alliances and centres of excellence as practical examples that create jobs, strengthen industries and build local capabilities .

He argues that the main gap in AI governance is not ambition or principles but the lack of practical mechanisms. Governance needs to move into concrete cooperation on standards, infrastructure and skills across all major dimensions of AI so that benefits become real development outcomes.

He says the gap today is not one of ambition but of practical mechanisms, and calls for value generated by AI to flow broadly across workers, languages, regions and generations . He explicitly frames the session around moving beyond high-level principles towards concrete conditions, cooperation, standards, infrastructure and skills, and sets out five lenses for discussion: social, economic, cultural and linguistic, technical and environmental .

He argues that many actors already share the ambition to make AI beneficial, but implementation is where progress stalls. What matters now is building the concrete machinery—standards, infrastructure and skills—that can turn AI’s promise into practical gains.

He says directly that the gap is not ambition or principles, but practical mechanisms . He then specifies that the task is to move beyond governance principles towards concrete conditions, cooperation, standards, infrastructure and skills that can translate AI’s productivity, scientific and sustainability gains into tangible development outcomes .

In his closing synthesis, he argues that merely accessing AI is not enough; countries need the capacity to shape outcomes. He distils the discussion into three action-oriented priorities: capacity rather than access alone, enabling environments rather than compute races, and a model where sovereignty coexists with openness.

He summarises the afternoon by saying access alone does not create prosperity or jobs and that the biggest risk is remaining spectators while others capture value . He adds that the right strategy is to build on digital public infrastructure and AI literacy rather than chase compute races, and that sovereignty and openness are partners through local models in local languages using data that belongs to users while still working with frontier models .

He argues that the right measure of AI success is social and developmental impact, not technical scale. AI should be assessed by whether it improves livelihoods, supports communities, keeps languages alive and spreads benefits broadly.

In his closing remarks, he says AI will not be judged by how many models are deployed but by how many jobs are created and uplifted, how many languages thrive digitally and how many communities see their knowledge reflected rather than erased . He links this to his earlier call to redesign systems and set new measures of success in response to AI’s transformative nature .

He argues that AI offers particular opportunities for developing countries by converting tacit and unstructured knowledge into usable form. However, these benefits will only be realised if AI is adapted to local languages, contexts and cultural realities instead of reproducing dominant Western assumptions.

He says AI can convert unstructured data into structured data, which can disproportionately benefit developing countries because much of their knowledge is tacit and unstructured, and can improve intergenerational knowledge transmission . He also notes that AI is largely shaped by Western culture and values, but that the real opportunity lies in local and regional adaptation from big AI to smaller AI that meets people where they are cognitively and culturally .

He argues that the prevailing model of AI development is neither sustainable for advanced economies nor feasible for lower-income ones. This means countries in the Global South need distinct AI pathways aligned with their own constraints, populations and resources.

He says advanced economies cannot sustainably keep up the current model and middle- and low-income economies cannot even aspire to doing AI in the same way as the biggest economies . He then argues for AI developments tailored to Africans, Latin Americans, Europeans outside dominant centres, small islands and the Caribbean, while noting that energy, water and even future data-centre regulation will need to be handled differently .

He argues that discussion alone is insufficient and should immediately be converted into concrete, resourced commitments. The dialogue should be judged by implementation, with participants able to report actual action by the next major meeting.

He challenges every delegation, company and institution not to wait for the summary and instead to take one of the three priorities and give it a name, a date and a budget before leaving Geneva . He adds that when the dialogue reconvenes in New York in 2027, participants should report what they actually did rather than what AI could do .

She frames AI governance around human-centred use rather than technological acceleration for its own sake. In drawing out Estonia’s example, she suggests successful societies will be those that keep people at the centre and focus on wise adoption rather than speed alone.

She says at the ITU that AI will only succeed if it remains centred on people and introduces Estonia as a country that has demonstrated this clearly . After President Karis’s response, she endorses his point that AI must be used wisely, not just quickly, and closes by stressing that AI’s future will be shaped by choices about investing in people, trusted digital foundations, innovation and cross-border cooperation .

He argues that calls for cooperation have become too abstract and repetitive. What is now needed is concrete agreement on what countries are cooperating on, what they want to achieve, and how they will act without either over-regulating or losing control of AI’s direction.

He says that although people often repeat that cooperation is needed, the missing element is clarity about cooperating on what, where to go and what to achieve . He adds that actions are lagging behind technology, and that policymakers face the challenge of avoiding over-regulation while still guiding AI onto the right track through specific and precise cooperation .

He argues that smaller countries have a shared interest in defending linguistic diversity and human rights in AI governance. At the same time, they must strike a balance between preventing harm and not stifling technological development through excessive regulation.

He says small states need to unite and not be underprivileged on language issues, while also highlighting human rights and Armenia’s participation in relevant international frameworks . He later says policymakers do not want to over-regulate and harm technological development, but still need to facilitate adoption without fear and send the technology in the right direction .

The speaker argues that AI can support sustainable development only if it is guided by ethics throughout its lifecycle. Multilingualism, multi-ethnicity and respect for human rights are presented as essential to building trust and avoiding exclusion or bias.

