Humber College

• I believe it should produce a shared baseline of principles that main stakeholders can publicly endorse. These principles should clearly address safety, transparency, accountability, and human rights, reducing fragmentation across the national frameworks. Also, the inclusion of diverse economic and cultural perspectives would help ensure policies are globally relevant and not just exported standards. The dialogue should result actionable commitments regarding the agreement on information-sharing mechanisms for AI risks, pilot frameworks for auditing advanced systems, or timelines for interoperable regulation. There is a need for roadmap for future meetings, working groups, or a coordinating body to manage governance efforts to continue evolving alongside the technology. The goal is to shift to proactive coordination. If the delegates leave with a roadmap for a permanent monitoring body, the event will be a historic win. The overall successful Global Dialogue on AI Governance is to deliver outcomes that move the world from fragmented conversations to coordinated, actionable progress. The pressing issues to address are data‑sharing frameworks for safety research
• early warning and incident‑reporting channels
• agreement on red lines for high‑risk AI uses and support for capacity‑building in low‑ and middle‑income countries. Such kinds of mechanisms that turn principles into practice.

• Safe, secure and trustworthy AI
• AI capacity-building
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Interoperability of governance approaches

These areas need urgent action and active engagement because AI is already shaping economies, societies, and decision-making at scale. Delays in addressing these issues can lock in harms that are hard to reverse. Here's why each one matters right now: 1. Safe, secure, and trustworthy AI: AI systems are increasingly used in high-stakes domains. If they are unsafe or unreliable, the consequences can be serious for e.g. Bias and discrimination can reinforce inequality; security risks (like data leaks or adversarial attacks) can be exploited, and the loss of public trust can slow beneficial adoption. 2. AI capacity-building: There's a growing global gap between those who can develop and use AI effectively and those who cannot. This will cause countries, organizations, and workers without AI skills risk being left behind; the over-reliance on a few tech leaders concentrates power. It is essential that the active investment in education, infrastructure, and skills ensures more equitable access and prevents a widening digital divide. 3. Social, economic, ethical, cultural, linguistic, and technical implications: AI is affecting the jobs and economies since automation is reshaping labor markets. The AI ethics raises concerns about the decisions made by AI fairness, accountability, & transparency. For the culture & language: Many AI systems underrepresent non-dominant languages and cultures, risking exclusion or homogenization. 4. Interoperability of governance approaches: We see that different countries and organizations are creating their own AI rules and frameworks. Without coordination, this leads to fragmentation, regulatory gaps, or conflicting standards. Interoperable governance ensures that rules align enough to be effective globally while respecting local contexts. Generally speaking: AI is advancing faster than the systems designed to guide it. Without urgent and coordinated action, then the risks scale quickly, the inequalities deepen, and the governance struggles to catch up.

Issues regarding the open-source software, open data and open AI models: Many widely used open-source projects rely on underfunded maintainers. As adoption grows, especially with AI models, there is a mismatch between commercial value extraction and contributor compensation, raising concerns about long-term viability and security. The open data and AI training datasets often include sensitive or scraped information. Questions around consent, ownership, and ethical reuse are intensifying, especially with regulations like GDPR influencing what can realistically be "open". The large organizations (e.g., OpenAI, Meta, Google) dominate compute resources, talent, and infrastructure which creates a paradox where openness exists, so the meaningful participation remains unequal. Even when the AI code or models are open, the missing documentation, datasets, or compute environments can limit true reproducibility which impact negatively the scientific research and public trust.

For Academia environments: • The absence of harmonized global frameworks for AI governance (e.g., differing interpretations of risk, safety, and transparency) creates uncertainty for universities. Researchers must navigate overlapping or unclear rules on data use, model release, and security, slowing collaboration and increasing administrative overhead, especially in cross-border projects. • The security concerns (e.g., dual-use risks in generative AI or cybersecurity tools) are leading to tighter controls on what can be published or shared. This challenges long-standing academic norms of openness, particularly in fields linked to AI safety and advanced models. • AI capacity building is uneven. Elite institutions—often partnered with major tech firms like Google, Microsoft, and NVIDIA—have access to large-scale compute and proprietary datasets, while smaller institutions face barriers. This widens research inequality and shapes whose questions get studied. • High salaries and resources in industry continue to draw top AI researchers away from academia. Governance gaps around public-interest research funding and career pathways exacerbate this, weakening universities' ability to lead independent, critical inquiry. • Universities are increasingly targets for cyberattacks, data breaches, and intellectual property theft. Weak or inconsistent security governance can expose sensitive datasets (including health or personal data), undermining trust and compliance. • While demand for AI expertise is surging, governance-related skills—such as AI ethics, auditing, and secure system design—are not yet systematically embedded in curricula. This creates a lag between technological capability and responsible deployment.

