Wouessi Inc.

A successful first AI Dialogue would deliver four concrete outcomes. First, a shared vocabulary and a baseline taxonomy of AI risks, capabilities, and governance levers that policymakers, SMEs, and civil society can all reference — reducing the conceptual fragmentation that today blocks meaningful comparison across jurisdictions. Second, a tangible commitment to capacity-building: a publicly funded mechanism or matched-funding program that supports small and medium enterprises (SMEs), public institutions, and underrepresented communities in adopting AI responsibly, with measurable participation targets for the Global South and minority-led firms. Third, the establishment of working groups with mixed stakeholder representation (government, private sector, civil society, technical community) that report back at the 2027 New York convening with draft interoperability standards on at least three priorities — model documentation, evaluation and red-teaming, and open data licensing. Fourth, a published roadmap that explicitly addresses linguistic inclusion (French, Arabic, Portuguese, indigenous languages) and that funds translation, evaluation datasets, and skills training in those languages. The Dialogue should resist the temptation to produce a single declarative document that papers over differences. Instead it should normalize structured disagreement: published positions from each stakeholder group, side-by-side comparison of approaches (EU AI Act, NIST AI RMF, Bletchley/Seoul commitments, Hiroshima Code, ASEAN guide, AU Continental AI Strategy), and a public-facing platform where SMEs and civil society can flag implementation gaps in real time. Success also means operational realism: the Dialogue should produce one or two pilots that demonstrate outcomes — for example, a Global South SME compliance toolkit, or a multilingual evaluation benchmark — rather than only normative texts. Finally, success requires that participation is not extractive: contributors from Africa, Latin America, and small island states must see their inputs reflected in the final outputs, not merely acknowledged.

• Safe, secure and trustworthy AI
• AI capacity-building
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Open-source software, open data and open AI models

We selected AI capacity-building, social/economic/cultural/linguistic implications, open-source/open data/open models, and safe/secure/trustworthy AI because these four areas determine whether AI governance is meaningful for the millions of SMEs and underrepresented communities that today sit outside the policy conversation. As a Canadian-headquartered AI consulting and AgriTech firm with operations across Canada, the United States, and Francophone Africa, we see daily how capacity gaps and language barriers translate into exclusion: a smallholder farmer in Cameroon or a francophone newcomer in Ontario rarely encounters tools, training, or standards in a form they can use. Capacity-building is the prerequisite for everything else - without funded skills programs, sandboxed compute access, and SME-focused implementation guidance, the most elegant governance frameworks remain inert. Social, economic, cultural, and linguistic implications matter because the harms and benefits of AI are not distributed evenly: high-resource languages, urban populations, and large enterprises are over-served, while frontier and emerging-market users face data scarcity and culturally misaligned models. Open-source software, open data, and open models lower the barrier to entry for Global South firms, public institutions, and civil society - they are also a structural counterweight to compute concentration and provide an essential testbed for safety research. Finally, safe, secure, and trustworthy AI underpins all of the above: governance without rigorous evaluation, red-teaming, and incident reporting becomes performative. We chose not to select interoperability or human rights as separate priorities only because we read them as cross-cutting requirements that should be embedded in each of the four areas above, not because they are less important. Wouessi's experience deploying precision agriculture and digital skills training tools in francophone markets shapes this view directly.

Three issues sit awkwardly across the listed themes and deserve explicit treatment. First, compute access and energy equity: the practical ability to fine-tune, evaluate, or even meaningfully test frontier models is concentrated in a handful of jurisdictions and firms. Without coordinated public-good compute (regional AI sandboxes, subsidized inference credits for SMEs and academic users, transparent allocation rules), governance will entrench rather than redistribute power. Second, the SME implementation gap. The AI policy conversation is overwhelmingly framed around frontier-lab obligations and very-large-platform risk, but the everyday economic reality of AI adoption is mediated by SMEs - consultancies, integrators, vertical-software vendors, and agricultural cooperatives - who lack the legal teams to interpret obligations and the engineering depth to implement them. Governance instruments should explicitly include SME compliance toolkits, simplified documentation templates, and proportionate obligations. Third, AI's effect on small-language and oral-tradition communities. Most evaluation benchmarks, copyright regimes, and training-data norms assume written, high-resource languages. Francophone, Bantu, Wolof, Lingala, and other linguistic communities are rarely covered, and indigenous knowledge systems risk being either ignored or extracted without consent. The Dialogue should commission work on linguistic data sovereignty, opt-in licensing frameworks for community knowledge, and minimum quality bars for non-English deployments. Two further emerging issues deserve mention: the convergence of AI with cross-border financial infrastructure (where remittance corridors, KYC, and credit decisioning increasingly use opaque models, with disproportionate impact on diaspora communities); and the agricultural climate-adaptation use case, where AI-enabled precision agriculture is one of the most concrete near-term opportunities for the Global South - but only if data, models, and devices are designed with farmer cooperatives, not imposed on them.

