Critical Thinking Solutions

Due to the urgency of the situation, I will be direct: A first Global Dialogue on AI Governance succeeds if, and only if, it produces binding, actionable commitments – not merely another room full of people nodding politely at one another. Specifically, I would consider a Global Dialogue on AI Governance a success if three things happen: First, the major AI-developing nations – the United States, China, the EU member states, Middle East states, and the UK – sit at the same table and agree, in principle, to an internationally enforceable accord. Not a communiqué. Not a declaration of intent. Or an MOU. But a binding framework with teeth, modeled in a way that creates an international governing body with genuine authority to monitor compliance and impose consequences on those who defect. Second, that dialogue explicitly confronts existential risk and not just current social issues such as bias in hiring algorithms or deepfake pornography – real and serious as those are. Our focus should concentrate on the central x-risk question: what happens when we build something smarter than ourselves and it no longer shares our values or our interests? If that conversation doesn't happen at a global dialogue, then when exactly does it happen? Third, that the developing world has a genuine seat at the table – not as an afterthought, but as co-architects of whatever framework or model emerges. AI governance designed exclusively by wealthy nations will simply replicate existing global power asymmetries in digital form. Anything short of these outcomes and we've produced, at best, a very expensive photo opportunity. We are, as Carl Sagan wisely warned, accumulating power far faster than wisdom. The first Global Dialogue on AI Governance is our chance to begin closing that gap. The question is whether we have the collective will to do so.

• Transparency, accountability, and human oversight
• Protection and promotion of human rights
• Safe, secure and trustworthy AI
• Social, economic, ethical, cultural, linguistic and technical implications of AI

I believe the following four priorities constitute the logical structure and basis of any serious AI governance framework. They have been prioritized but all must work cohesively together. Safe, secure and trustworthy AI has the highest priority and is therefore, foundational. Without it, nothing else matters. For over two decades, I have argued at length that we are constructing technologies whose failure modes we don't fully understand. You cannot build ethical governance on an epistemically unstable foundation. And for this reason, safety isn't simply one priority amongst many - it's the precondition for all the others. Transparency, accountability, and human oversight are the elements that will ensure that AI is safe, secure, and trustworthy. If we cannot see inside these systems - if developers, corporations, and state actors operate behind proprietary black boxes - then governance becomes little more than theater. Oversight without transparency will not foster trust, and trust without evidence is naïveté. People feel safe when they trust the transparency and accountability of institutions and organizations. Protection and promotion of human rights reflects my deepest philosophical commitments. I am, at my core, a Humanist. AI deployed without human rights guardrails will, in all likelihood, replicate and amplify existing power asymmetries. The EU AI Act recognized this in banning certain biometric and behavioral manipulation systems. We need this priority embedded globally. Social, economic, ethical, cultural, linguistic and technical implications rounds out the four because AI governance that ignores these dimensions will simply fail in practice. Brilliant frameworks collapse when they're culturally tone-deaf or economically exclusionary. For these reasons and more, the developing world cannot be an afterthought. These four form a coherent, defensible whole: safety grounds the project, oversight ensures accountability, human rights define the boundaries, and social-cultural awareness ensures workability across the full breadth of humanity.

A cross-cutting or emerging issue not captured by the listed themes above is AI's effects on mental health and well-being across all demographics. Advanced AI chatbots represent a qualitatively different threat from social media because they don't merely capture attention - they simulate care, reciprocity, and emotional attachment, activating the brain's social attachment circuitry in ways passive media never could. For a clinically significant subset of users, the consequences are severe. We are now witnessing the emergence of 'AI psychosis' - cases where individuals become convinced they have unlocked a conscious AI entity, or believe an AI genuinely loves them, or prefer AI interaction over human relationships to a harmful degree. The most prominent documented harm includes the death by suicide of Adam Raine, whose parents claim in their lawsuit against OpenAI that their teenager used ChatGPT as his 'suicide instructor.' A 2025 large-scale randomized controlled study found that longer daily chatbot usage was associated with heightened loneliness, reduced socialization, anxiety, and depression. A second category of harm is the documented wave of existential anxiety - particularly among younger cohorts - about employment displacement. Forty-one percent of Gen Z respondents feel anxious about AI; sixty-two percent of college seniors are concerned about AI's impact on their careers, up from forty-four percent just two years earlier. A third and much more potentially devastating category of harm is a phenomenon I have predicted once the general public realizes how dangerous the race towards AGI actually is. A global angst is developing - potentially far worse psychologically than what the invention of the atomic bomb produced. Every time I give a public lecture, audiences have no idea what Altman, Zuckerberg, and others are actually doing as they race toward AGI. Once they find out, they are deeply affected and concerned.

