Empowering Voices: Strengthening CSO and Youth Participation in AI Governance

Emaan argues that inclusive AI governance should begin with direct documentation of harms experienced by marginalised communities and with practical pathways for remedy. Her approach is to connect community evidence, complaint handling and policy input so that minority groups are not excluded from governance and accountability processes.

She explained that her pilot project works with Pakistan’s Christian community, partnering with churches, Christian schools, community members and CSOs to document harms linked to AI-facilitated violence . She gave concrete examples of those harms, including deepfakes leading to blasphemy allegations, incitement, mob violence and arrests, and said the project would then build a referral pathway into the National Commission for Human Rights complaint system and produce resources in Urdu and Punjabi so community input can shape policy .

Emaan argues that civil society organisations in Pakistan and similar settings face structural obstacles in building sustained evidence systems for AI governance. She stresses that fragmented funding and siloed projects make collaboration and institutional support necessary if CSOs are to contribute meaningfully.

She said that in countries like Pakistan, this work is almost entirely donor-funded, with short funding cycles and intense competition, which leads to repeated projects and siloed data collection rather than durable evidence systems . She added that a major solution would be stronger collaboration among CSOs and possibly a supporting body to assist them, while also noting that youth groups often spend their energy searching for donors instead of advancing the work itself .

Emaan argues that participation gaps are not abstract but already visible in who turns up to tech and AI opportunities. The dominance of one social group in a youth hackathon revealed deeper gender, class and minority exclusions that inclusive governance must address.

She described how her organisation hosted a tech hackathon for youth in Pakistan with the Supreme Court and found that all applications came from young Muslim men except for 20% from women, with no applications from Hindu, Christian or Sikh communities . She said these patterns reflected class and community structures in Pakistan and were a serious concern for the commission .

Emaan argues that youth participation in AI governance requires more than inviting young people into the room; it requires addressing unequal access to tools, language barriers and misconceptions about who is qualified to participate. She emphasises that rights-based education and local-language resources are necessary so young people understand that governance is not only for coders or elites.

She contrasted students from top schools, who had access to visas, language models and online courses, with students in government-funded and underprivileged schools where even explaining AI in Urdu was difficult . She recounted trying to translate AI as 'Nakli akal', which students found frightening, and said many students only associated AI with TikTok image generation and believed they needed coding skills to have a say in governance . She also pointed to resources such as Netmission Academy and the Asia-Pacific Youth Internet Governance Forum, and said her organisation has MOUs with 20 universities to develop human-rights-based AI modules .

Emaan argues that within five years, youth participation should be embedded formally in international accountability processes rather than left optional. She also calls for a national institutional mechanism in Pakistan that is independent, youth-led and publicly resourced to address digital harms and governance concerns.

She said youth participation should be mandatory in international reporting processes on digital issues, arguing that the UN should require youth voices in reporting cycles such as CAT and CERD so states must provide evidence of what young people are saying . She also proposed an independent, self-governed digital commission in Pakistan run entirely by youth and supported with state resources, citing concerns about low freedom of speech rankings and online violence affecting children, minorities and women .

Emaan argues that educational stakeholders are not resistant to AI but are actively looking for guidance and tools. She suggests that practical uptake depends on making learning materials free, easy to use and adapted to local languages and contexts.

She said teachers are overworked but welcome tools that can assist them, and she described parents at a high school in Pakistan asking how they could get their children into AI because they feel urgency and uncertainty about what AI means for the future . She argued that if teachers and students are given accessible curricula, workshop designs and language-friendly resources, they are willing to engage, and she cited an ITU survey showing particularly high excitement about AI in South Asia compared with fear .

Addie argues that inclusive AI governance depends on gathering local community evidence and converting it into practical evaluation tools. Her core claim is that when communities can assess AI systems through benchmarks and evaluations, trust is more likely to be built than when systems are simply imposed.

She said her organisation Kaleidoscope helps local communities evaluate and govern AI systems as they enter their environments . She described her approach as 'evidence plus infrastructure equals trust', explaining that evidence means collecting community inputs and infrastructure means turning those inputs into tools such as benchmarks and AI evaluations to test system performance, with the aim of restoring community trust in deployment .

Addie argues that civil society can play a strong role in evidence collection even if it does not control the final policy infrastructure. She also warns that governance systems must widen their understanding of evidence, because community knowledge may otherwise be undervalued compared with better-funded corporate evidence production.

She said CSOs can collect evidence and then plug that evidence into existing institutional infrastructure to turn it into action . As an example, she described Maori communities in New Zealand building speech recognition for their under-resourced language, with an independent CSO curating the data and the New Zealand government placing it in a trust so it benefits the community and is protected from surveillance and commercial exploitation . She also argued that there needs to be more discussion about what qualifies as evidence because for-profit AI actors have far more capital to generate the kinds of evidence that currently dominate governance processes .