The delegate says ethics should be a cross-cutting principle across design, development, implementation, use and assessment, and that this approach helps build public trust and prevent risks such as bias and exclusion . The delegate also stresses Guatemala’s multilingual and multi-ethnic reality, arguing that linguistic and cultural diversity broadens the perspectives from which relevant AI solutions can be developed .

The speaker argues that international and regional cooperation are necessary to make AI governance more inclusive and representative. Forums should be used to share experience, forge consensus and support governance aligned with different national contexts.

The delegate says Guatemala will continue to participate actively in international cooperation forums to share experiences, craft consensus and promote AI governance that is more inclusive, representative and human-centred .

The speaker argues that AI has broad potential benefits, but these will only be realised if countries invest in people and enabling conditions. Skills, trust, infrastructure and cooperation are presented as the foundations for equitable access to AI’s social and economic gains.

The speaker says that AI can improve public services, healthcare, education and public administration, but that realising these opportunities requires investment in skills, public trust and responsible deployment . The speaker also says Slovenia is focusing on infrastructure, knowledge transfer, financing and stronger cooperation between research, industry and the public sector, and concludes that international cooperation is necessary to ensure AI benefits are accessible to all .

The speaker argues that multilingualism and cultural diversity should be treated as strengths rather than obstacles in the digital age. AI must support smaller language communities so that digital transformation remains inclusive.

The speaker says Slovenia, as a small community, places particular importance on the cultural and linguistic dimensions of AI . The speaker adds that AI should support all languages, including those with smaller speaker communities, so that no language is left behind in digital transformation .

The speaker argues that broad AI adoption across smaller firms and public institutions depends less on hype than on solid enabling conditions. Infrastructure, financing, knowledge transfer and technical foundations are necessary if AI is to be widely and responsibly used.

The speaker says AI offers opportunities for productivity, innovation and competitiveness, particularly for SMEs, and that Slovenia’s focus is on creating conditions for wider adoption through infrastructure, knowledge transfer, financing and cooperation between research, industry and the public sector . The speaker also stresses that access to compute, high-quality data, interoperable infrastructure and digital skills remain essential, and points to national AI infrastructure investments including an AI factory and competence centre .

The speaker argues that inclusion in AI governance should be judged by whether developing countries can influence outcomes, not simply participate symbolically. This requires practical support tools such as lightweight models, procurement guidance and culturally inclusive interoperability frameworks.

The speaker says meaningful inclusion is measured not by participation alone but by influence, and argues that developing countries must be active contributors to data sets, standards and models . The speaker proposes practical outcomes including global repositories of lightweight AI models, public-sector AI procurement guidelines, and embedding cultural and linguistic benchmarks within interoperability and AI safety frameworks .

The speaker argues that expanding AI in developing contexts must not reproduce a high-consumption model. Instead, AI should be designed and deployed in greener, more resource-efficient ways.

The speaker says that bridging the AI divide requires compute infrastructure and sovereign data capabilities, but adds that AI development must also be environmentally sustainable through green AI and resource-efficient, frugal AI .

The speaker argues that AI’s benefits will not emerge automatically from technology deployment. Public institutions, workers and wider society need preparation through skills, transition support and literacy so that AI strengthens rather than displaces human capability.

The speaker says capacity building should extend across the entire public sector and that reskilling initiatives and just transition policies are needed to ensure AI enhances human capabilities and distributes economic benefits equitably . The speaker also highlights special attention to AI literacy for children and minors .

The speaker argues that meaningful access to AI is impossible without material and financial foundations. Developing countries require compute, data sovereignty and sustainable financing if they are to create their own AI ecosystems rather than rely permanently on others.

The speaker says meaningful access to AI requires more than capacity building and specifically calls for investment in compute infrastructure, sovereign data capabilities and sustainable financing mechanisms to enable developing countries to build their own AI application ecosystems .

The speaker argues that governance will remain unequal if developing countries are merely rule-takers. They need a direct role in standards, data and model development to ensure AI governance reflects global diversity rather than entrenched concentration.

The speaker says AI governance must reflect cultural diversity and prevent excessive technological concentration, and that developing countries must be active contributors to AI data sets, standards and models .

The speaker argues that children and minors deserve explicit attention within AI readiness efforts. Literacy for younger users should be elevated as a distinct public priority rather than assumed to follow from broader digital skills initiatives.

The speaker states that Egypt pays special attention to AI literacy for children and minors while discussing public-sector and social preparation for AI .

The speaker argues that AI touches nearly every aspect of social life and should be governed accordingly as a people-centred development issue. Developing countries must not just receive systems built elsewhere; they should participate in building, assessing and governing systems suited to their own societies.

The speaker says AI bears directly on economic participation, public services, culture, language, education, employment and rights, and that its implications must be understood through the varied ways people encounter technology . The speaker later says countries should not only receive AI systems delivered elsewhere but must be able to build, adopt, evaluate and govern systems reflecting their own languages, institutions, cultural context and development priorities .

The speaker argues that AI governance must address the treatment of cultural and creative material as more than just training inputs. Recognition, consent, attribution and fair remuneration are necessary to prevent exploitation of cultural expression and traditional knowledge.

The speaker says that in the creative sector governance must address recognition, consent, attribution and fair remuneration . The speaker adds that creative works, cultural expressions and traditional knowledge should not be treated merely as raw material for technological development .

