Human Decisions, Global Consultancy, Berlin

It will be critical for the UN led Global Dialogue to articulate its added value vis-a-vis industry and member state led initiatives. Specifically, global inclusion and giving voice to all citizens of the world is unique. It will be important to continue this broad engagement and ensure equal representation of voices across member states, citizenship and the industry.

• Safe, secure and trustworthy AI
• Transparency, accountability, and human oversight
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• AI capacity-building

These priorities reflect the need to approach AI governance not only as a matter of technological regulation, but as a question of how institutions, societies and economies make decisions in AI-enabled environments. The critical issue is not only whether AI systems are technically safe or efficient, but whether their use strengthens or weakens human judgement, public accountability, institutional legitimacy and equitable socio-economic outcomes. Transparency, accountability and human oversight are urgent because AI systems increasingly influence decisions affecting people's rights, opportunities, mobility, employment, access to services and public trust. Human oversight must be meaningful, not merely procedural. Decision-makers need the authority, literacy and institutional mandate to understand, question, override or contest AI-supported outputs. The social and economic implications of AI require particular attention, especially for labour markets, skills systems, public administration, migration governance and development policy. AI can enhance productivity and evidence-based decision-making, but it can also deepen inequalities, transform or displace work, and create new asymmetries between those who design systems and those affected by them. AI capacity-building is essential to avoid a governance divide in which some countries and institutions become rule-makers and system-builders while others remain passive adopters. Interoperability of governance approaches is also necessary so that global, regional and national frameworks can support innovation while preserving rights, accountability and institutional coherence.

A key emerging issue is the changing relationship between human and machine decision-making. Current AI governance debates often focus on model safety, data protection, transparency or innovation, but pay less attention to how AI changes institutional judgement itself. In practice, AI systems increasingly shape what decision-makers see, which options appear legitimate, which risks are prioritised, and which people or groups become more visible or invisible to institutions. This is particularly urgent in labour markets. AI is already affecting recruitment, skills matching, workforce planning, productivity measurement, public employment services, migration pathways, education-to-work transitions and social protection. These systems may improve efficiency, but they can also reproduce bias, misclassify skills, obscure informal or care work, undervalue migrant experience, and create new forms of exclusion for workers who do not fit standardised data profiles. The issue is not only job displacement. It is also the redistribution of decision-making power across employers, platforms, governments, workers and automated systems. A second concern is the risk of "false delegation". Institutions may formally retain human oversight while, in reality, human actors defer to AI outputs because of time pressure, limited technical literacy, lack of contestability mechanisms, or perceived machine objectivity. This can weaken accountability while preserving the appearance of responsible governance. The Global Dialogue should therefore address not only AI systems, but AI-enabled decision environments. This includes questions such as: which decisions should never be automated; when AI should only assist; how uncertainty should be communicated; who can challenge or override outputs; how affected persons can seek explanation or remedy; and how public institutions can build the human, legal and organisational capacity to remain accountable. The central governance challenge is to ensure that AI strengthens human capability, institutional responsibility and equitable development, rather than replacing judgement with automated authority.

From the perspective of a new Berlin-based initiative working globally on human-centred decision-making, governance and institutional capacity, the most significant governance gaps are currently visible in the transition from AI principles and regulation to practical implementation. In the EU, AI governance is becoming more structured and increasingly human-rights-oriented. This is an important and positive direction, and should remain a global priority. At the same time, many public institutions, smaller organisations and cross-border actors still lack the practical capacity to translate emerging rules into accountable decision-making systems. The key challenge is not only whether AI systems are technically compliant, but whether organisations know how to use them responsibly in real institutional settings. Transparency, human oversight and accountability require more than documentation. They require people with the authority, skills and confidence to understand AI-supported outputs, question them, override them where necessary, and explain decisions to affected persons. The social and economic implications are particularly significant in Europe's labour markets. AI is already influencing recruitment, skills matching, productivity analysis, public employment services, education and workforce planning. These developments create opportunities for better evidence, more personalised services and stronger anticipation of skills needs. However, they also create risks of bias, exclusion, over-standardisation of human capabilities, and reduced recognition of migrant, informal or non-linear work experience. For the EU and its partners, the opportunity is to make human-centred AI governance operational: connecting regulation with capacity-building, institutional design, labour-market policy and international cooperation. A Berlin-based initiative with a global outlook can contribute by supporting governments, international organisations and partners to strengthen human oversight, assess decision risks, build AI literacy, and ensure that AI adoption improves institutional responsibility and equitable development rather than weakening human judgement.

