AI Safety Summit: Five key takeaways from Bletchley Park

Rishi Sunak, experts and politicians at the AI Safety Summit on Thursday, Nov 2 at Bletchley Park (Getty Images)
Rishi Sunak, experts and politicians at the AI Safety Summit on Thursday, Nov 2 at Bletchley Park (Getty Images)

The hotly anticipated AI Safety Summit at the historic Bletchley Park in Milton Keynes is finally over. It's been a rainy and cold few days, much like the weather experienced by WWII codebreakers who famously beavered away at this venue.

Lots of speeches were heard and at the end of the two-day summit on artificial intelligence legislation, which drew in 28 political leaders from around the world, we now have a signed and ratified "world first" AI declaration promising to work together to research AI safety.

A plan was agreed between all parties to test out AI models both before and after launch, for critical national security – and to minimise the risk of societal harms.But what happens now? Here are the five most interesting things we learned from the AI Safety Summit:

AI regulation might stifle open-source tech

Mark Surman, president and executive director of the Mozilla Foundation (Lisbon Council)
Mark Surman, president and executive director of the Mozilla Foundation (Lisbon Council)

On the first day of the AI Safety Summit, the Mozilla Foundation, a US non-profit that promotes open-source technologies, published an open letter signed by 158 academics, politicians and tech firms. The letter called for politicians to avoid legislating in such a way that harms open source — i.e. non-proprietary — technologies from being created for everyone's benefit. In particular, ones related to AI itself.

"Even though openly available models come with risks and vulnerabilities, we believe this [open source] is the way forward to ensure safety in AI. We have seen time and again that increasing public access and scrutiny makes technology safer, not more dangerous," said Mark Surman, president and executive director of the Mozilla Foundation.

"The idea that tight and proprietary control of foundational AI models is the only path to protecting us from society-scale harm is naive at best, dangerous at worst," he added.

He and other tech industry insiders have been calling for more government investment into AI software development, with the view that software developed openly and not behind closed doors, which is observed by but not hindered by policy makers, is the best way to ensure the creation of safer AI systems that don't pose harms to society.

Last night, Mr Sunak interviewed controversial tech entrepreneur Elon Musk at an exclusive event in central London about AI safety and he specifically asked Mr Musk whether insisting that all AI models be open source would help improve transparency and the visibility of potential problems.

Rishi Sunak shakes hands with Elon Musk (Kirsty Wigglesworth/PA) (PA Wire)
Rishi Sunak shakes hands with Elon Musk (Kirsty Wigglesworth/PA) (PA Wire)

Mr Musk was not sure — in fact, he pointed out that open source software development tended to trail proprietary development by at least six-12 months, which matters a lot in the tech industry.

"I would lean towards open source as at least you know what is going on, but it should be said that even if AI is open source, can you really see what is going on?" he asked,

"You can run a bunch of tests and see what [the AI model is] going to do, but it's probablistic, not deterministic. It's not like other normal computer software."At the end of two days of panel debates, it is not clear whether much has changed here. Some EU countries includng France are keen on open source, while the UK and US remain undecided.

"Support from the big players, such as Google and OpenAI isn’t enough, as independent academic papers must also be considered as key sources for information," said Steve Elcock, founder and chief executive of Elementsuite, an AI-based HR and workforce-management platform.

"The open source community is strong and perceived as important for regulation; it must therefore be a critical voice in AI risk mitigation."

Lots of chatter, yet little clarity

Did Rishi Sunak meet the goals of his AI Safety Summit? Experts aren't so sure (Justin Tallis - WPA Pool/Getty Images)
Did Rishi Sunak meet the goals of his AI Safety Summit? Experts aren't so sure (Justin Tallis - WPA Pool/Getty Images)

Experts have not been mincing their words when it comes to the somewhat vague outcomes of the AI Safety Summit, for instance Forrester's VP principal analyst Martha Bennett.

“This declaration isn’t going to have any real impact on how AI is regulated... there’s a bit too much emphasis on far-out, potentially apocalyptic, scenarios," she said.

"While the declaration features all the right words about scientific research and collaboration, which are of course crucial to addressing today’s issues around AI safety, the very end of the document brings it back to 'frontier AI'."

Frontier AI refers to general-purpose AI at the leading edge of current capabilities, according to Dr Nicole Wheeler of the School of Computer Science, at the University of Birmingham and a Turing Fellow.

The other type of AI discussed at the Summit were "narrow AI models", which are specific AI systems with capabilities limited to a specific domain, that are considered to have dangerous capabilities.

There is also a third common type of AI already in use in the world today, called "foundation models", which refer to AI models trained on vast amounts of data and adaptable for a wide range of tasks, such as answering questions and summarising text.

"My only misgiving was the shift in focus from foundation models to frontier models, which does indeed limit our ability to meaningfully regulate existing deployed AI models with the potential to cause harm," she stressed.

"The distinction between frontier and foundation models is fuzzy, and leaves room for uncertainty about which capabilities any new regulations would apply to."

Some people in the global tech industry feel that neither the UK nor the summit as a whole have provided much clarity on what is likely to happen going forward with AI regulation — ironically the entire point of the summit in the first place.

"The Bletchley declaration is an encouraging first step towards the goal of safer AI development, but it needs to be followed by concrete actions and a focus on the short-term implications of AI, from systemic bias to its lack of transparency," added Mr Surman.

