University of Toronto Faculty of Law

Procedurally, success means demonstrating that multilateral AI governance is not merely a legitimating exercise for decisions already made elsewhere. As someone who trained in the Philippines and now works in Canada, I am acutely aware of how international legal frameworks can reproduce rather than remedy structural inequalities. The Dialogue must go beyond seating Global South states at the table and ensure their priorities — digital infrastructure deficits, linguistic and cultural marginalization, asymmetric data extraction — actually shape the agenda. Substantively, I would identify three markers of genuine progress: First, a shared diagnostic framework. Not premature consensus on solutions, but common language for naming harms. The current proliferation of national and regional AI governance regimes is, in part, a proliferation of incommensurable concepts. Shared vocabulary is a precondition for meaningful interoperability. Second, serious engagement with power asymmetries. AI governance risks replicating the pattern of international law in general - frameworks negotiated by powerful states whose terms others must absorb. A successful Dialogue would surface the ways AI development is concentrating economic and epistemic power. Third, conceptual ambition about what governance must address. Safety and interoperability are necessary but insufficient if frameworks ignore how AI systems actively generate new social norms, legal categories, and distributions of authority — not merely reflect existing ones.

Social, economic, ethical, cultural, linguistic and technical implications of AI;Interoperability of governance approaches;Protection and promotion of human rights;Open-source software, open data and open AI models;

First, protection and promotion of human rights is my primary commitment, and the lens through which I approach the others. My scholarship has increasingly focused on how AI systems do not merely implicate human rights in isolated instances, but reconfigure the conditions under which rights are exercised - reshaping dignity, equality, and epistemic autonomy at scale. Human rights should not be one thematic area among several but should be considered as the main framework within which all other areas are evaluated. Second, the social, economic, ethical, cultural, and linguistic implications of AI are where that rights commitment becomes concrete. The communities most exposed to AI's extractive and homogenizing tendencies are rarely those whose interests drive system design. The inclusion of cultural and linguistic dimensions is particularly significant and must not be treated as simply an appendix to the "real" economic and technical concerns. Third, open-source and open data matters because access to AI's technical infrastructure is a precondition for meaningful global governance and global participation, not merely an innovation policy question. Countries that cannot audit, adapt, or deploy AI systems independently will be rule-followers, not rule-makers, regardless of what the Dialogue formally decides. Finally, interoperability of governance approaches is important but we should also be careful because deciding which frameworks are compatible is simultaneously deciding which values prevail. This matters especially for countries in the Global South that are still building their governance infrastructure. If interoperability standards crystallize before those countries have developed and tested their own frameworks, they will be asked to harmonize with a baseline they had no role in setting.

Speaking from a position that straddles two jurisdictions and as an academic, I see distinct but interconnected patterns of impact. In Canada, the governance gap is less about absence than about incoherence. Canada has some meaningful institutional capacity but these existing frameworks were designed for a different technological moment and are struggling to keep pace. The more pressing challenge is that Canada sits in an asymmetric relationship with the United States which has actively retreated from multilateral governance commitments. Canadian regulators face real pressure to maintain interoperability with a US market that is deregulating not through formal agreement but through economic gravity. The opportunity here is for Canada to articulate a distinct governance identity, as it has done in areas like cultural policy and data privacy, rather than defaulting to de facto harmonization with the US. In the Philippines and the broader Southeast Asian region, the challenges are more foundational. Regulatory frameworks exist but enforcement capacity, technical expertise, and institutional independence remain constrained. The more significant challenge is structural. For example, the Philippines is a major site of AI's extractive economy. Content moderation labor, data annotation, and training data generation are disproportionately performed by Filipino workers under conditions that are poorly regulated, poorly compensated, and largely invisible to the governance frameworks being developed in the jurisdictions that benefit from that labor. The Global Dialogue's thematic areas do not adequately name this form of AI-related harm. It sits uncomfortably between labor rights, data governance, and platform regulation and falling through the gaps of each.