The speaker argues that AI risks are distributed unequally across society. Governance should therefore focus on the groups that are more exposed to displacement, exclusion or misuse and ensure stronger safeguards and support for them.

The speaker says AI can displace tasks, widen skill gaps, reproduce bias, enable manipulation, misuse personal data and weaken accountability . The speaker then identifies workers, children, persons with disabilities, indigenous peoples, creators, small enterprises and communities with limited digital representation as groups facing particular risks .

She argues that AI should be used as a practical tool for better services and modernisation rather than as an end in itself. At the same time, deployment must preserve human oversight, protect data and uphold dignity and non-discrimination.

She says Belarus is using AI to improve quality of life, modernise the economy and public administration, and deploy systems in healthcare, education, social services, logistics, urban planning and climate-related management . She also says human control over critical decisions remains extremely relevant, alongside responsibility for algorithmic decisions, data protection and the prevention of discrimination .

She argues that AI governance must keep humans at the centre and prevent algorithmic systems from making unaccountable or discriminatory decisions. Human societies, not machines, must remain responsible for the outcomes of technological change.

She says the issue of human control over critical decisions remains highly relevant and explicitly mentions responsibility for algorithmic decisions, data protection and the prevention of discrimination . She adds that human beings are at the heart of the technological revolution and that AI must develop for the good of humanity, on the basis of developer accountability and protection of human rights and dignity .

He argues that AI should be embedded in a broader national development strategy and treated as a tool for human rights and the public good. Governance should examine not only what AI can do, but what it does to human beings and democratic life.

He says Mexico believes it is important to address technologies such as AI in terms of what they can do with human beings, not just what they can do for us, and welcomes the scientific panel’s focus on human rights, democracy and cultural diversity . He adds that rapid AI development is changing the social contract, power relations, economic activity and social networks, and notes that Mexico has made technological development a cross-cutting pillar of its national development strategy and seeks public technological solutions as a common right .

She argues that AI development is not an inevitable path but a political choice that should be shaped around justice and fairness. Governance must include the material and territorial conditions of AI, not just its visible applications.

She says AI is often discussed as a tool for sustainable development, but that governance debates neglect the infrastructure making it possible, including data centres, water, energy, minerals, supply chains and labour . She concludes that AI development should not be treated as something already decided and that societies can still choose what type of AI to build and for whom, moving forward with technical responsibility .

She argues that environmental transparency must be built directly into AI governance as a technical and political requirement. Countries, especially in the Global South, need the capacity to measure, audit and respond to the territorial impacts of AI infrastructure.

She says there are huge gaps in measuring and comparatively disclosing energy use, water footprint, critical materials and e-waste across the AI lifecycle, and calls environmental transparency a technical principle of AI governance . She also argues that countries need technical capacity to evaluate, audit and adapt AI systems to their social, environmental and regulatory contexts, and notes that communities where minerals are extracted or data centres are built often bear concentrated environmental costs while being excluded from decision-making .

She argues that there can be no genuinely inclusive AI dialogue if environmental questions are marginalised. AI is materially tied to energy systems, water, minerals, waste streams and climate goals, so governance must address these interdependencies directly.

She says AI will shape and be shaped by energy systems, water systems, mineral supply chains, waste streams, climate goals and planetary boundaries, and that there is no truly inclusive AI dialogue if the planet appears only in the footnotes . She also lists environmental opportunities such as monitoring methane, biodiversity loss, deforestation, pollution and climate risks, while stressing the need to understand AI’s full end-to-end footprint including minerals, manufacturing, water, electricity, e-waste and rebound effects .

She argues that the structure of the dialogue itself needs to explicitly include the environment rather than treating it as secondary. An inclusive AI agenda must name the planet as a central concern in both current and future discussions.

She notes that the agenda lists many dimensions of AI but not the environmental one, and says that if the planet is only present in the footnotes, the dialogue cannot be truly inclusive . She closes by hoping that next year the environment is named clearly on the agenda because the future of AI cannot be separated from the future of the planet .

She argues that AI’s environmental burden is not incidental but built into how AI is currently developed and deployed. Governance should therefore use precaution and legally binding transparency and accountability across the full supply chain instead of treating AI expansion as unavoidable.

She says AI is not inevitable and that current development is driven mainly by the economic interests of a few companies and countries . She describes AI’s supply chain from mining to data processing as requiring immense natural resources, says its environmental footprints are structural rather than accidental, and argues for precaution, transnational transparency, accountability and redress, including mandatory disclosure of water, energy and supply-chain impacts .

She argues that manipulative design is embedded in commercial business models rather than being a rare technical flaw. Because those incentives are structural, oversight must rely on enforceable obligations, independent scrutiny and rights-based assessment rather than voluntary self-regulation.

She says manipulative design is not a bug and combines social engineering with technology inside business models that depend on data and attention for revenue . She argues that self-regulation has not been sufficient, and calls for mandatory transparency, access to disaggregated data, reporting when risks are detected, and human rights impact assessments throughout system lifecycles .

She argues that waiting for full scientific consensus can leave affected communities without protection while harms continue. Civil society, especially those directly affected, must therefore be included in documentation, oversight and redress mechanisms from the outset.