The AI Dialogue can play a pivotal convening and bridging role in advancing international cooperation on AI governance by addressing fragmentation and fostering trust among stakeholders. Many actions needed: • Creating a neutral, multi-stakeholder platform which can bring together governments, academia, civil society, and industry to align on shared principles. In a landscape shaped by differing approaches (such as the EU's regulatory model), the Dialogue can facilitate mutual understanding and reduce policy divergence. • Promoting interoperability of governance frameworks: Rather than pushing a single global standard, the Dialogue can help identify areas of convergence across frameworks (e.g., risk classification, transparency requirements, safety testing). This supports regulatory interoperability, making it easier for AI systems to operate across jurisdictions. • Advancing responsible openness: The Dialogue can help clarify norms around open-source AI, open data, and model sharing—balancing innovation with safeguards. • Supporting capacity building and inclusion: This is a key role is amplifying the voices of underrepresented regions and supporting underrepresented participation. By facilitating knowledge transfer, funding partnerships, and technical training, the Dialogue can help reduce global disparities in AI capacity and governance readiness. • Building trust through transparency and confidence-building measures: The Dialogue can encourage voluntary disclosures (e.g., model capabilities, risks, and evaluation results) and promote best practices for auditing and accountability. These measures can reduce mistrust • Enabling rapid response to emerging risks: The Dialogue can act as an agile forum to surface emerging risks (e.g., misuse, systemic bias, security threats) and coordinate timely, non-binding responses or guidelines.

• The AI Dialogue should build on existing global, regional, and multi-stakeholder efforts to avoid duplication and accelerate impact. Key initiatives include the OECD AI Principles and AI Policy Observatory, UNESCO's Recommendation on the Ethics of AI, and the UN system's work through bodies such as the ITU and UNESCO. It should also connect with regional frameworks (e.g., the EU AI Act), industry-led standards (e.g., ISO/IEC AI standards), and multi-stakeholder forums like the Internet Governance Forum (IGF). Public–private partnerships, academic research networks, and civil society coalitions working on AI accountability, safety, and inclusion are also critical anchors. • The added value of an AI Dialogue lies in its ability to act as a bridge across these fragmented efforts. It can foster interoperability by aligning principles, standards, and regulatory approaches across jurisdictions. Also, it can provide an inclusive platform that amplifies voices from SMEs, and the underrepresented communities that are often missing from technical and policy discussions. The AI dialogue can facilitate practical cooperation by translating high-level principles into actionable guidance, pilot projects, and shared best practices. • The Dialogue can serve as an early-warning and horizon-scanning mechanism, identifying emerging risks and opportunities in AI development. By convening policymakers, industry leaders, researchers, and civil society, it can build trust, reduce policy silos, and encourage responsible innovation. Ultimately, its value lies not in replacing existing initiatives, but in connecting them to create coherence, filling governance gaps, and accelerating collective progress toward safe, inclusive, and human-centric AI.

1. Different stakeholders can contribute to the AI Dialogue by bringing complementary expertise, perspectives, and accountability mechanisms: • Governments can share policy experiences, align regulatory approaches, and commit to interoperable frameworks. • Private sector actors can provide technical insights, disclose best practices, and pilot responsible AI solutions. • Academia and researchers can contribute independent evidence, foresight analysis, and evaluation methodologies. • Civil society and affected communities can highlight real-world impacts, human rights considerations, and inclusion gaps. • International organizations can help coordinate efforts, ensure global representation, and link outcomes to existing frameworks. 2. Recommendations for format and structure: • Multi-tiered structure: Combine high-level plenaries (for strategic alignment) with thematic working groups (e.g., safety, governance, inclusion, innovation). Working groups should produce concrete outputs such as guidelines, toolkits, or policy briefs. • Hybrid and inclusive participation: Ensure both in-person and virtual engagement, with funding or support mechanisms to include participants from underrepresented regions. • Continuous, not one-off: Establish the Dialogue as an ongoing process with regular convenings, iterative feedback loops, and clear milestones. • Action-oriented outputs: Move beyond discussion by incorporating pilot projects, voluntary commitments, and measurable outcomes. • Transparency and accountability: Publish summaries, recommendations, and progress reports; allow public input or consultation phases. • Interoperability focus: Design sessions that explicitly map and connect existing initiatives to avoid duplication and promote alignment. • Light governance model: Use a rotating secretariat or steering group representing diverse stakeholders to maintain neutrality and adaptability.

