This article is written by Octavia Reeve, Associate Director, Ada Lovelace Institute, UK.

As part of the recently launched Government AI Campus, Apolitical is publishing a series of articles exploring AI adoption in government. These articles take on many forms, from op-eds written by academic experts to interviews with public sector leaders working on AI adoption themselves. See more by visiting the AI Campus.

This article will argue that in developing AI governance, regulation and legislation, it is vital not only to listen to people’s perspectives and experiences in relation to AI — alongside expertise from policymakers and technology developers and deployers — but also to involve them in decision-making in ways that are legitimate, trustworthy and accountable.

The EU AI Act represents a milestone in aligning artificial intelligence technologies with the needs of people and society: it is the first comprehensive attempt to legislate AI. As it passes into law, it’s worth noting that it has no definitive provisions for direct public involvement in its implementation. Article 58a commits only to setting up an advisory forum of diverse commercial and non-commercial stakeholders to support the European Commission in the implementation and application of the AI Act, including civil society and ‘social partners’ and the potential for subject-specific sub-groups that_ could_ include members of the public. We shouldn’t take public involvement in policy decision-making for granted, or assume that legislators and policymakers know when, how — and why — to do it well.

💬 Let the author know what you think about involving public perspectives in AI governance by leaving a comment below

"As technologies become commonplace at work, in homes and education, and the relationship between AI, government and people using and affected by these technologies grows ever-more complex, it is increasingly important to meaningfully involve people in AI policy decision-making."

In the UK, public trust in political decision-making is at a low ebb. In 2021, the British Academy described the UK public’s perceptions as ‘actively at odds with the sense that the central governing system is serving their needs and reflecting their voice’, and the CDEI’s 2023 tracker shows that trust levels in Government data practices remain low (41%) compared to other actors. Public trust will be vital to building trust and legitimacy in decision-making about AI and realising societal benefits. When the UK moves towards AI legislation of its own, it will be imperative that policymakers in the UK Government and other jurisdictions have a robust understanding not just of relevant public attitudes but also of how to involve people in decisions.

Current research from the Ada Lovelace Institute shows some consistent public views in relation to AI: there is strong support for the protection of fundamental rights (for example, privacy) and a belief that regulation is needed. People have positive attitudes about some uses of AI (for example, in health and science development), but there are concerns about AI for decision-making that affects people’s lives (for example, eligibility for welfare benefits).

As technologies become commonplace at work, in homes and education, and the relationship between AI, government and people using and affected by these technologies grows ever-more complex, it is increasingly important to meaningfully involve people in AI policy decision-making.

Importantly, in the public’s mind, there isn’t only one AI: they have nuanced views and differentiate between the benefits, risks, harms and opportunities of existing and potential uses of different technologies. Also significant is that some concerns are associated with socio-demographic differences. This means that there cannot be one model for AI governance — it must be context-specific and take account of specific benefits and harms for different people and communities. For example, the use of facial recognition technologies in public spaces would need different consideration to care homes employing autonomous care workers to look after elderly, vulnerable people.

The so-called ‘deliberative wave’ has ebbed and flowed in the last 20 years and has not yet reached the goal of meaningful, institutionalised involvement and empowerment of people in decision-making. However, it does seem to be lapping at the shores of UK policymaking. In 2023, UK regulator ICO took account of evidence from the British Youth Council and the Ada Lovelace Institute Citizen’s Biometrics Council in the development of their biometric data guidance, and the Department of Culture, Media and Sport (now DSIT) commissioned a mini-public dialogue to consider the UK Government’s approach to managing its Trust Framework and rolling out its digital identity scheme through private-sector technology and service providers.

These are positive steps — Chief of Staff Sue Gray recently hinted that an incoming Labour government might use citizens’ assemblies — but there is still some way to go before governments regularly and systematically involve people as active participants in policy development and delivery. This means not just contributing to public consultations, but understanding and using validated social research methods that enable people’s knowledge and experience to contribute from the beginning of policymaking processes.

