AI Safety is a Social Problem
Introducing the AI and Society Summit
The most important questions for AI governance relate to the kinds of societies we want to live in, and how AI can support, not undermine, human and planetary flourishing. Come and join us at the AI and Society Forum in London on 31 October to be part of the conversation.
AI Safety is, primarily, an issue of power and influence, but it’s very often rendered as a technological puzzle that needs to be solved. After all, the people who care the most about technologies tend to be technologists, and when you’re holding a hammer everything looks like a nail.
In today’s FT, columnist John Thornhill opined the lack of a “political Alan Turing” to come and “crack the code” of safety. But social safety is not a matter for a lone male hero — it’s a question for the whole of society, and whoever ultimately sets the rules for AI will have a great deal of power.
Some (including the UK government) are pushing for a global approach to governance, and if they’re successful that will confer global-scale power. It won’t simply be the power of setting the guardrails for those technologies, but the power of operating outside the limits of existing social contracts and setting new norms.
This is the context for the AI Safety Summit on 1-2 November, which is currently preoccupying everyone who works in UK tech policy. In particular, the matter of who will and won’t be in the room in Bletchley Park has been a matter of much speculation and debate since the event was announced in the summer with a press release that quoted five corporates and not a single voice from academic or civil society.
The two-day Summit is a closed-door, invitation-only event that will probably have little to do with the nitty gritty of technology governance and more to do with geopolitics and attracting big business to the UK. Recent speculation indicates that the overall goal is to put Brexit Britain on the technology map as the home of a new AI Safety Institute, and perhaps an extremely cynical person might suspect Sunak’s sudden interest in tech points to an interest in post-election job opportunities. But whatever the Summit is for, it seems unlikely to be the place where the really crunchy issues concerning AI governance are resolved.
There is, perhaps, another post to be written that forensically picks apart the focus of the Summit, but it’s Saturday night so that can wait. However it’s worth saying that the guidance published by government so far on what will be covered hovers in that odd territory between vague and extremely specific that often indicates someone’s not entirely sure what they’re talking about. (The diagram below is from the official literature. As far as I’m aware, “dangerous capabilities” and “cutting-edge LLMs” are not, as yet, regarded as technical terms.)
Image taken from DSIT’s AI Safety Summit: Guidance
Many of us in civil society — the space between the market and the state — have been working over the Summer to try and increase the scope and transparency of the Summit. With less than a month to go, it’s clear that won’t happen; while the invitation list has, apparently been widened from the five tech companies quoted in the press release, I don’t get the sense it will be a space where challenge will be welcomed or encouraged.
So — with help from many others — my team at Promising Trouble is producing the AI and Society Forum on 31 October. It’s a free event for changemakers, campaigners, activists and experts by experience — for people who aren’t supported by big institutions, who are fighting for equity and justice in spaces both in and beyond the technological. The focus will be on the political, social and environmental impacts of AI and include topics such as the climate impacts of data storage and compute; the use of AI on policing and borders, workers’ rights, democracy, misinformation, privacy and education. We’re inviting a couple of keynote speakers, but most of the day will be shaped by the attendees. There are 154 free tickets, and we’ll release the first tranche on Tuesday 10 October. Sign up to get yours.