What have we learned from the first AI Safety Summit?

With 100 attendees over two days, the UK AI Safety Summit is over. What have we learned? 

Here’s a summary from the things I have read/seen/heard this week.

The positives

A diplomatic coup: AI knows no borders, so mitigating the risks requires global collaboration. The Bletchley Declaration is a (tentative) first step towards that goal. Getting the US and China together with 26 others was a “major diplomatic coup” for the UK. For those who want to dig into the philosophical meaning of some of the terms in the declaration, I’d highly recommend Dr. Brennan Jacoby‘s post here.

Meeting objectives: the Summit met all four of its pre-conference goals, including an agreement by the leading companies to work with governments to conduct pre- and post-deployment testing of their next generation of models – described as “huge” by the chair of the UK AI Frontier Model Taskforce Ian Hogarth.  

But some missed opportunities

Marginalised groups: on day one of the conference, more than 100 civil society stakeholders wrote to the Prime Minister to say the “communities and workers most affected by AI have been marginalised by the Summit.” Since launching, the Appraise Network has argued for all those impacted by AI to be part of such policy discussions. 

Ignoring our own: alongside this, by focusing on frontier models and existential risks, the Summit side-lined its domestic AI industry, argues Keegan McBride of the Oxford Internet Institute. But our homegrown companies are vital if we want to remain a global AI leader and realise the benefits of the technology (see looking forward further below).

Stealing a march: the US President’s Executive Order on AI, announced on Monday, focused on here-and-now risks such as privacy, bias and guidance on watermarking to combat misinformation in contrast to the UK’s Summit focus on existential threats. Arguably, by making such an announcement, the US has helped positioned itself as a global leader on AI policy and safety, in competition with the UK. 

Looking forward

Driving adoption: deployment of AI by UK firms lags behind Germany, France, Italy and Spain. As Daniel Wilson of BT argues: “We have an AI deployment deficit” that we must address to fully realise the benefits of AI to “improve everyday life.” 

A national approach: reinforcing this point, Sam Gilbert of the Bennett Institute of Public Policy has argued in a recent report that the UK should use its “strengths in fintech, cybersecurity and health-tech to build software – the apps, tools and interfaces – that harness AI for everyday use.” 

Reimagining the state: and it is not just business that can benefit. Benedict Macon-Cooney of the Tony Blair Institute argues that “by building on the AI-era infrastructure of cloud, compute and high-value data, a wholly new concept of the state is possible: a state that is truly strategic, harnessing the power of science and technology to perform its functions better and at lower cost.”

With hundreds of op-eds and probably thousands of LinkedIn posts, this just scratches the surface. Keen to hear what other have made of this week.