She says effective mechanisms require meaningful engagement of civil society, particularly affected communities, at the transnational level . She also argues that while scientific panels are important, documentation of impact cannot be restricted to academic consensus because communities affected by AI harms do not have time to wait, and calls for support for independent documentation and stronger UN human rights monitoring mechanisms .

She argues that education and literacy are important but cannot substitute for regulation of harmful systems. Vulnerable users should not bear the burden of protecting themselves from abusive or exploitative AI design.

She says manipulative design especially exploits people who are already vulnerable, including children and historically marginalised groups . She adds that digital, data and AI literacy are important, but they cannot become an excuse for impunity and that placing responsibility only on end users is not enough .

He argues that governance of AI’s resource footprint is impossible without shared measurement rules. International standardisation should begin even with seemingly basic units such as the definition of a token, because current inconsistency undermines comparison and cost transparency.

He says the scientific panel’s report includes a diagram showing that models and the computers they run on consume substantial energy and electricity both in training and use, and that there is currently no standardised measurement of these costs . He adds that one basic area needing standardisation is the definition of a token, because different organisations define it differently, which makes cost and usage hard to compare .

He argues that AI governance must address both the use of AI to support environmental goals and the environmental cost of AI itself. This means looking across the full lifecycle of AI systems—from minerals and land through compute and models—and embedding sustainability in governance design.

He frames the issue as both AI for green and green AI, noting opportunities such as improving food systems and farmer resilience while insisting that environmental sustainability must be part of AI’s core design . He then says governance must go from ‘mine to model’, covering water, minerals, land, compute and even child labour in mines, and calls for standards and reporting across the AI lifecycle .

He argues that global governance needs common standards and shared implementation mechanisms to make AI’s environmental footprint governable. The UNEA resolution should be translated into practical standards exchange, measurement and capacity building.

When asked about the UNEA resolution, he says it was difficult to negotiate and that what must happen next is to include environmental sustainability in core processes and AI governance . He calls for standard measurements and reporting across the AI lifecycle and references the UN High-Level Advisory Body’s idea of a global environmental AI standards exchange, stressing that governance should support transparency and capacity building rather than only punishment .

She argues that AI offers major benefits, including in sustainability and public welfare, but these gains depend on safe, secure and trustworthy governance. The right approach is flexible and nuanced: enabling innovation while managing risks responsibly.

She says AI offers opportunities for sustainability and inclusive development, including disaster risk reduction, healthcare and climate change mitigation applications . She also says irresponsible AI use presents risks, and that Australia’s approach is to capture opportunities while building systems, infrastructure and frameworks that enable confidence in the use of AI .

She argues that AI governance should be anchored in established rights and legal protections rather than only technical performance. Human oversight and protection for vulnerable groups must remain central as AI systems spread.

She says AI governance should realise opportunities in ways that are safe, inclusive and consistent with international human rights law, while addressing privacy, equality, freedom of expression, security, cultural and linguistic diversity and meaningful human oversight . She also highlights the need to pay close attention to groups already disadvantaged by digital and economic gaps, including First Nations people, women, people with disabilities and remote communities .

She argues that AI governance must explicitly account for the unequal exposure of different groups to digital and economic disruption. Inclusion requires designing policy around those already facing structural disadvantage.

She says policymakers need to pay close attention to cohorts already disadvantaged by digital and economic gaps and at higher risk of AI- and automation-driven disruption . She then lists First Nations people, women, people with disabilities and regional and remote communities, and notes Australia’s commitment to align AI action with its national agreement on closing the gap in partnership with First Nations communities .

He argues that successful AI adoption depends on social trust, widespread skills and solid digital infrastructure rather than enthusiasm alone. Without these conditions, AI is likely to reinforce inequality and leave parts of society behind.

He says AI should prompt societies to rethink how they work and that the main issue is not using it first and fast but using it wisely . He emphasises trust in society and government, transparency of data use, skills for the whole society, and digital infrastructure, warning that without these conditions inequality will grow and that access remains especially difficult where even electricity is absent .

He argues that education is the best place to begin national AI preparation. Teachers and students need structured support to learn what AI can and cannot do, and smaller countries can use partnerships with platform providers instead of trying to build every system themselves.

He describes how Estonia decided to begin with education, first training teachers through summer courses on AI’s opportunities and risks and then starting with upper secondary schools so students would have these skills before university . He says Estonia collaborated with OpenAI and Google to build a platform for schools because, as a small country, it could not build its own alone .

He argues that language survival in the AI era depends on access to contemporary linguistic material, not just historical texts. Without this, speakers of small languages may increasingly shift to dominant languages such as English.

He says language is extremely important for a small country and that AI platforms must understand Estonian and have access to modern literature, newspapers and other contemporary language sources . He gives the example of a major newspaper allowing one institution to use all of its articles from the nineteenth century to the present because modern language is needed, and warns that without such efforts young people may switch to English .

He argues that AI education in schools should not merely familiarise students with tools but should strengthen critical thinking and judgement. Literacy should begin with teachers so that schools can teach students how and when to use AI wisely rather than passively rely on it.

