UK PM wants to write AI rules for everyone

As the UK opens its global AI summit at Bletchley, the White House is already waving a 110-page executive order on AI, and the G7 has agreed a code of conduct for responsible AI development.

The flurry of activity shows western nations are converging on AI rule-making, but is also a signal the UK government is being outflanked in its attempt to position itself a global leader in AI governance.

The UK is a relative AI minnow compared with the US and China. It has little to show in the way of AI regulation either, beyond a “pro-innovation” white paper published in August. The EU by contrast has been working on regulating AI since 2021 and will likely finalise its extensive legislative bill by the end of 2023.

Biden’s AI executive order, which mandates safety testing for large AI models and guidelines for “watermarking” AI-generated content to identify deep fakes, means the UK is now behind all major international players (the US, EU and China) in AI governance.

The UK’s summit therefore focuses heavily on existential risk from “frontier” artificial intelligence – AI-engineered bio-weapons, automated cyber attacks and super-intelligent systems that humanity may lose control of – a new, and until recently highly speculative, area where Sunak can plausibly present the UK as a global leader.

Sunak is proposing a new UK-based Global AI safety institute that would conduct research into the safety of new AI models before they are made public.

Western leaders are not all playing ball – Macron, Trudeau and Olaf Scholz have all declined to attend the summit. Biden is sending Kamala Harris in his place.

Following his executive order, Harris declared that on AI the US could “catalyse global action and build global consensus in a way that no other country can”. Leaked documents obtained by Politico suggest attending nations will signal AI safety research efforts should continue under existing institutions like the UN and the Global Partnership on AI.

Such a move may leave Sunak’s proposed Global AI Safety Institute dead on arrival.