In my view, the AI Dialogue's most significant potential contribution is creating the conditions under which meaningful international cooperation becomes possible at all. First, it can serve as a norm-clarification forum. The current landscape of AI governance is characterized by proliferating frameworks that use similar language — safety, trustworthiness, human-centered AI — to mean substantially different things. The Dialogue can do the patient conceptual work of surfacing those differences honestly, which is a prerequisite for any genuine convergence. Second, it can institutionalize Global South voice in governance agenda-setting. Most existing AI governance frameworks such as the Hiroshima Process or the OECD Principles were developed in forums where Global South states were absent or peripheral. The Dialogue's universal membership, combined with meaningful participation support, offers a structural corrective. Third, it can establish the human rights framework as the baseline, not one option among several. International human rights law already provides a comprehensive normative vocabulary such as dignity, equality, non-discrimination, participation that applies to AI's most significant harms. The Dialogue can consolidate the emerging consensus that AI governance is not a sui generis field requiring entirely new norms, but an application domain for obligations states have already undertaken.

Emerging regional frameworks such as the African Union's AI continental strategy, ASEAN's AI governance framework, and emerging initiatives in Latin America represent pluralistic governance knowledge that is systematically undervalued in global conversations. The Dialogue should elevate them and cast a more prominent light on them as sources of genuinely different regulatory imagination beyond the usual plethora of soft law we already have (EU AI Act, OECD principles, etc.) The Dialogue's added value is its capacity to hold all of these existing frameworks in relationship simultaneously and create a forum where those differences are negotiated transparently, with every country present and the human rights framework as common ground.

First of all, I would resist the conventional tripartite model of states, industry, and civil society as if these are equivalent categories. Industry actors bring important technical knowledge and resources, but also direct financial interests in governance outcomes The academic and scientific community needs a distinct and structured role, not absorption into the civil society category. The relationship between the Dialogue and the Independent Scientific Panel should also be operationalized concretely: Panel findings should arrive in advance of sessions with sufficient time for deliberation, not presented ceremonially. Second, the Geneva session should combine smaller thematic working groups with plenary sessions that consolidate rather than initiate. Working groups should be structured around problems rather than themes, with explicit mandates to produce actionable recommendations rather than descriptive surveys. Finally, travel support is necessary but insufficient. The Dialogue should invest in regional preparatory processes (I'm thinking of Southeast Asia or Latin America) that allow regional positions to be developed collectively before Geneva, so that Global South voices arrive as interlocutors with consolidated perspectives rather than as individual national delegations negotiating in isolation against far better-resourced counterparts.

Ironically, it is the communities most exposed to AI's harms who are least present in governance conversations. This includes content moderators and data annotators in the Global South whose invisible labor makes AI systems function; indigenous communities whose lands, images, languages, and knowledge systems are extracted into training data without consent or compensation; and persons with disabilities, who experience both AI's greatest accessibility potential and its most acute discriminatory failures. Another community or group not represented in these global discussions are non-English or non-Western epistemic traditions.

I have three suggestions. One, problem-centered working groups with drafting mandates. Instead of thematic discussions that produce summaries, working groups should be assigned specific governance problems, e.g. cross-border algorithmic harm, data rights for linguistically marginalized communities, etc.— and tasked with producing draft frameworks. Deliberation anchored to concrete drafting tasks generates different and more useful outputs than open-ended exchange. Two, feature sessions where the governance challenges can be presented to industry, academic, civil society experts by those who are authoritative voices in a specific space, e.g. a Kenyan or Filipino labor advocate. The industry experts or OECD regulators or technical researchers can then take a position as questioners/learners rather than explainers. And third, as I mentioned previously, pre-session regional deliberative processes feeding consolidated positions into the Dialogue. Structured citizen consultations organized regionally would ensure that the Dialogue reflects more than the views of whichever stakeholders can afford to attend in Geneva.

One example that I have come across are algorithmic impact assessments, as developed by organizations like the Algorithmic Justice League and theorized in the academic literature, offer a procedural mechanism that is both transferable and adaptable. Requiring prospective analysis of who bears the risks of a system before deployment, with public accountability for that analysis - can be calibrated to different institutional contexts without requiring uniform substantive standards. I would also highly suggest the crafting and adoption of an International AI Bill of Rights in order to set the baseline for norm expectations using human rights language, catalyze complementary local/national regulatory efforts, and provide a symbolic list for protecting humanity in the age of AI. Such a document can only be deemed legitimate if it comes from an internationally-convened body such as the Dialogue. I'm happy to participate in the drafting of this and I've also suggested to Canadian regulators to have adopt a similar Charter of Rights and Freedoms for the age of AI in the same spirit as this one.