He says Estonia began by training teachers and upper secondary students, with courses covering AI’s possibilities and risks and resulting in full familiarity among teachers and students with available options . He also explains that the school platform was designed not simply to give answers but to discuss with users and make them think, linking this directly to the need for critical thinking .

He argues that simply having access to AI tools does not automatically generate jobs, value or prosperity. What matters is whether countries have the capacity and enabling systems to shape, govern and locally adapt AI in ways that serve their own development priorities.

He says the real question is no longer who has access to AI but who has the capacity to shape it, and repeats that access alone does not create prosperity, jobs or value . He then describes Smart Africa’s systems approach, saying AI success requires connectivity, trusted data systems, digital identity, effective institutions, governance and trust, and must be built on digital public infrastructure and human capacity .

He argues that Africa’s comparative advantage lies not in duplicating frontier-model competition but in adapting AI to local realities. AI should be judged by whether it solves practical problems for citizens in agriculture, health, education and government while respecting language and cultural context.

He says Africa’s opportunity is not necessarily to compete in building every frontier model but to apply AI where it matters most to citizens, such as agriculture, healthcare, education and government services . He also stresses that language, culture and context matter, noting Africa’s more than 2,000 languages and saying locally adapted solutions can outperform imported ones because they better understand local needs .

He argues that middle-power and Global South countries should avoid wasting resources trying to compete directly in infrastructure-heavy AI races they are unlikely to win. Instead, they should focus on leadership, workflows, skills and the use of widely available tools that can already create value.

He advises ministers not to spend their budgets trying to win the compute or infrastructure game, saying they are not going to win it and do not need to . He says they should invest in the enabling environment and complementary factors, calling AI a human and societal problem of relearning skills, reinventing workflows and leadership . He also notes that powerful AI is already available through free resources such as YouTube, Ollama and open models, and compares this to how the Global South made mobile technologies especially useful despite not inventing frontier technologies .

The speaker argues that countries will benefit most from AI not by moving first, but by becoming better at adapting, practising decisions and correcting course. Scenario planning is presented as a practical way for governments to prepare for both opportunities and risks before systems become locked in.

The speaker says countries that benefit from AI will be those that can practise, adapt and course-correct fastest, and explains that scenario planning is not just crisis preparation but also preparation for opportunities . The speaker gives examples of readiness questions around jobs, cyberattacks, language failures and worker pressure, and says public officials in Southeast Asia want practical help with procurement, evaluation, inter-agency coordination, cyber-risk management and public communication .

He argues that countries themselves are best placed to encode their own language, culture and values into AI systems. Rather than expecting foreign firms to understand local context better than domestic actors, countries should build internal capability to do this work.

He says the cultural and linguistic dimension of AI is very important and that in each country local people are the best experts in their own culture, language and values . He adds that the necessary tools already exist and that people, especially younger people, within each country can create solutions for their own cultures and languages, noting that Oman has done this and believes every country can .

He argues that linguistic and cultural diversity is central to meaningful inclusion in AI. Countries that do not build systems reflecting their own realities risk marginalisation within the AI economy.

He says the cultural and linguistic dimension of AI is very important and that countries cannot expect foreign companies to know their language, culture and values better than they do themselves . He closes by calling on participants to shape AI with wisdom so that the technology reflects the diversity of the planet .

The speaker argues that the pace of AI development is outstripping social and institutional preparedness. Governance should therefore focus on helping classrooms, labour markets and public consensus adapt quickly enough to keep up with the technology.

The speaker says the gap between the speed of AI development and readiness is becoming apparent because AI is advancing faster than laws, institutions and society can adapt . The speaker gives examples from Korea’s own classroom AI rollout, where stakeholders drew different lessons, and points more broadly to self-driving vehicles, misleading AI-generated content and young people’s worries about jobs as evidence of the urgency of faster social adaptation .

He argues that equitable AI development requires access to technology, compute, data and open-source tools on a non-discriminatory basis. Without this, countries risk becoming dependent on dominant powers, which he characterises as a form of digital neocolonialism.

He says Russia is committed to bridging the digital divide, raising labour productivity, ensuring non-discriminatory access to technologies, computing capacities and data, and developing open-source computer models while upholding security requirements and human rights . He later says non-state stakeholders should help build trust among countries and peoples, and that only such work will safeguard against digital neocolonialism .

He argues that AI governance must align with state sovereignty and the laws of the countries where systems are deployed. Trusted AI, legal compliance and non-interference are presented as essential components of a fair international regulatory system.

He says it is essential that AI system developers comply with the laws of the countries in whose territories those technologies are applied . He also says Russia supports the creation and deployment of trusted AI systems and seeks a fair and equitable system of international AI regulation consistent with the UN Charter, especially sovereign equality and non-interference in internal affairs .

He argues that the current dialogue should not start from scratch but should build on earlier multilateral efforts. Continuity across summits and UN processes is necessary to create a fair and effective global regulatory framework for AI.

He welcomes the launch of the Global Dialogue as a UN platform and notes the work already done by member states, the UN Secretariat and the scientific panel to agree on modalities and objectives . He specifically calls for the outcomes of the AI Summit in New Delhi to be built upon in further UN work .

He argues that publishing AI strategies is not enough if countries lack the means to execute them. Real leadership should be judged by implementation capacity, budget and measurable public value rather than formal policy announcements alone.

