Let’s give people power over AI

The predominant emotion associated with AI by Panel members at the start of their deliberation was fear. Narratives that emphasise the brand new, frontier or complex nature of technologies intentionally disrupt our ability to understand and interrogate how we want technologies to fit into our lives and societies. When AI’s creators are hailed as the only people who understand its impacts, they get to set the terms for its deployment and regulation. When AI is characterised as the solution to the world’s problems, those who question or resist it are cast as Luddites, holding back progress. Technology becomes something that is done to us, rather than something in service to us. No wonder it makes people worried.

Simply learning about AI can help to counter these narratives. Getting hands-on experience that, for example, ChatGPT has a knowledge cut-off date, and can’t answer questions about today’s weather, helped to foster within our Panel members a clearer understanding of what a generative AI system is (a model trained on selected data inputs chosen by its creator and constrained by the costs of re-training the model), and is not (the live equivalent of a human agent or brain). And perhaps counterintuitively, we found that with this greater understanding of the limits of current systems, people felt more hopeful about the potential uses of AI. It became less magical, and more of a tool they could wield.

Scaling this democratisation is the challenge. AI is often seen as a global and cross-cutting phenomenon and therefore in need of global and cross-cutting rules and regulation. This can lead us to forms of decision-making that either ignore people altogether or aim for shallow but broad, often technology-mediated, participation only possible with a level of digital literacy most people lack.

But the decisions we need to make about incorporating data and AI into our workplaces, schools, hospitals, and local communities are highly context specific. The details of the technology matter: what data is collected? How is it used? Who is it shared with? How is automation embedded into bureaucratic processes? What are the business models? How is it monitored and governed? The details of each community matter too, with different demographics, sensitivities, and norms. For example, predictive policing may be acceptable when targeting patrols to streets suffering a rash of burglaries, but unacceptable when used in ways that increase the use of police powers to ‘Stop and Search’ people in neighbourhoods with large ethnic minority communities. AI chatbots may be welcome when they give us faster access to help with our broadband connections, but undesirable as a replacement for proper support for mental health conditions.

These context-specific decisions about the deployment of data and AI systems require context-specific dialogues with the communities affected by them. As well as rightly giving those affected by AI a say in its deployment, smaller, distributed and local engagements can meet people where they are, and provide an opportunity for self-directed, goal-oriented learning. The challenge then shifts from the unachievable goal of ensuring everyone is equipped to make decisions individually and independently, to one of enabling groups and communities to learn, deliberate and decide together. It can then progress to getting people who currently control data and AI to take notice.

An active civil society is vital to enable distributed dialogues to happen. Existing formal and informal organisations such as worker unions, school governing boards, patient associations, and local community groups are catalysts with established and understood engagement mechanisms that should be brought to questions of data and AI. As importantly, grassroots civil society organisations are uniquely positioned to take collective action and demand a powerful voice in how AI affects the communities they represent, when the organisations rolling out these technologies won’t listen. This might take the form of campaigning initiatives that seek to influence the rollout of AI systems, or the re-analysis and interpretation of data and AI by the communities they concern. It could even be the collection and ownership of data that challenge official sources by better representing those communities’ lived experiences – what the authors of Data Feminism, Catherine D’Ignazio and Lauren Klein, term ‘counterdata’.

To make this work, current mainstream narratives about AI – that it is new, complex, and the preserve of technical experts – need to be challenged. We need to strongly assert the primacy of the public in decisions about technology, over the companies and governments seeking to exploit it. We should highlight the stories of people with lived experience of AI and automated decision-making as the real experts on AI’s impact, rather than focussing on the developers in AI labs. We should reject the notion that automated decisions or generative AI systems are neutral or objective, when there is plenty of evidence that they can be biased or buggy. Whilst the Horizon computer system developed by Fujitsu for the Post Office didn’t use AI, that scandal has much to teach us about the way in which ordinary people need to be heard and trusted when they highlight that software has gone wrong, as well as the power of collective action.

Our narratives must also be hopeful rather than fearful. This does not mean blind techno-optimism: far from it. As our People’s Panel participants showed us, hope comes from being realistic about AI’s faults and limitations and from having agency over its implementation. We need hope for better lives, and societies to direct the development of AI towards those goals. Finally, our hope must not be passive, but instead a hope we work towards through collective action and solidarity to bring about the world we want to see.