Addie argues that even well-intentioned AI deployments can reproduce power imbalances when they are designed from a US perspective and exported elsewhere. She suggests that local accountability and stronger safeguards are needed so 'AI for Good' does not simply transmit external biases into other communities.

She drew on her background as an AI engineer and former AI lead at a philanthropy funding AI for Good deployments, and said that although AI for Good is possible, the movement often lacks guardrails . She specifically criticised the tendency for technologies developed in California or the United States to carry US biases and then be deployed in other regions 'in the name for good' without enough local accountability .

Addie argues that young people should not merely be symbolically included in AI governance. Their involvement should affect real decisions, especially because they will live longest with AI’s long-term consequences and can offer perspectives that differ from older generations.

She said the functional difference between a 25-year-old and a 50-year-old in using AI may be less stark than in older industries because AI is relatively new, making youth perspectives especially important . She referred to a UNICEF essay by Roger Hart describing eight rungs of children’s participation, contrasting tokenism with youth-initiated shared decision-making, and argued that AI governance should give young people genuine authority and test whether their presence changes outcomes rather than serving a performative function . She also noted that long-term effects, including environmental impacts, particularly affect youth .

Addie argues that within five years, governance should include formal participatory channels where people can express expectations for AI systems and see whether those systems meet them. She emphasises transparency and independence so evaluation is not controlled by AI companies themselves.

In her five-year vision, she proposed transparent participatory voting systems through which people can input what they want from AI tools affecting their lives . She added that these systems should be paired with transparent objective scores and independent evaluations showing how well the tools meet those needs, specifically avoiding self-assessment by AI companies .

Addie argues that AI expertise cannot be assumed on the basis of age or social role because the technology is new to nearly everyone. This means intergenerational governance and education should be shared, with youth having an equal voice rather than being treated only as learners.

Responding to the audience, she said she does not think teachers and parents are currently equipped to teach AI because it is a new skill set and understanding it has very little to do with age . She argued that this makes youth participation more credible and necessary, while also saying youth should not have an exclusive role because intergenerational collaboration remains important .

Vasileios argues that inclusive AI governance requires operational mechanisms that make regulation usable in practice. Rather than remaining at the level of principles, governance should be supported by concrete repositories of risks, controls and mitigation techniques that enable accountability.

He said Credo AI works by offering the operational layer that translates regulatory requirements into actions organisations can actually implement, with the goal of making AI governance more robust, holistic and sustainable . He specified that this includes creating risk libraries, control libraries and risk mitigation techniques, co-shaped with customers, so there is a substantive inventory of evidence that can support accountability structures in AI governance .

Vasileios argues that evidence and infrastructure are mutually reinforcing rather than competing approaches. He maintains that civil society’s influence depends on independent and trustworthy evidence, but also on shared language and common frameworks that prevent fragmentation across organisations and regulators.

He said the issue should not be framed as a binary between infrastructure and access to evidence, but as a combined approach in which CSOs provide independence and evidence integrity within a common taxonomy or language . He warned that fragmented ways of discussing AI governance across CSOs and regulators create a 'Tower of Babel' situation, and argued that fragmentation is especially harmful for a global issue like AI .

Vasileios argues that AI adoption is outpacing governance capacity inside organisations. This creates systemic risk because firms rely heavily on AI tools without the internal maturity or controls needed for responsible use.

He cited a proof point that more than 70% of companies rely on AI tools, while only about 4-5% of those companies know how to conduct AI governance . He concluded that these organisations are using AI without mature infrastructure to control it, which he described as deeply problematic .

Vasileios argues that youth are often kept far from real decision-making despite being among the groups most affected by AI. He says meaningful participation requires early involvement, concrete influence over decisions and sustained follow-up so youth are not merely symbolically consulted.

He said there is significant distance between decision-makers and youth, criticised tokenisation that gives young people visibility without power, and argued they must be involved in processes that truly matter . He stressed that youth participation should be anchored in locally deployed AI systems and that young people must be brought in proactively before decisions are predetermined . He also noted that youth are heavily exposed to AI through proctoring tools, smartphone systems and future hiring algorithms, making it crucial that they are aware of and able to claim decision-making space .

Vasileios argues that meaningful youth inclusion in the next five years will depend on resourcing, implementation and institutional change. He stresses that symbolic consultation must be replaced by funded, impact-driven participation supported by AI literacy embedded in everyday institutional practice.

He said that one of the main barriers identified earlier was access to evidence and funding, and argued that more funds should be invested in this direction over the next five years . He criticised tokenisation in the EU context as performative participation without follow-through and said he wants to see impact-driven inclusion of youth in decision-making, potentially with strengthened rights . He further argued that AI literacy must become ingrained in the daily thinking and actions of institutions because future generations will need to manage AI’s implications .

Vasileios argues that schools and educators should be a central focus of AI literacy policy because children are already using AI in ways that institutions are not prepared for. He frames this as an obligation, not an optional add-on, because educational settings are directly affected by AI use.

