Co-creating a Shared Human Rights Agenda for AI Regulation and the Digital Welfare State

On September 26, 2023, the Digital Welfare State and Human Rights Project at the Center for Human Rights and Global Justice at NYU Law and Amnesty Tech’s Algorithmic Accountability Lab (AAL) brought together 50 participants from civil society organizations across the globe to discuss the use and regulation of artificial intelligence in the public sector, within a collaborative online strategy session entitled ‘Co-Creating a Shared Human Rights Agenda for AI and the Digital Welfare State.’ Participants spanned diverse geographies and contexts—from Nigeria to Chile, and from Pakistan to Brazil—and included organizations working across a broad spectrum of human rights issues such as privacy, social security, education, and health. Through a series of lightning talks and breakout room discussions, the session surfaced shared concerns regarding the use of AI in public sector contexts, key gaps in existing discussions surrounding AI regulation, and potential joint advocacy opportunities.

Global discussions on the regulation of artificial intelligence (AI) have, in many contexts, thus far been preoccupied with whether to place meaningful constraints on the development, sale, and use of AI by private technology companies. Less attention has been paid to the need to place similar constraints on governments’ use of AI. But governments’ enthusiastic adoption of AI across public sector programs and critical public services has been accelerating apace around the world. AI-based systems are consistently tested in spheres where some of the most marginalized and low-income groups are unable to opt out – for instance, machine learning and other technologies are used to detect welfare benefit fraud, to assess vulnerability and determine eligibility for social benefits like housing, and to monitor people on the move. All too often, however, this technological experimentation results in discrimination, restriction of access to key services, privacy violations, and many other human rights harms. As governments eagerly build “digital welfare states,” incorporating AI into critical public services, the scale and severity of potential implications demands that meaningful constraints be placed on these developments. 

In the past few years, a wide array of regulatory and policy initiatives aimed at regulating the development and use of AI have been introduced – in Brazil, China, Canada, the EU, and the African Commission on Human and Peoples’ Rights, among many other countries and policy fora. However, what is emerging from these initiatives is an uneven patchwork of approaches to AI regulation, with concerning gaps and omissions when it comes to public sector applications of AI. Some of the world’s largest economies – where many powerful technology companies are based – are embarking on new regulatory initiatives with impacts far beyond their territorial confines, while many of the groups likely to be most affected have not been given sufficient opportunities to participate in these processes.

Despite these shortcomings, ongoing efforts to craft regulatory regimes do offer a crucial and urgent entry point for civil society organizations to seek to highlight critical gaps, to foster greater participation, and to contribute to shaping future deployments of AI in these important sectors.

In hosting this collaborative event on AI regulation and the digital welfare state, the AAL and the Center sought to build an inclusive space for civil society groups from across regions and sectors to forge new connections, share lessons, and collectively strategize. We sought to expand mobilization and build solidarity by convening individuals from dozens of countries, who work across a wide range of fields – including “digital rights” organizations, but also bringing in human rights and social justice groups who have not previously worked on issues relating to new technologies. Our aim was to brainstorm how actors across the human rights ecosystem can, in practice, help to elevate more voices into ongoing discussions about AI regulation.

Key issues for AI regulation in the digital welfare state

In breakout sessions, participants emphasized the urgent need to address serious harms that are already resulting from governments’ AI uses, particularly in contexts such as border control, policing, the judicial system, healthcare, and social protection. The public narrative – and accelerated impetus for regulation – has been dominated by discussion of existential threats AI may pose in the future, rather than the severe and widespread threats that are already seen in almost every area of public services. In Serbia, the roll-out of Social Cards in the welfare system has excluded thousands of the most marginalized from accessing their social protection entitlements; in Brazil, the deployment of facial recognition in public schools has subjected young children to discriminatory biases and serious privacy risks. Deployments of AI across public services are consistently entrenching inequalities and exacerbating intersecting discrimination – and participants noted that governments’ increasing interest in generative AI, which has the potential to encode harmful racial bias and stereotypes, will likely only intensify these risks.

Participants also noted that it is likely that AI will continue to impact groups that may defy traditional categorizations – including, for instance, those who speak minority languages. Indeed, a key theme across discussions was the insufficient attention paid in regulatory debates to AI’s impacts on culture and language. Given that systems are generally trained only in dominant languages, breakout discussions surfaced concerns about the potential erasure of traditional languages and loss of cultural nuance.

