Sunak sounds the alarm on AI

‘Dr Death’ was the nickname bestowed on Rishi Sunak by one scientist during Covid. But ‘Dr Doom’ seemed a more apt sobriquet at certain points during his big speech today on artificial intelligence. The Prime Minister evoked the spectre of humanity ‘losing control of AI completely’ to a ‘superintelligence’ that could result in ‘extinction’. He warned of a world in which AI facilitated chemical weapons, disinformation and child sexual abuse. Therefore, governments ought to step up, he argued, as ‘only nation states have the power and legitimacy to keep their people safe’.

How best to do this then? Sunak this morning announced the world’s first AI safety institute in the UK to explore ‘all the risks – from social harms like bias and misinformation, through to the most extreme risks’. But fundamentally he believes the answer is not a ‘rush to regulate’, asking: ‘How can we write laws that make sense for something we don’t yet fully understand?’ Instead, he believes, the solution lies in research and collaboration: researching AI’s capabilities and working with foreign governments to mitigate its risks.

Sunak trumpeted next week’s conference at Bletchley Park and his AI Taskforce, modelled on the Covid one led by Kate Bingham. Yet as James Phillips argues in this week’s Spectator, the AI Taskforce has a much trickier challenge: ‘Everyone knew the goal was to roll out a vaccine to combat Covid. This was something tangible – and obviously there was a sense of urgency that Covid needed to be tackled. There is, by contrast, no consensus among governments about whether AI is a big risk, let alone if states need to regulate its development.’

Some in No. 10 think that Sunak can score political wins when it comes to technology too, contrasting his fluency on the subject with other politicians’ hesitancy. The PM is an enthusiast for innovation but today’s speech was a world away from the brillo and boosterism offered by Boris Johnson. Sunak seems to prefer to lean into his role of speaking ‘hard truths’ – balancing the potential gains of AI with its very real risks.