He says that the AI for LAC regional dialogue made clear the need for the region to move from adoption to agency and that this requires stronger institutions . He then cites ILIA 2025, saying that while 9 of 19 countries have national AI strategies, only a few have budgets, implementation plans or impact indicators, and concludes that leadership should be measured by whether countries can convert policy into trusted, scalable and measurable public value .

He argues that high-performing AI on generic global benchmarks may still fail in local contexts if it does not understand regional language and institutions. Local relevance is therefore a condition for technological agency and for resisting externally imposed priorities.

He says culture and language are not peripheral concerns and that systems failing to understand local language, institutions and social realities will produce weaker outcomes regardless of their performance on global benchmarks . He also notes that the AI for LAC regional dialogue warned that the region must avoid becoming merely a market, a source of data or a testing ground for external priorities .

He argues that regional inequalities in investment and compute demonstrate that AI opportunity is not evenly distributed. To avoid remaining dependent consumers, regions such as Latin America need stronger institutions and technological agency.

He says opportunity is not universal when compute, investment, standards, languages and decision-making authority are concentrated in a small number of countries and companies . He provides regional data, stating that Latin America and the Caribbean account for 6.6% of global GDP and 8.8% of the world’s population yet receive only a small fraction of AI investment, and that more than 90% of the region’s high-performance computing capacity is concentrated in a single country .

He argues that the region’s challenge is not a lack of ambition but a lack of agency over infrastructure, development and value capture. The response should be to move beyond adoption of imported systems towards building local capabilities and governance power.

He says structural imbalance is visible in AI investment and compute concentration and describes this as a gap in technological agency rather than ambition . He then states that the AI for LAC regional dialogue concluded that the region must move from adoption to agency by adapting models, developing solutions, governing strategic data, evaluating impacts and shaping standards .

He argues that AI inclusion is incomplete if systems only understand words but not the social worlds in which those words are used. Cultural understanding of values, institutions and lived practices is necessary for trusted and sovereign AI.

He says that language is only the first dimension, and that true symmetry requires culturally inclusive AI capable of recognising local value systems, traditions, institutions, farming practices, legal systems and cultural norms . He adds that AI must understand not only how people speak but how they live within their societal constructions .

He argues that smaller and developing nations should help build the knowledge base of future AI rather than simply adopt systems built elsewhere. Their languages and cultural perspectives are essential inputs into more inclusive AI development.

He says several hundred languages and dialects remain underserved and warns that, if left to market forces, AI will evolve around the largest languages, leaving many communities on the margins of the AI economy . He later states that smaller and developing nations must not be constrained to being consumers of AI but should be empowered as contributors to the knowledge, language and cultural perspectives on which future AI systems are built .

He argues that meaningful inclusion requires sovereign AI capacity so that countries can deploy systems aligned with their own goals and constraints. This sovereignty should protect independence, resilience and sensitive data while still enabling participation in AI progress.

He says deep and meaningful inclusion raises the imperative of sovereign AI, defined as the ability for countries to deploy AI in ways reflecting national priorities, cultures and languages without compromising independence, sensitive data or national resilience . He also describes Sri Lanka’s digital transformation architecture, which treats AI-powered language equalisers as digital public infrastructure and integrates them into multilingual government and agricultural advisory services .

He argues that current AI markets are excessively concentrated, creating strategic vulnerabilities and unequal opportunity. Global AI governance should therefore help create fairer competitive conditions across the full AI stack.

He says Finland underlines open competition and innovation and notes a high degree of market concentration in AI chips and models, largely in the hands of a few global entities . He adds that this concentration creates strategic dependencies and vulnerabilities in critical supply chains and asks how governance can help build a more level playing field where businesses of different sizes and regions can compete and innovate fairly .

He argues that a healthy AI ecosystem should preserve space for both open and closed models to compete fairly. Open-source models in particular can lower costs, support local language use and help countries maintain digital sovereignty.

He says the first priority is to protect the open global AI ecosystem so that frontier models, whether open-source or closed-source, can compete fairly and users worldwide can benefit from technological advancement and competition . He gives examples from Chinese open-source models, saying they have gained substantial global downloads and usage because of high performance and low cost, support local languages including over 100 languages, and allow users to keep data locally, thereby supporting digital sovereignty .

He argues that multilingual open-source models can help under-served regions build AI on their own terms. By supporting local languages and local data retention, they contribute both to inclusion and to sovereignty.

He says open-source models support digital sovereignty because users can keep data locally . He adds that models such as QN support over 100 languages and provide a foundation for AI developers in Africa and Southeast Asia that have long been overlooked .

He argues that while sovereignty has a role in AI governance, overreliance on sovereignty-based approaches can fragment data and undermine cooperation. AI, like the internet before it, requires interoperable and international governance to avoid splintering.

He compares the current AI debate with the earlier choice between a single interoperable open internet and a patchwork of national splinternets, arguing that the open model prevailed and enabled global innovation . He says AI requires even greater global cooperation because sovereignty-based restrictions can frustrate collaboration on bias and diversity and can fragment the data needed for training by imposing restrictive localisation regimes .

She argues that meaningful governance depends on shared concepts and standards before any assessment or verification can work. Common vocabularies are the basis for comparing outcomes across jurisdictions and building robust standards in multiple areas of AI governance.