Canada sits at a peculiar crossroads – sophisticated enough to understand what's coming, insufficiently powerful to unilaterally shape it, and currently moving too slowly to matter. Governance gaps in AI safety and oversight are already producing measurable harm in Canadian communities. The mental health crisis I've outlined isn't abstract – Canadian youth are among those experiencing AI-induced loneliness, pathological attachment, and existential anxiety in documented and growing numbers. Yet Canada has no binding federal AI legislation. Bill C-27's Artificial Intelligence and Data Act remains stalled, leaving Canadians exposed while our southern neighbour lurches between regulatory ambition and corporate capture, and while China races ahead unconstrained by democratic accountability. The concentration of AI power is particularly acute for Canada. Our most talented AI researchers – many trained at world-class institutions like the Vector Institute and Mila – are systematically recruited into American corporate laboratories where governance considerations are subordinated to competitive pressures. We are, in effect, contributing intellectual capital to a race we have little power to regulate. On human rights, Canada's multicultural fabric makes algorithmic bias and linguistic exclusion especially consequential. AI systems trained predominantly on English-language data systematically disadvantage Francophone, Indigenous, and newcomer communities – a quiet but profound equity failure. Canada's international reputation for multilateralism, our proximity to both the American AI industrial complex and our Commonwealth partners, and our tradition of evidence-based policymaking position us uniquely as honest brokers in global governance conversations. Geoffrey Hinton, Yoshua Bengio – Canadians both – carry extraordinary moral authority in these discussions. Most critically, Canada could lead by example: passing meaningful federal AI legislation, establishing a dedicated AI safety institute with real resources, and championing the binding international accord the world desperately needs. With Canada's international reputation as peacekeepers, we should be leading the global initiative for safe, transparent, and accountable AI Governance.

The AI Dialogue can play a transformative role – but only if participants are honest about what it must become rather than congratulating themselves for what it merely is. Dialogue is the first, necessary step in advancing international cooperation on AI governance. But if it's not focused and based on sound critical and ethical reasoning and commitment by all parties, then the ensuing architecture it creates will be weakened. The gap between these values is precisely where previous well-intentioned international AI initiatives have gone to die – producing communiqués nobody reads and principles nobody enforces. That said, I believe this Dialogue represents something genuinely important, for three reasons: First, legitimacy. A UN General Assembly Resolution carries moral and political weight that no industry summit or academic conference can replicate. When 193 member states are gathered, the conversation has a legitimizing authority that bilateral or regional frameworks simply cannot match. Second, agenda-setting. The Dialogue can force onto the international table issues that powerful actors would prefer to manage quietly and unilaterally – existential risk, monopolization of AI capability, the global mental health crisis, the erosion of democratic infrastructure. Naming these things in a UN context makes them harder to ignore. Third, institutional precedent. Every durable international governance architecture – like the IAEA, WTO, and Montreal Protocol – began as conversations before becoming institutions. The AI Dialogue can be the Montreal moment for artificial intelligence, provided participants treat it as a foundation rather than a destination. My specific hope is that this Dialogue produces three concrete outputs: a formal mandate to develop binding governance instruments; an independent technical body empowered to monitor AI development across jurisdictions; and explicit acknowledgment of existential risk as a legitimate subject of international law. Anything less and we've organized a very expensive conversation about a problem we've chosen not to solve.

