Governing AI for the society we want

First published in Civil Service World on 24 September

It is an understatement to say that developments in artificial intelligence are moving quickly.

In a few years, AI has moved from a specialist concern for technologists into the mainstream of public policy. Governments are investing heavily in its adoption, businesses are redesigning services around it, and people are encountering it and using it more often in their lives.

So far, much of the debate about AI governance tends to centre on three familiar concerns: supporting innovation, improving productivity and managing risk.

All three are important, but at Carnegie UK we think they should be preceded by a more fundamental question: what outcomes do we want AI to deliver for society?

Our new report, Governing AI for Collective Wellbeing, argues that the UK should start from that question and make collective wellbeing the organising principle for AI governance.

This is not a case for embracing or resisting AI as a matter of principle, nor for slowing innovation. It is a case for judging decisions about AI by the outcomes that matter to people and communities.

At Carnegie UK, we define collective wellbeing as the conditions that enable everyone to live well now and in the future. We describe those conditions through four interconnected domains: social wellbeing, economic wellbeing, environmental wellbeing and democratic wellbeing.

Applying a collective wellbeing lens changes the policy conversation by shifting attention from what AI can do to the outcomes it should deliver. It asks what an AI system will mean in practice for people, communities, democracy and the environment.

Will it help people get the support and services they need? Will it contribute to fair work and decent living standards? Are its environmental costs justified by the benefits it creates? And will it strengthen public trust, accountability and democratic participation?

AI is already influencing how decisions are made, services are delivered, information circulates and power is exercised. Policy and governance choices made now will help determine whether AI narrows or widens inequalities, strengthens or weakens democratic institutions, and supports or frustrates environmental goals.

The UK still has time to influence that direction.

Public expectations, institutional norms and governance arrangements around AI are still taking shape. That gives governments a window to embed wellbeing principles before poor practice becomes entrenched. Building wellbeing, transparency and accountability into AI systems from the start is far easier than trying to retrofit them later. Governments do not need to wait for new institutions or sweeping legislation. They can begin now by using the powers and processes already available.

Ministers and civil servants already have useful levers at their disposal. Business cases, procurement processes, assurance frameworks, public registers and regulatory reporting can all be used to judge whether high-impact AI systems will strengthen or undermine collective wellbeing. The test should not stop at whether a system works, it should also ask whether the system creates public value and improves lives.

To guide that work, our new report sets out seven core principles for AI governance in the UK:

• Wellbeing centred design

• Accountability, transparency and redress

• Human-in-the-loop governance

• Fairness and non-discrimination

• Democratic oversight and participation

• Environmental sustainability, and

• Long-term stewardship for future generations.

These principles turn familiar public values into practical tests that governments can apply when deciding whether to develop, procure, deploy or regulate AI systems.

The debate on AI governance is framed as a trade-off between innovation and regulation but that misses the point. The task is to make sure innovation serves a clear public purpose.

As AI takes on a larger role in our lives, governments need to be explicit about the outcomes it should support. Collective wellbeing provides a coherent basis for making those choices.

There is little doubt now that AI will change many aspects of how we live and work. The task before us is to shape that change so that everyone can live well, now and in the future.