In an ideal world, policy decision-making will proactively consider how different public attitudes and participatory approaches can help policymakers to collect robust evidence about what the public wants from AI, find new ways of generating evidence and build legitimate and productive relationships with people using and affected by technologies. Without these approaches, elected representatives and public servants are more likely to approximate an unrepresentative, subjective sense of public opinion, leaving room for misunderstandings and missing information about specific people and communities.

The stakes are undoubtedly high for policy folks, in an environment where decision-makers are held accountable by parliaments, courts, media and public representatives. To support participatory methods as part of policymaking, decision-makers must be able to rely on evidence being high quality, robust, reliable and attributable to specific populations and groups. They must also understand that participatory methods have distinct and important aims compared to quantitative studies: they don’t aspire to be statistically representative, but — designed well — they can include diverse perspectives and provide the opportunity for democratic deliberation that enables reflection and collaborative thinking about big, societal challenges.

Quantitative social science methods can look attractive, providing important information about what broad groups of people think about particular issues. In a fast-moving technology environment, they provide broad-brush evidence to provoke further thinking and questions. But how these studies are constructed is important — in particular how they sample groups who may already be minoritised, underrepresented or vulnerable, and whether it is possible to isolate views of particular communities and perform intersectional analyses (what do Scottish children or Black women in a range of locations across the UK think about AI, for example?).

All this points to all kinds of reasons, including institutional and cultural barriers, that might incentivise policymakers to ignore public voices. These barriers raise questions of power and privilege — who is heard and who is not, who is included in data and who is missing. Changing this requires a shift away from a culturally embedded reliance on specific methods to deliver accurate representations of the public view. Jennifer Gabrys has argued for citizen-science data to be considered ‘just good enough data,’ challenging the reliance on narrow ways of producing, valuing, and analysing datasets.

The world we live in is messy, fluid and hard to pin down, and all research methods freight political and philosophical beliefs. However, if policymakers retain an attachment only to large survey data — and by implication to a holistic, substantial reality that can be reported — they will miss the opportunity to derive value from studies that involve and empower people, like public deliberations or peer research, which can tell us not only what people think, but why.

AI technologies can undoubtedly deliver benefits of efficiency and computational problem-solving and will quickly become part of everyone’s lives — for example, in schools, hospitals and welfare provision. Evidence of public views tells us that it is particularly important to take account of context, which means it is essential for policymakers to have a strong understanding of participatory approaches that reach underrepresented people in communities, and to use them with rigour and integrity.

Alongside all the opportunities that AI presents for societal benefits, there is an opportunity specifically for policymakers: to listen to and engage with the views of the public, so that their decision-making can navigate effectively the complex and fast-moving world of AI with legitimacy, trustworthiness and accountability.

For more information, see our recently published evidence review: https://www.adalovelaceinstitute.org/evidence-review/what-do-the-public-think-about-ai/

Key takeaways:

  • Taking into account people’s perspectives and experiences in relation to AI — alongside expertise from policymakers and technology developers and deployers — is vital to ensure AI is aligned with societal values and needs in ways that are legitimate, trustworthy and accountable.
  • As the UK Government and other jurisdictions consider AI governance and regulation, it is imperative that policymakers have a robust understanding of relevant public attitudes and how to involve people in decisions.
  • There is some evidence that policy decision-makers are open to hearing and including public voices.
  • Public attitudes research and participatory research are different approaches to hearing public voices that gather distinctive insights with different uses. Around AI, they can deliver legitimacy in decision-making.
  • Because of AI's pervasive nature, it is particularly important to take account of context which means that really understanding participatory approaches and using them with rigour and integrity is essential.

Done reading? Make sure to share your own thoughts on the role of public input in AI policymaking by leaving a comment below ⬇️

(Image credit: Unsplash)