The AI Dialogue can play an important role as a trusted global platform for moving AI governance from principles to practical cooperation. Its added value should be to connect normative guidance, regulatory learning, institutional capacity-building and evidence on social and economic impacts across regions and sectors. First, the Dialogue can help build a shared understanding of what human-centred AI governance means in practice. This includes transparency, accountability, meaningful human oversight, human rights protection, and mechanisms for contestability and redress. These concepts need to be translated into operational guidance that public institutions, employers, service providers and communities can actually apply. Second, the Dialogue can support interoperability between different governance approaches. The EU, Council of Europe, OECD, UNESCO, G7, African Union and other actors are developing complementary frameworks. The Dialogue can help identify areas of convergence, reduce fragmentation and support countries that need to align national approaches with international standards without simply importing models that may not fit local realities. Third, the Dialogue should strengthen AI capacity-building, especially for countries and institutions that risk becoming passive adopters of technologies developed elsewhere. Capacity-building should include not only technical skills, but also institutional design, public procurement, data governance, labour-market analysis, impact assessment and human oversight. Finally, the Dialogue can provide a space to examine emerging impacts, including on work, skills, migration, public services and inequalities. It should elevate evidence from diverse regions and affected groups, not only from technology providers or advanced economies. Its core contribution should be to ensure that AI governance remains cooperative, inclusive and implementation-oriented, while keeping human judgement, accountability and equitable development at the centre of technological change.

The AI Dialogue should build upon existing initiatives, partnerships and mechanisms rather than create a parallel governance space. Relevant foundations include UNESCO's Recommendation on the Ethics of Artificial Intelligence, the OECD AI Principles, the G7 Hiroshima AI Process, the EU AI Act, the Council of Europe Framework Convention on AI, and regional approaches such as the African Union Continental AI Strategy. It should also connect with sectoral and development-focused mechanisms addressing digital public infrastructure, data governance, labour markets, education, migration, human rights, social protection and sustainable development. The added value of the AI Dialogue should be threefold. First, it can provide an inclusive bridge between existing frameworks. Many initiatives are valuable but fragmented by region, sector, institutional mandate or level of technical maturity. The Dialogue can identify areas of convergence, clarify where approaches differ, and support interoperability without imposing a single model. Second, it can connect governance principles with implementation. A major gap is not the absence of high-level principles, but the lack of institutional capacity to operationalise transparency, accountability, human oversight, risk assessment and redress in real decision-making environments. Third, the Dialogue can elevate perspectives that are often underrepresented in AI governance: developing countries, smaller public administrations, workers, migrants, affected communities, civil society and institutions that adopt AI systems without having shaped their design. Its distinctive contribution should be to ensure that AI governance remains global, human-centred and development-oriented. It should help translate existing standards into practical capacity-building, shared learning, evidence on impacts, and cooperation mechanisms that support responsible AI adoption while protecting human rights, institutional accountability and equitable socio-economic development.

Different stakeholders should contribute to the AI Dialogue according to their respective mandates, expertise and lived experience. Governments can contribute by sharing regulatory approaches, implementation challenges, public-sector use cases, capacity-building needs and lessons from national AI strategies. They should also identify where international cooperation is needed to address cross-border risks, interoperability and unequal access to AI capabilities. International organizations can help connect AI governance with human rights, sustainable development, labour markets, migration, education, public administration and digital cooperation. They can also support evidence generation, technical assistance and coordination across existing normative frameworks. Academic and scientific institutions can provide independent analysis on AI capabilities, limitations, risks, uncertainty, bias, human-machine interaction and long-term social impacts. Their role is important in ensuring that governance discussions remain grounded in evidence rather than technological optimism or fear. Civil society, workers' organizations, migrant groups, youth, affected communities and human rights actors should have a central role in identifying risks that may not be visible from institutional or technical perspectives. Their participation is essential for legitimacy, accountability and meaningful oversight. The private sector, including technology developers and deployers, should contribute transparency on system design, deployment contexts, risk management, data practices and operational lessons. However, the Dialogue should ensure that commercial perspectives do not dominate the agenda. The AI Dialogue should combine several formats: high-level plenary sessions for political direction; thematic working groups for implementation-focused discussion; regional consultations to reflect different governance contexts; and sector-specific tracks on labour markets, public services, education, migration, health, social protection and digital public infrastructure. It should also include practical case clinics where stakeholders examine real AI-enabled decision-making systems and identify lessons on transparency, accountability, human oversight and impact assessment. Outputs should be concrete: policy briefs, implementation guidance, capacity-building tools, model governance checklists, and periodic synthesis reports. The structure should remain inclusive, evidence-based and action-oriented, with clear channels for underrepresented groups to shape the agenda rather than merely comment on pre-defined priorities.