In the AI consulting and AgriTech sectors that Wouessi operates in, three governance gaps translate directly into operational and equity costs. In Canada, the absence of clear, harmonized SME guidance under AIDA (and now its successor instruments) and provincial frameworks creates compliance ambiguity for firms below the AI lab tier. We routinely see Canadian SMEs delay or shrink AI deployments not because of risk, but because of legal uncertainty about documentation, third-party model use, and procurement obligations. The opportunity is significant: a Canadian SME compliance pathway aligned with NIST AI RMF and the EU AI Act would unlock adoption in agri-food, digital training, and public-sector contexts. In our Francophone African markets — Cameroon, Gabon, Guinea — the dominant gap is foundational: limited national AI strategies, scarce sectoral guidance, and almost no regulatory capacity to oversee AI in agriculture, fintech, or public services. The result is two-tier risk: well-resourced multinationals deploy with little oversight, while local SMEs and cooperatives lack both the regulatory clarity and the capacity-building support to participate. The opportunity is large: precision agriculture, mobile health, and cross-border payments are exactly the use cases where AI can deliver measurable welfare gains, but only if governance frameworks include open data norms, language inclusion, and capacity-building budgets. Cross-cutting both regions, the open-source and open-data gap matters enormously: without baseline open evaluation datasets in French, Pidgin, and African indigenous languages, models deployed in those markets cannot be properly tested, and harms accumulate quietly. For Wouessi as a CAMSC-certified minority-owned firm, the governance gap also has a procurement dimension: minority- and Black-led tech SMEs are often invisible to public AI procurement, despite being uniquely positioned to deliver inclusive deployments.

The AI Dialogue's most valuable role is convening — not norm-setting in the first instance. Multiple normative tracks already exist (UNESCO Recommendation, OECD Principles, GPAI workstreams, G7 Hiroshima Process, Council of Europe Framework Convention, AU Continental AI Strategy, ASEAN Guide). Their proliferation is now itself a source of fragmentation. The AI Dialogue can play four convening functions that no other forum cleanly delivers. First, mapping and reconciliation: produce and maintain an authoritative public crosswalk between major frameworks, highlighting genuine divergence versus terminological mismatch. This single deliverable would be enormously useful to SMEs, regulators, and developing-country policymakers who today must reconstruct it themselves. Second, capacity-equalizing convening: structure participation so that smaller member states, SMEs, civil society from the Global South, and linguistic minorities can engage on equal terms — through translation, travel funding, asynchronous participation, and structured stakeholder seats rather than ad-hoc consultations. Third, pilot incubation: rather than producing only declarations, identify three to five practical pilot projects (e.g., multilingual evaluation benchmarks, SME compliance toolkits, regional AI sandboxes, agricultural data trusts) and shepherd them through funding, partnership, and reporting. Fourth, public legitimacy and accountability: as a UN-anchored forum the Dialogue can hold a unique mirror to existing initiatives, asking publicly whether commitments have translated into action — particularly on capacity-building, linguistic inclusion, and SME participation. The Dialogue should be cautious about substituting itself for technical bodies (ISO, NIST, IEEE) or duplicating UNESCO's normative work. Its comparative advantage is political legitimacy plus universal participation. Used well, this combination can resolve the coordination problems that bilateral and plurilateral tracks cannot — and translate cooperation into outcomes that reach SMEs and underrepresented communities, not just the largest labs and largest states.