1. Several voices are still underrepresented: • Global South stakeholders, particularly from low- and middle-income countries, who face distinct challenges around infrastructure, data governance, and economic dependency. • Small and medium-sized enterprises (SMEs) and startups, which often lack the capacity to engage in policy forums despite being key innovators and adopters. • Workers and labor organizations, especially those in sectors most affected by automation and algorithmic management. • Marginalized and historically excluded communities (e.g., Indigenous peoples, racialized groups, persons with disabilities), who are disproportionately impacted by biased or poorly designed AI systems. • Youth voices, who will live with long-term consequences of AI deployment. • Non-technical disciplines (e.g., social sciences, humanities), which are critical for understanding societal impacts but are often overshadowed by technical perspectives. 2. How to include them: • Targeted outreach and funding: Provide travel grants, stipends, and capacity-building programs to enable meaningful participation from underrepresented groups. • Decentralized engagement: Host regional and local dialogues and feed their outcomes into global processes to ensure context-specific insights are reflected. • Accessible formats: Use multilingual, hybrid, and non-technical formats to lower barriers to participation. • Institutionalized representation: Allocate dedicated seats or quotas in working groups, steering committees, and advisory bodies. • Partnerships with trusted intermediaries: Collaborate with local organizations, community leaders, and civil society networks to surface grounded perspectives. • Feedback and accountability loops: Ensure participants can see how their input shapes outcomes, building trust and sustained engagement.

To foster meaningful and dynamic engagement, the AI Dialogue should move beyond traditional panels and adopt more interactive, participatory formats: • Co-creation labs: Small, diverse groups collaborate intensively over short periods to develop concrete outputs for e.g. model policies, governance frameworks, or technical guidelines—encouraging hands-on problem solving. • Scenario-based simulations: Participants engage in role-play exercises around realistic AI governance challenges (e.g., responding to a major AI system failure or cross-border regulatory conflict). This helps surface trade-offs, test coordination mechanisms, and build shared understanding. • Deliberative citizen assemblies: Randomly selected, demographically representative participants learn about AI issues, deliberate, and produce recommendations. This brings informed public perspectives into high-level discussions. • Stakeholder takeovers: Instead of experts speaking, affected communities, workers, or youth lead sessions, with policymakers and industry representatives listening and responding. • Interactive foresight workshops: Use horizon scanning and futures-thinking methods to explore long-term risks and opportunities, helping participants align on proactive governance approaches. • Challenge-driven hackathons: Multidisciplinary teams (policy, technical, legal) work together on predefined challenges, producing prototypes, policy tools, or risk assessment methods. • Digital collaboration platforms: Ongoing online workspaces enable asynchronous input, document co-drafting, and broader participation beyond live events. • Real-time polling and feedback loops: Use live surveys and iterative feedback during sessions to capture diverse views and adjust discussions dynamically. Combining these formats in a modular way and linking interactive sessions to concrete outputs will ensures the Dialogue remains inclusive, action-oriented, and responsive, while maintaining participant engagement and producing tangible results.

Several policies, practices, platforms, and approaches illustrate how AI governance can be made effective and actionable: • Regulatory frameworks and guidelines: The EU AI Act establishes a risk-based classification system for AI, mandating transparency, safety, and human oversight for high-risk applications. Similarly, UNESCO's Recommendation on the Ethics of AI provides globally recognized principles on fairness, accountability, and inclusivity. • Standards and technical specifications: ISO/IEC JTC 1/SC 42 develops international standards for AI, covering risk management, robustness, and trustworthiness. Industry-led initiatives like IEEE's Ethically Aligned Design offer guidance for embedding ethical considerations into AI systems. • Multi-stakeholder platforms: The Global Partnership on AI (GPAI) fosters collaboration among governments, industry, and academia to address responsible AI development, safety, and innovation. Similarly, AI4People and the Partnership on AI facilitate research sharing and ethical best practices across sectors. • Auditing and accountability mechanisms: Practices such as algorithmic impact assessments and independent auditing of AI systems promote transparency and accountability, particularly in sensitive areas like hiring, criminal justice, or healthcare. • Public participation and inclusivity approaches: Deliberative processes, citizen assemblies, and participatory design workshops-used in countries like Finland's AI Strategy engagement process-ensure diverse perspectives inform AI governance. • Open-source and collaborative tools: Platforms like AI Incident Database and OpenAI's Model Card approach provide practical resources for risk reporting, monitoring, and documenting AI system capabilities and limitations. • Cross-border cooperation mechanisms: Initiatives such as the OECD AI Policy Observatory support harmonization of AI policies, offering benchmarks and guidance for responsible deployment while fostering international alignment. Collectively, these examples demonstrate a combination of regulation, standards, accountability, multi-stakeholder collaboration, and inclusive participation. They highlight the importance of balancing technical rigor with ethical principles, public engagement, and global coordination to address AI's complex challenges effectively.