Techbros party on

The 'Safer AI Summit' comes to Bletchley Park

Here’s a new take on an old meme. If you are hosting your perfect ‘safer AI summit’, who do you invite to the party?

If you’re the UK prime minister, the answer is depressingly predictable. Guest of honour is Elon Musk, current emperor of X (the ‘anti-woke virus’) and until recently a public advocate of AI regulation. Turns out, if this involves measures against discrimination and inequity, he is less of a fan.

Musk will not only be airing his views at the summit but ‘in conversation with’ the Prime Minister, and I’m sure we are all on tenterhooks to learn what parts of ‘anti-woke’ they plan to enshrine in regulation, or indeed in non-regulation. Of course gilded invitations absolutely must go to senior representatives of DeepMind, Google, Anthropic and OpenAI, who are deeply committed to the safety of their companies’ market dominance and making sure regulation is as tough as possible for anyone wanting to get into AI on a smaller, more open and more specialist scale. Meta’s president of global affairs, Sir Nick Clegg, will enjoy revisiting old haunts, and if the last 13 years have left the UK a little… haunted, well it’s time to put his role in that unpleasantness to one side, because Nick has his eyes on wider shores and more expansive futures. And talking of expansive futures, you simply must invite Geoffrey Hinton, Godfather of AI, because his concerns are so futuristic you can practice Looking Very Concerned over the starters and be laughing it all off by the time the vegetables arrive.

And just to be clear, it is only the future we will be discussing at this party. Present harms, such as the oppressive labour relations involved in training AI models, or the downgrading of professional work entailed by their use, or the loss of livelihood to creative workers will not be mentioned over the wine and canapes. It would be positively rude to bring up the injustices visited on racial minorities when AI is used in policing, border control, profiling and deciding whose children are at risk. The massive environmental costs of generative AI, the concentration of power and knowledge in a few giant corporations, surveillance, bias, political leverage and disinformation, non-consensual pornography - these are very much off-topic.

In an interview with Politico, hostess-with-the-mostest, technology secretary Michelle Donolan:

said the summit would be “exclusively” focused on the risks and opportunities of frontier AI, defined usually by the industry as AI systems that exceed the capabilities of today’s leading models, and so will not be the forum for broader conversations on other topics.

Michelle is performing her favourite dinner party turn here, which is telling people what they must not talk about. She has been knocking it out of the park in this role recently, telling scientists what they can’t research and telling academic communities what they can’t discuss or criticise, and in the process I think we can all agree really bringing new meaning to the terms ‘minister’, ‘science’, ‘technology’ and ‘research’, as well as ‘freedom of speech’.

So, ‘broader conversations’ are out, along with any guests who might find it difficult to mix in this exalted company. Some people who are not on Michelle and Rishi’s list are kicking off about it, but that’s only to be expected. After all, the best parties are defined by the people who aren’t allowed in. When Michael Birtwhistle of the Ada Lovelace institute complains that the focus is on ‘models that don't exist yet and capabilities that those models don't yet have’, I think we can all hear the FOMO talking. And we can write off as sheer party-pooper pique the open letter signed by trade unions and professional bodies, Amnesty International and Liberty, open society representatives, privacy activists, data rights groups, open source developers and civil society groups that have denounced the summit as a front for the very organisations most in need of external scrutiny and public regulation, and most unwilling to hear from the people their products put at risk.

Update (10pm, 1 November)

Well, I think that went swimmingly. The Americans and the Chinese managed to be polite to one another, nobody brought up any awkward topics, and the hosts were made to feel important even while the US was planning its own party to which nobody else was invited. And Michelle’s parting words will be useful to any hostess who feels the small talk beginning to slip out of her comfort zone:

“We need to properly understand the problem before we apply the solutions.”

Indeed we do. Of course, it’s quite hard to understand the problem of ‘frontier’ AI, because by definition it doesn’t exist yet, and everyone is absolutely committed to not understanding the problems that do. Still, the AI corporations agree that regulation is needed, so it’s really only a matter of asking them not to approach any kind of frontier until we know where it is and what is on the other side and how to regulate it.

Meanwhile, party on dudes.