• The AI Dialogue should connect, not duplicate. Priority touch-points include: UNESCO's Recommendation on the Ethics of AI and its Readiness Assessment Methodology — already adopted by ~70 states and a natural foundation for capacity-building work
• the OECD AI Principles and the OECD AI Policy Observatory, which house the most mature comparative dataset
• the Global Partnership on AI (GPAI) and its working groups, particularly responsible AI and data governance
• the G7 Hiroshima AI Process Code of Conduct and its reporting framework
• the Council of Europe Framework Convention on AI, which provides a binding template
• the AU Continental AI Strategy and the Smart Africa AI Blueprint, essential for African inclusion
• the ASEAN Guide on AI Governance and Ethics
• ISO/IEC JTC 1/SC 42 and NIST AI RMF for technical standards
• the AI Safety Institute network (UK, US, Canada, Japan, EU and others) for evaluations and red-teaming
• the Bletchley/Seoul/Paris AI Safety Summit commitments
• ITU AI for Good and the Partnership on AI capacity-building work
• and the Internet Governance Forum's Policy Network on AI for multistakeholder process learning. Civil society and technical-community initiatives that should be formally included are MLCommons (benchmarking), the Partnership on AI, Mozilla's open-source AI work, Hugging Face's responsible-deployment tooling, the African NLP communities (Masakhane, Lelapa AI), and the indigenous data sovereignty movement (CARE principles). The AI Dialogue's added value over these initiatives is fourfold: universal political legitimacy (193 member states)
• the ability to commission cross-initiative crosswalks
• convening power to seat under-resourced stakeholders
• and a mandate that can incubate pilots across the technical-normative divide. Its risk is duplication — which can be managed by an explicit non-duplication principle and a public dashboard showing how the Dialogue's outputs link to existing instruments.

• The AI Dialogue should adopt a structured-pluralism design: governments retain decision authority, but stakeholders are formally seated rather than consulted ad-hoc. Concretely, we recommend a four-pillar structure. Pillar one: a governmental track for member-state representatives, with the Co-Chairs convening. Pillar two: a private-sector track stratified to give voice to SMEs and emerging-market firms, not only to large platforms — for example, reserved seats for SMEs from each UN regional group, for minority- and women-led firms, and for firms based in least-developed countries. Pillar three: a civil society and rights-holder track including academia, journalism, indigenous organizations, and labor — with explicit translation and travel support. Pillar four: a technical community track including standards bodies, AI safety institutes, open-source maintainers, and benchmark organizations. Each pillar should produce its own input document for each thematic working group, published side-by-side. To avoid talk-shop dynamics, structure should be outcome-oriented: each thematic working group writes a problem statement, a draft solution, and a proposed pilot — then reports against measurable indicators. Format recommendations: open consultation periods of at least 90 days
• multilingual submission portals (UN languages plus key regional languages)
• structured templates with optional free-form sections
• commitment to publishing all written submissions
• rotating regional convenings beyond Geneva and New York to include African, Asian, and Latin American host cities for thematic sessions
• and asynchronous participation channels (recorded testimony, written input, video Q&A) so that resource-constrained stakeholders can engage without travel. The Dialogue should also adopt a transparent declaration-of-interest regime and a public submissions registry. Wouessi as an SME would value a clear, low-friction pathway for SME participation — most SME voices are absent from international AI fora today not because they decline but because the entry barrier is too high.

• Several communities are systematically underrepresented in global AI governance discussions. First, SMEs and firms from the Global South. The current conversation is structured around frontier labs and large-state regulators
• the SMEs that actually deploy AI to end-users — agricultural cooperatives, vertical-software vendors, integrators — have almost no formal seats. Second, linguistic minorities. Francophone Africa, Lusophone communities, Arabic-speaking populations outside Gulf states, indigenous-language communities, and signed-language communities are barely represented in benchmark design, evaluation work, or training-data norms. Third, smallholder farmers and rural communities, despite agriculture being one of the highest-impact AI use cases for the Global South. Fourth, minority-owned and Black-led technology firms, who bring distinct deployment context (community trust, language fluency, local distribution) but rarely sit on international AI panels. Fifth, frontline workers whose tasks AI most directly affects — care workers, smallholders, gig-economy drivers, public-service clerks — are routinely discussed about, rarely invited. Sixth, civil society organizations from least-developed countries and small island states. Seventh, indigenous data-sovereignty advocates, whose work on the CARE principles and First Nations OCAP principles deserves a structural seat. Inclusion mechanisms that work: reserved stakeholder seats by region and sector with rotating membership
• capacity-building grants to support participation (not just observation)
• translation and interpretation in working sessions, not just plenaries
• rolling open consultations with structured templates so that submissions can be received continuously
• partnership with diaspora-led firms (such as Wouessi) and minority-business certification bodies (CAMSC, NMSDC) to surface SME voices
• explicit youth and women's representation
• and outcome-based accountability — the Dialogue should publish, after each convening, who participated and how their inputs shaped outputs. Without these mechanisms, multistakeholder remains aspirational.