Several voices remain underrepresented in global discussions on AI governance, particularly those most affected by AI-enabled decisions but least able to influence system design, procurement or regulation. First, workers and jobseekers are insufficiently represented, although AI is increasingly used in recruitment, skills matching, productivity monitoring, platform work, workforce planning and public employment services. Their inclusion is essential to understand how AI affects access to work, bargaining power, skills recognition, income security and workplace dignity. Second, migrants, displaced persons and mobile populations are often absent from AI governance discussions, despite being affected by data-driven systems in border management, visa procedures, identity systems, humanitarian assistance, labour intermediation, social protection and integration services. Their exclusion risks reinforcing invisibility, misclassification and unequal access to rights and opportunities. Third, small public administrations, local governments and frontline service providers need stronger representation. They are often responsible for implementing AI-related systems but may lack technical capacity, procurement leverage or institutional safeguards. Fourth, developing countries and smaller economies should be included not only as recipients of capacity-building, but as contributors to governance models reflecting different legal, linguistic, cultural, economic and infrastructural contexts. Fifth, civil society organizations, trade unions, youth, women's organizations, disability advocates, linguistic minorities and communities affected by algorithmic bias should have structured channels to shape the agenda. Inclusion should go beyond invitations to speak. The AI Dialogue should establish regional and sectoral consultations, funded participation for underrepresented groups, multilingual engagement, accessible formats, and mechanisms for written submissions from grassroots and frontline actors. It should also use case-based consultations focused on real AI-enabled decisions affecting employment, mobility, access to services and rights. Most importantly, affected communities should be involved early enough to influence priorities, not only asked to validate conclusions. Meaningful inclusion requires agenda-setting power, feedback loops, transparency on how inputs are used, and continued participation in implementation and review.

The AI Dialogue should move beyond conventional panel formats and use engagement methods that support practical learning, diverse participation and problem-solving across regions and sectors. One useful format would be AI governance case clinics, where participants examine real or realistic AI-enabled decision-making systems, such as recruitment tools, public employment services, migration procedures, social protection targeting or education pathways. These clinics could identify risks, accountability gaps, human oversight needs and possible safeguards. A second format could be human oversight simulation exercises. Participants would be asked to respond to AI-supported recommendations under conditions of uncertainty, time pressure or incomplete data. This would help reveal how human judgement can be strengthened or weakened in practice, and what institutional conditions are needed for meaningful oversight. Third, the Dialogue could establish regional and sectoral listening labs involving workers, migrants, youth, local authorities, civil society, small public administrations and affected communities. These labs should be multilingual, accessible and designed to feed directly into the formal agenda. Fourth, the Dialogue could use multi-stakeholder design sprints focused on concrete outputs, such as model accountability checklists, public procurement safeguards, AI literacy modules, labour-market impact assessment tools or guidance on contestability and redress. Fifth, the Dialogue could create evidence-to-policy sessions where researchers, policymakers, technology providers and affected groups jointly review emerging evidence on AI's social and economic impacts, including labour markets, skills, mobility and inequality. Finally, the Dialogue should include a living digital consultation platform that allows stakeholders to submit use cases, governance challenges and proposed solutions between formal meetings. This would help maintain continuity, transparency and broader participation. The most effective engagement formats will be those that connect principles with practice, give underrepresented groups agenda-setting space, and produce usable outputs rather than only declarations or summaries of discussion.

Several existing policies and practices offer useful models for effective AI governance. The EU AI Act provides an important example of a risk-based legal framework that combines prohibited practices, obligations for high-risk systems, transparency duties and human oversight. Its value lies in linking innovation with fundamental-rights protection and institutional accountability. The emerging European AI Office can also support more coherent implementation across countries and sectors. UNESCO's Recommendation on the Ethics of AI and its Readiness Assessment Methodology offer a practical model for helping countries identify institutional, legal, social and technical gaps before AI systems are widely deployed. This is especially relevant for capacity-building and for countries that need context-sensitive governance support. The OECD AI Principles and OECD.AI Policy Observatory are also valuable because they combine normative principles with comparative policy evidence, helping governments learn from each other and monitor emerging governance practices. At the organisational level, the NIST AI Risk Management Framework provides a useful voluntary approach for identifying, measuring and managing AI risks across the lifecycle of AI systems. It is especially relevant for translating high-level principles into internal governance practices. More concrete tools are also needed. Canada's Directive on Automated Decision-Making and Algorithmic Impact Assessment show how public administrations can assess risk before deploying automated systems, including through structured questions on system design, impact, data and mitigation measures. Effective AI governance should build on these examples while adding stronger attention to real decision environments: who uses AI outputs, who can challenge or override them, how uncertainty is communicated, and how affected people can seek explanation, remedy and accountability.