UK technology secretary Michelle Donelan seemed to indicate there is a way to go before the Government fully understands the problems of AI (Getty Images)
UK technology secretary Michelle Donelan seemed to indicate there is a way to go before the Government fully understands the problems of AI (Getty Images)

Sachin Dev Duggal is the founder of, a London-based AI-powered app development platform that is one of the most-watched AI startups in the UK.

He is concerned that the UK Government isn't taking into consideration the opinions of homegrown tech companies, as compared to the global tech firms.

"This country is a hotbed of innovation. However, I see a concerning trend of focusing on the opinions of established tech giants from the US to the detriment of home-grown brands. Companies like Graphcore, Stability AI, Wayve, FiveAI, and my own – collectively worth an estimated £10 billion – are testament to the UK's potential," he said.

He feels the focus on generative AI and specifically "frontier models" means the UK is "unprepared for the broader implications of AI on society".

"UK AI businesses must have a voice in shaping their own future," added Mr Duggal.

Perhaps most telling was something the UK's technology secretary Michelle Donelan said on Thursday afternoon: “We are interested in applying the solutions, once we fully understand the problems. Are we ruling out legislation? Absolutely not.”

However, the UK does seem to want AI innovation, as last weekend Mr Sunak announced £100m in investment to accelerate the use of AI in healthcare and life sciences, in order to speed up transformational breakthroughs in treatments for previously incurable diseases.

"We can’t afford to be scared of AI, as we simply can’t solve humanity’s biggest challenges, like climate change and the biodiversity crisis, without embracing safe and effective AI," said Will Cavendish, former DeepMind and UK Government advisor, who is now global digital leader at sustainable development consultancy Arup.

"When examining regulation, attendees at the summit must remember to consider an ‘ethic of use’ – where we have an obligation to use technology that will help humanity – rather than only an ethic of ‘do no harm’.”

Bias and disinformation run wild

One key message that kept being mentioned by delegates at the AI Safety Summit over the two days was that tackling disinformation and preventing AI from being used to make the problem worse is a serious concern.

"Whilst further meetings are already planned for 2024, there are also immediate risks of AI, particularly in the context of forthcoming elections, where AI generated text, and 'deep fakes', have the potential to disrupt the democratic process," said Anthony Cohn, Professor of Automated Reasoning at the University of Leeds and Foundational Models Theme lead at the Alan Turing Institute.

He added that there was an "urgent need" to educate the public on the risks and limitations of AI and to be wary of believing everything they see or read in the unregulated media.

"Moreover, anyone using an AI system, like a chatbot, should clearly be made aware that they are engaging with an AI, not a human," said Prof Cohn.

"These short-term risks, which also include problems deriving from biased training data and resulting biased outputs are more urgent to address than possible, but still distant, risks relating to the conceivable creation of AGI which the declaration is more focused on."

Nobody knows how to safely test AI

At the end of the second day of the AI Safety Summit, the UK announced that the 28 nations had signed a plan to safely test AI models. And yet it is unclear what that means in reality.

No government in the world currently knows how to safely test AI models, so how will this plan work? (Toby Melville / Getty Images)
No government in the world currently knows how to safely test AI models, so how will this plan work? (Toby Melville / Getty Images)

“Having governments involved in the testing of models sounds like a good idea in order to avoid the companies marking their own homework but it does raise the obvious questions around enforceability – how would that government involvement be implemented?" asked Sarah Pearce, a data privacy lawyer who heads up Hunton Andrews Kurth’s data-privacy practice in London.

She added that lawmakers would need to look at dividing up the AI models into certain categories, such as jurisdiction, and that increased funding would likely be needed. Post-pandemic and amid worsening global economic conditions, it might be difficult to find the money to pay for this.

Elena Simperl, a Professor of Computer Science at King’s College London, is concerned that there is a lack of standardisation in the AI industry, many varying practices but also many different people behind these AI models, who can be prone to bias.

“We have seen many frameworks and approaches to audit AI systems at design, development and deployment time. Yet studies with AI engineers or compliance teams or decision makers using AI show that there is a lot of ambiguity in how to operationalise them and limited support in the development environments, tools, and processes that these professionals use," she said.

“AI systems make mistakes in a different way than people do. We need to get used to a world where prompting, AI copilots, and people co-auditing AI outputs will become the norm."

We're all going to lose our jobs

This statement by Elon Musk's during his fireside chat with Rishi Sunak sounds stark, but he was very clear about what he thinks the future holds for the world of work. Here's a clue: it's not pretty.

"I think we are seeing the most destructive force in history here. We will one day have AI as smart as a human. It's hard to say when where that will be, but there will come a time where there are no jobs — the AI would be able to do everything," he told Mr Sunak at the event on Thursday night.

Even so, Mr Musk caveated that he felt the standard of living across society would be vastly higher. He didn't specify whether this would apply to developing countries as well.

"One of the challenges of the future will be how do we find meaning in life? We won't have universal basic income... we will have universal high income," he added.

Ms Donelan, the UK's technology secretary, was also keen to assauge doubts from reporters at Bletchley Park during the AI Safety Summit: "I really do think we need to change the conversation when it comes to jobs … what AI has the potential to do is actually reduce some of those tedious administrative part of our jobs, which is particularly impactful for doctors, our police force, our teachers.”

Mr Musk talked-up the benefits of implementing more AI into the workplace: "There are a lot of jobs that are uncomfortable, or dangerous, or tedious and the computer will have no problem coping with that."

And so long as the bots don't complain about the tedium of their new roles, then who are we to feel otherwise? The real issues arise if they grow sentient and tire of acting as our underlings.