Get all your news in one place.
100’s of premium titles.
One app.
Start reading
Evening Standard
Evening Standard
Business
Simon Hunt

Key questions ahead of Rishi Sunak's AI safety summit

At a tech conference in Oxford a few days ago, a row broke out between two panellists over AI regulation.

One of them, Dr Keegan McBride, an academic from the Oxford Internet Institute, wanted AI models to be democratised so no big tech firm like Google could have a stranglehold on the industry.  

But Carl Miller, research director at think tank Demos, warned how dangerous the technology could be in the wrong hands and urged tougher rules. Opening AI up to the masses was a hippie fantasy, blasted Miller, and police enforcement on cyber crime was already under strain. 

Who cares what the UK thinks anyway? McBride snapped back. All the real innovation was happening in San Francisco. 

The exchange highlighted the tightrope Rishi Sunak must walk as he summons technologists and diplomats from the far corners of the globe to Bletchley Park for his much-trailed AI Safety Summit, which starts tomorrow. 

On the one hand, there are clear threats posed by advances in AI, both nationally and globally, ranging from online scams to national security threats. On the other, no politician wants to create rules that will cripple a very valuable industry domestically -- and in any case, are the UK’s own capabilities big enough for it to take a lead on the world stage? 

So far, the government has tried to capture global attention with warnings of extreme, cataclysmic scenarios that could unfold if frontier AI – the most advanced form of the technology – is left to develop unchecked. 

An AI apocalypse just feels like another thing to add to our already-big pile of potential apocalypses

In his speech last week, Sunak discussed AI-powered state espionage, misinformation and political interference, the development of chemical, biological and radiological weapons, the proliferation of deepfakes and child sex abuse imagery, and the prospect of societal unrest sparked by all of the above. 

“And in the most unlikely but extreme cases, there is even the risk that humanity could lose control of AI completely,” Sunak said, adding that he didn’t want to be "alarmist." 

“The risk is that if Rishi’s speech was trying to attract everyone to the UK it was a strange way of doing it because it sounded very downbeat,” said James Ball, a technology journalist who this week published a report calling for fresh licensing rules on AI models. 

“An AI apocalypse just feels like another thing to add to our already-big pile of potential apocalypses.” 

AI leaders are hoping the government will tone down its focus on extreme scenarios and instead strike a balance with the opportunities open to the sector in the UK, which has shown signs of promise. 

The UK already has 27 AI unicorns, more than Germany and France combined, according to the 2023 State of AI Report by tech investor Air Street Capital, while around a fifth of the most-cited academic research on AI globally was written here. 

“I would like us to make sure we don’t just talk about hypothetical risks and whether robots will kill us all,” said Poppy Gustaffson, the CEO of British cybersecurity firm Darktrace, a summit attendee who will be joining a meeting at Bletchley on the risk of a loss of control on frontier AI. 

“Our imagination is at its best when we’re thinking about all the risks but we’re less good at thinking about all the problems it can solve. 

“I’m a massive AI optimist – our whole business is reliant on AI. There is a risk of knee-jerk regulation [and] regulatory capture by big tech.” 

A focus on the long-term risks of future advances by big tech could be diverting attention from the ones that are already here. Deepfake imagery, scams and hacking attempts are proliferating at pace, and the “menu of regulatory options” to put a stop to them “has already largely sold out,” as Demos research puts it. 

UK-based research firm Aligned AI last month reported that multiple AI chatbots, including OpenAI’s ChatGPT, showed clear signs of gender bias. That already presents a huge societal problem if technologies like these see widespread use, such as in making hiring decisions, assessing welfare claims or marking student exams. 

According to Aligned co-founder Rebecca Gorman, a greater focus on fine-tuning the models that are in use today means there is much less scope for the technology to grow to produce catastrophic outcomes in the years ahead. 

“As we integrate AI into consumers’ lives in industry and commerce then some of the limitations that are inherent to machine learning become more significant,” she told the Standard. 

“They’re not trained to handle outliers – we choose to optimise them on the average situation we get unintended harmful behaviour. Consumer platforms that optimise for engagement are making teenagers depressed and even suicidal. 

“[But] if you don’t have a way to stop today’s AI models from malfunctioning, you won’t be in a good position to handle frontier AI.” 

What can we expect when delegates emerge from hours of intense discussion on Thursday? There are some promising signs.

A US executive order on AI safety revealed last night felt carefully timed, and while world leaders appear to have snubbed the event, the world’s best-known tech billionaire, Elon Musk, has confirmed his attendance. Summit insiders say a joint communiqué has been prepared and is ready for sign-off. But will it be bold? 

“My impression is they’ve scaled back the ambition a bit,” Ball said. 

“Given how polarising this debate is, if they came out with much more than saying this is an annual thing we’ll reconvene, I’d be quite surprised. 

“The more they’ve looked into this, the more they’ve realised that it’s very complex and there’s a lot of conflicting interest.” 

Sign up to read this article
Read news from 100’s of titles, curated specifically for you.
Already a member? Sign in here
Related Stories
Top stories on inkl right now
Our Picks
Fourteen days free
Download the app
One app. One membership.
100+ trusted global sources.