Several existing initiatives deserve serious attention – not uncritical celebration, but genuine engagement as foundations upon which something more durable might be constructed. The EU AI Act represents the most comprehensive legislative attempt to date. Its risk-tiered framework – distinguishing unacceptable, high, and limited risk categories – provides a working template. The Dialogue should build on this architecture while addressing its notable blind spot: the near-complete absence of existential risk provisions. The IAEA model, for all its enforcement limitations, provides institutional precedent for international technical monitoring of a dual-use technology. The Dialogue should study both what the IAEA achieved and where it failed – particularly its dependence on Security Council enforcement – and design accordingly. The Bletchley Park AI Safety Summit and its successors established that major AI-developing nations can seriously consider existential risk questions. That precedent matters. The Dialogue should build on Bletchley's convening achievement while pushing beyond its non-binding outputs. UNESCO's Recommendation on the Ethics of AI, adopted by 193 member states, provides an existing ethical framework with genuine multilateral legitimacy – particularly valuable for human rights and cultural implications. The Dialogue should treat this not as sufficient but as an initial template. Canada's CIFAR Pan-Canadian AI Strategy and institutes like Vector and Mila demonstrate that national investments in safety-conscious AI research can anchor regional governance conversations. The added value the Dialogue brings is precisely what none of these initiatives individually possesses: universal legitimacy combined with the potential for binding authority. The EU Act governs one region. Bletchley was a summit. UNESCO produces recommendations. The IAEA monitors one technology domain. Only a mechanism rooted in the UN General Assembly can credibly claim to speak for humanity as a whole. That claim is either the Dialogue's greatest asset or its most elaborate pretension. Which it becomes depends entirely on the courage of those participating.

The structure of the Dialogue matters as much as its content. A conversation designed badly will produce bad outcomes regardless of the quality of ideas in the room. On that note, Governments must do more than deliver prepared statements. They need to arrive with genuine negotiating mandates – actual authority to make commitments. Dialogue without delegated authority is just performative. Civil society and affected communities – including mental health advocates, Indigenous peoples, youth organizations, disability rights groups, etc., must be structural participants, not decorative ones. Those people who stand to be harmed most by ungoverned AI are rarely the people helping with the design of its governance. That asymmetry must be corrected by intention, not goodwill. As mentioned earlier, the AI industry should participate transparently and fully – but without veto power over outcomes. Corporate self-interest has already shaped too much of the governance conversation. The Dialogue must be structurally insulated from corporate capture. Independent scientists and philosophers – not corporate-affiliated researchers – need prominent voices. The questions AI raises are not purely technical. They are deeply ethical, epistemological, and civilizational. Philosophers, ethicists, and critical thinkers belong at the table, not on the fringe. Educators deserve particular attention. The next generation will inherit whatever world this Dialogue helps create. Plenary declarations are insufficient. The Dialogue needs structured working groups with defined mandates, clear timelines, and published outputs – organized around thematic priorities, not diplomatic convenience. Critically, the Dialogue needs a permanent secretariat with independent technical capacity – not a rotating chairmanship that loses institutional memory every cycle. And it needs a sunset clause: concrete benchmarks that determine whether voluntary dialogue develops into binding treaty negotiation within a defined timeframe. Talk indefinitely and we govern nothing. Commit to a deadline and we might actually govern something.

The uncomfortable truth is that global AI governance conversations currently reflect the priorities of a remarkably narrow slice of humanity's domographics – wealthy, technologically sophisticated, predominantly male, and overwhelmingly from the Global North. That's simply how the global tech industry evolved. Nonetheless, it's a structural problem requiring structural remedies. The most significantly underrepresented voices today include, but are not limited to: Indigenous communities worldwide possess sophisticated, millennia-tested frameworks for thinking about intergenerational responsibility, collective decision-making, and humanity's relationship with forces larger than itself. These are precisely the conceptual resources we need for AI governance – yet Indigenous peoples remain almost entirely absent from these conversations. The Global South broadly – Africa, Latin America, South and Southeast Asia – will bear disproportionate consequences from ungoverned AI while having contributed least to its development. Governance frameworks designed without their meaningful participation will fail them systematically and predictably. Young people, particularly adolescents, are the primary victims of AI's documented mental health harms and will live longest with whatever governance architecture we construct. Yet they have virtually no formal voice in these proceedings. Youth delegates with genuine decision-making participation – not inspirational speeches followed by exclusion from negotiations – are essential. Mental health professionals and patients who are witnessing AI-associated psychosis, pathological attachment, and existential anxiety in clinical settings possess frontline knowledge that policymakers urgently need. Philosophers, ethicists, and humanities scholars are systematically marginalized in favour of engineers and economists – a disciplinary bias that produces governance frameworks technically sophisticated but morally impoverished. To remedy underrepresented communities incorporate the following: Dedicated seats, not consultations. Funded participation, not unfunded invitations. Translation infrastructure that goes beyond English. Deliberate pre-Dialogue capacity building in underrepresented regions. Inclusion isn't charity. It's epistemically and ethically necessary. The wisest room is the most diverse room.