She says governance is strengthened by staying lifecycle-oriented and notes that shared standards begin with shared terminology . She adds that without common vocabularies and taxonomies, jurisdictions measuring different things under the same label cannot compare results, and she links this need to interoperability, cultural and linguistic inclusion, environmental measurement and child safeguards .

She argues that governance should not stop at design intentions or developers’ self-assessments. Instead, it should follow AI systems through their whole lifecycle and evaluate real-world impacts with external verification.

She says the scientific panel observes that evaluation of agentic AI systems faces standardisation and reproducibility challenges and that developer-defined risk thresholds currently operate without standardised evaluation or external verification . She then argues governance is stronger when it is lifecycle-oriented from conceptualisation through retirement and assesses actual impacts alongside stated intent .

She argues that child protection in AI should be built directly into technical design rather than added later. Age-appropriate safeguards can and should be engineered into systems as concrete, testable requirements.

She says the panel finds that current AI systems can amplify risks to children even as deployment outpaces evidence and regulatory safeguards . She then argues that the answer is protection by design, and that age-appropriate design can be translated into testable technical requirements without constraining innovation .

She argues that unequal countries can still cooperate effectively if each contributes what it does best within shared frameworks. Interoperability is not about identical capacities, but about connecting complementary strengths through governance and public-private collaboration.

She says DCO was created to convene countries, cooperate and share best practices, based on the idea that every country has a competitive advantage . She gives the example that not every country needs to build a 100-megawatt data centre; some may provide compute infrastructure while others offer brain power, and shared IP and governance frameworks can allow local content to develop faster with private-sector partnership .

She argues that countries will remain dependent if they do not improve their foundational inputs into AI. High-quality data, local model creation and resilient infrastructure are all necessary if countries are to become AI producers rather than passive consumers.

In the lightning round, she says the first priority is data quality because increasingly intelligent AI is built on that fundamental data . She then says nations must create their own models and algorithms instead of remaining only users, and adds that resilient infrastructure is needed to withstand environmental catastrophes, political tensions and economic crises .

He argues that data governance sits underneath AI governance and should be treated as its basic layer. Interoperability is both necessary and achievable, and it does not require countries to give up sovereignty if cooperation is designed properly.

He says that data is the foundation upon which AI governance will be based and describes the CSTD working group on data governance and its four tracks, including interoperability and cross-border data flows . He explains that interoperability is highly desired from both development and business perspectives and says it does not necessarily require harmonising national laws because cooperation, agreements, contracts and common understanding can safeguard sovereignty while promoting interoperability and data flows .

He argues that interoperability should be treated as a practical enabler for both economic activity and development policy. Cross-border data cooperation is necessary to realise AI’s benefits while keeping national legal systems intact.

He says interoperability is desired from both a development perspective and a business and entrepreneurial perspective, and that one cannot choose one over the other . He also notes that cross-border data flows and interoperability are closely related tracks in the working group and that all of these are being approached as developmental issues requiring capacity, skills and financing .

He argues that AI’s appetite for data must not override the rights of the people from whom that data comes. Governance should therefore protect privacy and consent while ensuring the benefits of data use are shared fairly with data producers and creators.

He says AI ‘gobbles up’ data but that data belongs to people, reflecting their privacy and rights, including children’s rights, and is not necessarily freely available or consented for use . He calls for a UN-based architecture and decent framework to treat data with utmost care and to share the benefits of data with those who produced it, also noting misappropriation of creators’ data in AI processes .

As moderator, she frames the first panel around how AI can be built and governed to include linguistic diversity, local knowledge, environmental accountability and open access. Her role is not to advance a single substantive position, but to define the central governance questions for discussion.

She opens the panel by asking how large-scale AI breakthroughs can ensure systems reflect linguistic diversity, local knowledge and regional realities, referencing Estonia’s concern with preserving language and culture . She then asks questions on environmental footprint, open data and open AI models, local job creation and manipulative design, thereby setting the panel’s agenda around inclusion, sustainability, openness and accountability .

She frames the second panel around the practical machinery needed for AI governance rather than repeating general principles. She warns that serious blind spots remain in evidence, especially where systems are not assessed using sex-disaggregated data.

She says the panel is about the machinery that makes governing possible because governments cannot govern what they cannot measure, measure what they cannot access, or scale what does not interoperate . She also warns that half of what matters is still not being measured and specifically notes that most systems discussed have never been evaluated on data disaggregated by sex .

The co-moderator argues for a structured speaking process that balances state and stakeholder participation while keeping the session manageable. Procedural discipline is presented as necessary to maximise the number and diversity of voices heard.

The co-moderator explains that interventions will alternate between one representative from member states and one from stakeholders in order to promote broad participation from all stakeholders . The co-moderator also notes that a timer is visible, microphones will switch off automatically when time expires, and speakers are requested to respect the limit so that the largest possible number can be accommodated .

This procedural intervention highlights that multilingual participation requires logistical support and clear time limits. The interpreters’ flexibility enables broader participation, but only within practical constraints.

When asked for indulgence because the meeting was running late, the interpreters responded that interpretation could be provided until 6:15, thereby setting the operational limit for continued multilingual participation .