• Effective formats blend the deliberative, the technical, and the participatory. We suggest five. First, structured working groups with rotating chairs across stakeholder pillars, producing deliverables on a rolling 6-month cadence — modeled on IETF working groups but adapted for political legitimacy. Second, regional thematic convenings hosted in non-traditional locations (Yaoundé, Kigali, Bogotá, Suva, Dhaka) with hybrid participation, ensuring that travel cost is not a gate. Third, structured red-teaming and evaluation sprints that bring policymakers, technical experts, and civil society together to actually probe deployed systems against governance criteria — making the work concrete rather than abstract. Fourth, policy hackathons where SME developers, regulators, and rights advocates co-design compliance toolkits, evaluation harnesses, or transparency artifacts, producing reusable open-source outputs. Fifth, citizen assemblies and lived-experience panels — randomly selected or community-nominated participants from underrepresented communities, given preparation time, expert briefings, and rapporteur support — to bring frontline perspective into substantive sessions, not just opening ceremonies. Cross-cutting tools: a multilingual public submissions portal with AI-assisted translation that preserves source-language records
• a public dashboard tracking which inputs influenced which outputs
• structured deliberation platforms (Pol.is, Decidim) for asynchronous consultation
• recorded oral-tradition submission channels (audio, video) for communities whose knowledge is not primarily textual
• and youth-facing engagement tracks linked to formal outputs rather than parallel symbolic events. Avoid: opaque expert-group formats that recreate existing power asymmetries
• English-only working sessions
• observation-only seats marketed as participation
• and one-shot global summits without follow-through. Finally, embed evaluation: at each convening, publish honest data on regional, gender, and stakeholder representation, and on whether SME and civil society inputs measurably shaped outputs. Format design is governance design — process determines whose problems get solved.

We highlight five examples. First, NIST AI Risk Management Framework (US) and its companion Generative AI Profile - a voluntary, function-based framework that has become a practical reference for SMEs because it is implementable without legal counsel. Its translation into French, Spanish, and other UN languages would multiply impact. Second, the EU AI Act's tiered, risk-based obligations and its forthcoming SME guidance - useful as a model for proportional regulation, though its compliance burden on SMEs without dedicated legal teams remains a real concern that the Dialogue could help address. Third, Canada's Algorithmic Impact Assessment for federal public-sector AI deployments - a mature, publicly available tool that operationalizes accountability and could be adapted by other jurisdictions. Fourth, UNESCO's Readiness Assessment Methodology, which gives governments a structured pathway from principles to implementation, and which the AI Dialogue should fund and scale, particularly in Africa and small island states. Fifth, the AU Continental AI Strategy and Smart Africa Blueprint as a regional template that takes seriously capacity, infrastructure, and language inclusion. On the technical side: MLCommons benchmarks for evaluation transparency; Hugging Face model cards and dataset cards as practical documentation standards; the CARE principles for indigenous data governance; and Masakhane's African-language NLP work as a model of community-led research. On the operational side: Singapore's AI Verify toolkit (open-source testing and reporting); the UK and US AI Safety Institutes' published evaluations; and Mozilla Foundation's responsible-AI grant program for SMEs. Wouessi's own deployments - precision agriculture in francophone Africa, employer-sponsored digital skills training in Ontario - show that practical governance hinges on three things: usable documentation in the user's language, proportional obligations sized to the SME, and capacity-building budgets attached to every regulatory ask.