What does the success of the second global AI summit look like? As the industry’s great and good (and I) gathered last week at the Korea Institute of Science and Technology, a sprawling hilltop campus in eastern Seoul, that was the question I kept asking myself.
If we rank the event by the number of ads generated, then it is a great success. In less than 24 hours – starting with a virtual “leaders summit” at 8 p.m. and ending with a joint press conference with the science and technology ministers of South Korea and Britain – I counted no less than six agreements , pacts, promises and declarations, all of them demonstrating the success of the event in getting people to the table to reach an agreement.
There were Frontier AI’s security commitments:
The first 16 companies have signed up to voluntary AI security standards unveiled at the Bletchley Park summit, Rishi Sunak said on the eve of the follow-up event in Seoul.
“These commitments ensure that the world’s leading AI companies provide transparency and accountability in their plans to develop safe AI,” Sunak said. “It sets a precedent for global standards on AI safety that will unlock the benefits of this transformative technology.”
Seoul’s (deep breath) Statement of Intent Towards International Cooperation in AI Safety Science:
These institutes will begin to share information about models, their limitations, capabilities and risks, as well as monitor specific “AI security incidents and damages” where they occur and share resources to advance the global understanding of AI security science. AI.
At the first “full” meeting of those countries on Wednesday, (Michelle Donelanthe UK technology secretary) He warned that the creation of the network was only a first step. “We should not rest on our laurels. As the pace of AI development accelerates, we must match that speed with our own efforts if we are to address the risks and seize the limitless opportunities for our audiences.”
He Seoul Ministerial Declaration:
Twenty-seven countries, including the United Kingdom, the Republic of Korea, France, the United States, the United Arab Emirates and the European Union, have committed to developing proposals to assess the risks of AI in the coming months, in a series of agreements . What AI Seoul brings summit until its end. The Seoul Ministerial Declaration sees countries agree for the first time to develop shared risk thresholds for the development and deployment of frontier AI, including agreement on when model capabilities could pose “serious risks”without adequate mitigations. This could include helping malicious actors acquire or use chemical or biological weapons, and the ability of AI to evade human oversight, for example through manipulation and deception or autonomous replication and adaptation.
There was also the Seoul Declaration for Safe, Innovative and Inclusive AIwhich described the common ground on which 11 participating nations and the EU agreed to proceed, and the Seoul Declaration of Intent Towards International Cooperation in AI Safety Sciencewhich presented a rough idea of what the objectives really were.
And the Seoul AI Business Commitmentin which 14 companies (only partially overlapping with the 16 companies that had previously signed up to Frontier’s AI Safety Commitments) “committed to efforts for the responsible development, advancement, and benefit sharing of AI” .
Call to action
It’s understandable if your eyes glaze over. I’m confused and I was there.
The two hosts of the event do not help solve the problem. For the UK, the Frontier AI Security Commitments were the big ticket, announced with a comment from Rishi Sunak and combined with the offer of an interview with the technology press secretary in Seoul. In the English-language press, the Seoul AI Business Commitment was virtually ignored, despite being an important element of the South Korean delegation’s achievements. (My guess, for what it’s worth, is that the focus on “frontier AI” in the first set of commitments slightly let down South Korea’s tech sector, including, as it did, only Samsung and Naver; the commitment from Seoul, on the other hand, there were six national companies and eight international ones.)
Perhaps it is easier to explain everything by dividing it into groups. There are the companies’ two competing commitments, each detailing a slightly different voluntary code they intend to follow; there are the three overlapping statements from the nations, laying out what they really wanted to get out of the AI summit and how they will get there over the next six months; And there’s the National AI Safety Institutes’ concrete action plan, detailing how and when they will work together to understand more about the cutting-edge technology they’re trying to examine.
after newsletter promotion
There is an obvious objection here, and that is that none of these agreements have any teeth, or even enough details to identify whether someone is trying to follow them or not. “It is problematic for companies to determine what is safe and what is dangerous and voluntarily choose what to do about it,” Francine Bennett, acting director of the Ada Lovelace Institute, told me.
Similarly, it is strange to consider the agreements the success of the summit when they were largely set in stone before the delegates even arrived in South Korea. At times, as agreements and communications continued to arrive in my inbox with little relation to the events happening in person, it felt like a summit that could have been an email.
Just to be invited
The truth is slightly different: the success of the summit is that it happened.
That sounds like faint praise, but it’s true for any event like this. Yes, all of the summit’s key agreements were hammered out before it began, but by offering a strict deadline and a stage from which to announce success, the event gave everyone motivation to sign up.
And while the agreements are certainly ineffective, they are also a starting point for the real work that will need to be done once the summit ends. Having companies and governments sign a shared description of reality is the first step in having the difficult technical conversations necessary to solve problems. “We all want the same thing here” is a powerful statement when it is true. And, when the boundaries are drawn carefully enough, it still is.