The Chancellor presents the cluster not as an isolated event but as part of a wider process intended to influence later plenary work and future action. The emphasis is on carrying conclusions forward rather than ending with discussion.

At the end, the Chancellor thanks the co-chairs for capturing key outcomes and reporting them back to the main plenary in the next day’s ‘Dialogue of Dialogues’ . The Chancellor adds that participants look forward to reconvening in order to turn these insights into action .

He argues that multilingual AI inclusion depends on three things: solving data scarcity through technical advances, collecting data directly with local communities, and designing systems for deep cultural nuance. Language support alone is not enough unless models are evaluated against culturally relevant benchmarks.

He says nearly half of the training data of major AI models is in English even though English accounts for only 20% of spoken languages, which illustrates the scale of linguistic imbalance . He then describes Google’s three-pillar approach: machine-learning breakthroughs including a thousand-language initiative and 110 new Google Translate languages including 60 African languages, community-based spoken-language data collection in Africa, and open-sourced speech data in India spanning 150,000 hours across 773 districts, plus benchmark design for cultural nuance .

He argues that policymakers should distinguish between areas where evidence is already sufficient to justify immediate action and areas where more transparency is needed to strengthen the evidence base. Delay is unacceptable in either case, though the policy response may differ.

He uses a climate-governance analogy to warn that AI may be repeating history on a compressed timeline, moving rapidly from technical breakthroughs to widespread use and serious risk . He says delay would be a dereliction of duty, argues that action is already justified on harms such as sexual violence against women and children and on interoperability failures in digital health, and says that where evidence is thinner the international community should build transparency-enabling standards and comparable metrics .

The speaker argues that AI governance has focused too much on whether systems are safe before release and too little on what happens once they are deployed in real institutions. Effective governance must therefore include responsibility, appeal and transparency at the point where systems actually affect people.

The speaker says most governance architecture over the past five years has focused on the moment before release and rarely asks whether people later affected in a clinic, benefits office or classroom can contest or correct decisions made about them . The speaker gives examples of automated systems flagging citizens for fraud or denying benefits without identifiable decision-makers and argues standards must operationalise accountability across the supply chain, due process through appeal and remedy, and openness backed by guaranteed access for independent researchers .

She argues that children are among the most exposed users of AI and therefore need explicit design protection. Systems affecting them should be built around inclusion, child rights and developmental needs rather than around commercial incentives.

She says UNICEF research from 10 countries shows children are adopting AI three times as fast as the adults raising them, and that AI is invisibly woven into nearly every system and service they access . She then argues for three responses: design for inclusion from the start, treat children’s data as a rights issue because data profiles are built before children can even spell their own names, and ensure AI in schools and clinics follows child-development logic rather than commercial logic .

She argues that protecting children requires stronger evidence and enforceable assessment tools. Without child-specific data and benchmarking, AI deployment effectively becomes a large-scale uncontrolled experiment on children.

In the lightning round, she calls for a global evidence baseline on children’s access to and use of AI, their literacy and developmental impacts, saying that without it the world is running an experiment with children as guinea pigs . She also calls for child-centric benchmarking, clear child-rights red lines such as zero tolerance for child sexual abuse material, and mandatory child-rights impact assessments in education, health, migration, welfare and child protection systems .

He argues that developing countries need regional structures to build AI capability collectively rather than individually. Shared education, finance, infrastructure and legal frameworks can help them create locally useful AI capacity.

He says that after four years his country has moved from words to action through new governance, a specialist institution, GPU-based data centres, a national AI platform, and training for more than 3 million students and over 100,000 teachers . He then proposes five regional actions: a regional AI academy, a regional innovation and AI development fund, shared regional computing infrastructure for science and public good, a regional digital free zone and a legal framework to prevent harmful outcomes .

He argues that AI legitimacy depends on including developing countries in governance rather than allowing power to follow infrastructure concentration alone. Global rules should not be written only by the states with the biggest computational resources.

He says AI is legitimate only when it serves humanity, dignity and justice . He then says directly that global AI governance must include developing countries and cannot be shaped only by those with the largest data centres .

He argues that fragmented national approaches risk repeating earlier mistakes seen in telecommunications. West African countries should therefore cooperate regionally to pool investment, infrastructure and regulatory frameworks for AI development.

He says AI requires considerable investment and cross-cutting coordination and warns against repeating the regulatory fragmentation that delayed market development in telecommunications . He then points to community frameworks and common sectoral policies in West Africa and calls for regional cooperation, pooling of efforts and investment to build infrastructure, management tools and harmonised systems so AI can support health, security and education .

The speaker argues that Africa is not peripheral to AI because it provides key minerals underpinning global infrastructure. Fair AI governance must therefore include technology transfer, youth participation and protection of sovereignty rather than treating Africa only as a resource supplier or consumer market.

The speaker says AI is already an immediate reality and historical opportunity for African countries, and adds that AI history cannot be written without Africa because resources such as cobalt, coltan and copper from the DRC sit at the heart of AI infrastructure . The speaker then says the vision rests on opportunity and fair partnership, requiring obligatory technology transfer, involvement of youth, respect for national sovereignty and governance that protects democratic space from disinformation, electoral manipulation and technological dependency .