As advocates work not only to remedy some of these existing harms, but also to anticipate the impacts of the next iterations of AI, many expressed concern about the dominant role that the private sector plays in governments’ roll-outs of AI systems, as well as in discussions surrounding regulation. Where tech companies – who are often protected by powerful lobby groups, commercial confidentiality, and intellectual property regimes – are selling combinations of software, hardware, and technical guidance to governments, this can pose significant transparency challenges. It can be difficult for civil society organizations and affected individuals to understand who is providing these systems, as well as to understand how decisions are made. In the welfare context, for example, beneficiaries are often unaware of whether and how AI systems are making highly consequential decisions about their entitlements. Participants noted that human rights actors need the capacity and resources to move beyond traditional human rights work, to engage with processes such as procurement, standard-setting, and auditing, and to address issues related to intellectual property regimes and proliferating public-private partnerships underlying governments’ uses of AI.

These issues are compounded by the fact that, in many instances, AI-based systems are designed and built in countries such as the US and then marketed and sold to governments around the world for use across critical public services. Often, these systems are not designed with sensitivity to local contexts, cultures, and languages, nor with cognizance of how the technology will interface with the political, social, and economic landscape where it is deployed. In addition, civil society organizations face additional barriers when seeking transparency and access to information from foreign companies. As AI regulation efforts advance, a failure to consider potential extraterritorial harms will leave a significant accountability gap and risk deepening global inequalities. Many participants therefore noted both the importance of ensuring that regulation in countries where tech companies are based includes diverse voices and addresses extraterritorial impacts, but also to ensure that Global North models of regulation, which may not be fit for purpose, are not automatically “exported.”

A way forward

The event ended with a strategizing session that revealed the diverse strengths of the human rights movement and multiple areas for future work. Several specific and urgent calls to action emerged from these discussions.

First, given the disproportionate impacts of governments’ AI deployments on marginalized communities, a key theme was the need for broader participation in discussions on emerging AI regulation. This includes specially protected groups such as indigenous peoples, minoritized ethnic and racial groups, immigrant communities, people with disabilities, women’s rights activists, children, and LGBTQ+ groups, to name just a few. Without learning from and elevating the perspectives and experiences of these groups, regulatory initiatives will fail to address the full scope of the realities of AI. We must therefore develop participatory methodologies that bring the voices of communities into key policy spaces. More routes to meaningful consultation would lead to greater power and autonomy for previously marginalized voices to shape a more human rights-centric agenda for AI regulation. 

Second, the unique impacts that public sector use of AI can have on human rights, especially for marginalized groups, demands a comprehensive approach to AI regulation that takes careful account of specific sectors. Regulatory regimes that fail to include meaningful sector-specific safeguards for areas such as health, education, and social security will fail to address the full range of AI related harms. Participants noted that existing tools and mechanisms can provide a starting point – such as consultation and testing requirements, specific prohibitions on certain kinds of systems, requirements surrounding proportionality, mandatory human rights impact assessments, transparency requirements, periodic evaluations, and supervision mechanisms.

Finally, there was a shared desire to build stronger solidarity across a wider range of actors, and a call to action for more effective collaborations. Participants from around the world were keen to share resources, partner on specific advocacy goals, and exchange lessons learned. Since participants focus on many diverse issues, and adopt different approaches to achieve better human rights outcomes, collaboration will allow us to draw on a much deeper pool of collective knowledge, methodologies, and networks. It will be especially critical to bridge silos between those who identify more as “digital rights” organizations and groups working on issues such as healthcare, or migrants’ rights, or on the rights of people with disabilities. Elevating the work of grassroots groups, and improving diversity and representation among those empowered to enter spaces where key decisions around AI regulation are made, should also be central in movement-building. 

There is also an urgent need for more exchange not only across the human rights ecosystem, but also with actors from other disciplines who bring different forms of technical expertise, such as engineers and public interest technologists. Given the barriers to entry to regulatory spaces – including the resources, long-term commitment, and technical vocabularies imposed – effective coalition-building and information sharing could help to lessen these burdens.

While this event brought together a fantastic and energetic group of advocates from dozens of countries, these takeaways reflect the views of only a small subset of the relevant stakeholders in these debates. We ended the session hopeful, but with the recognition that there is a great deal more work needed to allow for the full participation of affected communities from around the world. Moving forward, we aim to continue to create spaces for varied groups to self-organize, continue the dialogue, and share information. We will help foster collaborations and concretely support organizations in building new partnerships across sectors and geographies, and hope to continue to co-create a shared human rights agenda for AI regulation for the digital welfare state.

As we continue this work and seek to support efforts and build collaborations, we would love to hear from you – please get in touch if you are interested in joining these efforts.

November 14, 2023. Digital Welfare State and Human Rights Project at NYU Law Center for Human Rights and Global Justice, and Amnesty Tech’s Algorithmic Accountability Lab.