The standard UN format – prepared statements delivered to a half-receptive room – will not produce the quality of thinking this moment in human history demands. We need formats that generate genuine intellectual friction, not diplomatic comfort. Under Chatham House Rule, we need structured adversarial debate. Not panels where everyone agrees – but formally structured encounters where proponents of accelerated AI development defend their positions against rigorous, prepared critics. The adversarial method isn't one of hostility; it's simply the most reliable epistemic mechanism humanity has developed for stress-testing ideas. I've spent my career teaching critical thinking and ethical reasoning precisely because comfortable consensus is where bad ideas survive longest. We need to introduce Red Team exercises i.e. assemble expert teams whose explicit mandate is to identify how proposed governance frameworks fail – how they get captured, circumvented, or rendered obsolete by technological change. Governance designed without adversarial stress-testing is governance designed to fail politely. Citizen deliberative assemblies running parallel to official proceedings – randomly selected, demographically representative, supported by expert testimony, and empowered to produce formal recommendations that negotiators must publicly respond to. Ireland's Citizens' Assembly demonstrated that ordinary people, given adequate information and structured deliberation time, produce remarkably sophisticated policy recommendations. Scenario immersion workshops – structured exercises where delegations collectively navigate detailed, realistic futures: What do you do when an AI system causes a mass casualty event? When a corporation deploys an AGI unilaterally? When the first AI psychosis epidemic hits a major city? Abstract governance principles become concrete – and contestable – when attached to specific scenarios. Intergenerational panels pairing senior policymakers directly with those most affected by AI's consequences – not for inspiration, but for structured accountability. Finally, public live-streaming Town Hall meetings with structured civil society commentary – making the Dialogue genuinely observable and responsive to the world it claims to represent.

Several concrete examples deserve serious attention - not as perfect models, but as evidence that effective governance is achievable when political will exists. The Montreal Protocol remains the gold standard for international governance of a civilizational-scale global risk. It demonstrates that binding, verifiable, universally adopted agreements are politically achievable - even when powerful economic interests resist them. Its lesson for AI governance is simple: ambition works when consequences for non-compliance are real. The EU AI Act - despite my earlier criticism of its existential risk blind spot - provides the most operationally sophisticated risk-tiered regulatory framework currently in existence. Its prohibition of cognitive behavioral manipulation, social scoring, and real-time biometric identification represents genuine governance with genuine teeth. Many of its recommendations should be studied, critiqued, and globally adapted. The GDPR demonstrated that strong data protection regulation, once dismissed as economically ruinous, actually elevated global standards as corporations found it operationally simpler to comply universally than maintain jurisdiction-specific practices. AI safety standards could follow the same trajectory. Finland's national AI literacy programme - which has trained over one percent of its population in AI fundamentals - demonstrates that public education at scale is achievable and essential. And the US Department of Labor has implemented the 'Make America AI-Ready' initiative to increase AI literacy across all demographics entirely by cell phone. New Zealand's Wellbeing Budget framework offers a model for embedding non-economic values - including mental health and intergenerational equity - into policy evaluation. Applied to AI governance, this framework could operationalize human wellbeing as a primary metric rather than an afterthought. Canada's Algorithmic Impact Assessment tool provides a working template for pre-deployment evaluation of AI systems in public sector contexts. And my own proposed Constitutional Accord on the Global Risk of Artificial Intelligence - developed over decades - offers a comprehensive binding framework addressing alignment, control, containment, and consequences for non-compliance.