He recounted visiting his niece in Greece and discovering that she was using Claude on her tablet for homework and had effectively taught herself by asking the AI tool directly, with no direct supervision from teachers . He used this example to argue that teachers and educators must be heavily invested in AI literacy efforts and that schools should be treated as a top-tier priority, ideally within public-law obligations around AI literacy .

Pedro frames the session around the need for AI governance that is more inclusive, diverse and democratic. He asks the panellists to explain what concrete tools and mechanisms they are developing to widen participation by civil society and other parts of society.

He introduced the session by saying it would discuss AI governance and how to build a more inclusive and diverse governance of AI . He then asked the panellists what concrete tools and mechanisms they were building to ensure more plural and democratic participation of civil society and other sectors in AI governance, and to explain the gaps these efforts address .

Pedro frames the second part of the discussion around the role of civil society in evidence production and governance design. He asks whether CSOs need independent evidence infrastructures or alternative ways for their knowledge and lived experience to shape expert-led AI frameworks.

He identified a common thread in the panellists’ projects around how evidence is produced and embedded in AI governance infrastructure . He then asked how CSOs could participate more actively in current expert-group settings and whether independent evidence infrastructure is needed to shape decision-making or whether other pathways could translate CSO knowledge into present AI frameworks .

Pedro repeatedly steers the conversation towards inclusion, highlighting that AI governance failures are tied to exclusion from consultation and participation. His moderation emphasises that governance quality depends on who is able to contribute evidence, expertise and lived experience.

He framed the overall discussion around democratic participation by civil society and other parts of society in AI governance . Later, he explicitly summarised the panel’s recurring themes as the need for collaboration, state participation and evidence, which linked governance gaps to who is included in these processes .

Pedro explicitly raises youth participation as a governance design question rather than a side issue. He asks whether youth should be engaged through the same structures as experts or through more tailored mechanisms that respond to their specific needs.

He asked how AI governance could be made in a way that mirrors youth participation and how young people could be brought into the conversation . He also asked whether existing expert-oriented methods are sufficient or whether something more specific is needed to cater to youth .

Pedro concludes the panel by prompting the speakers to move from diagnosis to design. He asks them to imagine concrete institutional arrangements, frameworks and benchmarks that should be in place within five years to support youth participation in AI governance.

He introduced a final exercise asking the panellists to imagine five years from now and say what they would want to see implemented in terms of youth participation in AI governance . He specifically asked about the kinds of bodies, frameworks and benchmarks they hoped would exist by then .

Pedro’s moderation consistently links AI governance to public understanding and participation, not only to expert regulation. By posing questions on youth inclusion and then inviting audience questions on education, he helps connect governance debates to practical literacy and social capacity-building.

He asked how AI governance could better include youth and whether young people need specific mechanisms for participation beyond the expert model . He then opened the floor to audience questions, which led directly into discussion of whether parents and teachers are equipped to teach AI and how intergenerational dialogue can be fostered .

The audience raises the concern that current governance mechanisms may be overtaken by rapid advances in AI capability. The question challenges the panel to consider whether the tools being discussed are future-proof if very advanced AI emerges soon.

An audience member referred to the belief of Google’s co-founder and CEO that highly advanced AI may arrive before 2030, and asked what that would mean for AI governance if it comes within five years .

The audience argues that broad agreement on intergenerational dialogue is not enough without practical methods for making it happen. The question seeks concrete strategies for connecting generations in AI governance and education.

One audience member said they agreed that intergenerational dialogue needs to take place, but asked directly how that dialogue can be fostered in practice .

The audience argues that the possibility of rapid AI progress increases the urgency of current governance work. Rather than assuming existing institutions will cope, the question implies that governance capacity must be strengthened immediately.

The audience question linked the prospect of highly advanced AI arriving before 2030 to the practical challenge of how governance should respond if such systems emerge in five years . This framing treated future AI capability as a reason to test and strengthen governance capacity now .

The audience argues that effective youth participation depends on whether the adults around young people can support them. The question highlights schools and families as crucial intermediaries in building the intergenerational dialogue needed for meaningful AI engagement.

An audience member asked whether teachers and parents are equipped to teach AI to the next generation and, if not, how society can upscale that capacity .

The audience challenges the panel’s five-year planning horizon by introducing the possibility that AGI-level systems may emerge during that same period. This implies that future recommendations should not only improve participation but also remain responsive to potentially transformative technological change.

The audience member explicitly asked that if highly advanced AI arrives within five years, what that development would mean and how AI governance could be done under those conditions .

The audience argues that governance cannot succeed without building practical capacity in the adults who shape children’s learning and development. This positions AI literacy for teachers and parents as a prerequisite for meaningful social participation in AI governance.

The audience member asked whether teachers and parents are equipped to teach AI and, if not, how that capacity can be developed, explicitly tying intergenerational support to the needs of the next generation .