UK Prime Minster Rishi Sunak is questioned [about AI safety] by the Liaison Committee of Parliament. 04 JULY 2023.
TRANSCRIPT (Youtube): Greg Clarke on Science and Technology and AI.
CLARKE. Thank you sir good afternoon prime minister. Prime minister a few days ago you told London Tech week that the possibilities of AI are extraordinary. What do you see as the biggest positive impact of artificial intelligence?
PM SUNAK. Well I think just in the last few weeks in the news reporting you know you’ve seen some of the promise of AI paralyzed people being able to walk the the kind of cracking of the structure of almost every known protein and you know starting to better model reactions in nuclear fusion so it just gives you a sense of the variety and that is there and also when you combine that with a computational power of quantum I think you can really start to imagine a future in which of an incurable diseases like cancer and Dementia or new ways to grow crops all might be possible I look for me I’ve kind of put it in a couple of different categories I think straightforwardly on the economy side the interesting report from PWC recently that estimated 10 GDP benefit over the coming decade um you’re starting to see that in the way that companies are using Ai and indeed leading companies like palantir opening that they European AI HQ here and the quality of our research Etc so that’s on the economic side and I’ve spoken in the past about AI holding out the promise of being a general purpose technology and the economic research is it’s very clear that that leads to a J curve effect and productivity that we are on the cusp of hopefully realizing and then in public services I’d say health and education are the two areas where you know I’m excited particularly in healthcare I think reasonably people are reasonably familiar with the opportunity to speed up diagnosis of a range of different conditions but also improve the productivity and accuracy of for example people looking for skin cancers is interesting dumb AI technology that has been deployed or robotic surgery and then in education the opportunity to reduce workload for teachers whether it’s lesson planning or marking but then also provide personalized tuition for children the Khan Academy is starting to roll out an AI chat bot which has I think enormous potential because we know that that more personalized approach to learning which has huge benefits for children particularly disadvantaged children and tutoring in the physical sense is hard to scale but the technology allows us to provide that and I think that would be transformational absolutely
CLARKE. So lots of upsides. But in that speech in London Tech Week you said that we must do AI safely. So what you see is the biggest danger of AI?
PM SUNAK. So I would categorize the risks into a few different buckets I think first you know it’s the socioeconomic risk from the large-scale societal shift that the technology will bring and that’s often what technology does that doesn’t mean you should stand in the way of it but it just means to make sure that we’re cognizant of it and provide people with the skills they need to flourish in a world that is being changed by technology I think the second is the risk of misuse and that is whether it’s open source models or otherwise you know the ability that they well they can be used as tools to generate this information or identify and exploit cyber vulnerabilities or create harmful content or deep fakes or child sexual abuse that’s a kind of second category I think the third question would be tool use whereby Foundation models that can be used to activate capabilities in the real world and then the last category is what people would describe as capability overhangs and that kind of Trends into the area of more existential risk because there’s just a you know a lack of understanding at this point about what the potential of these models might be now now obviously separate to all of that the National Security risk which I’m sure Donald acknowledge I appreciate and limiting what I can say but I think thinking about it in those four categories is probably the right way to to go about it.
CLARKE. Thank you and you’ve convened an international safety Summit uh for the Autumn are you intending that to be a summit of like-minded countries like the US and Japan perhaps or is it for all countries a bit like the international atomic energy Authority that includes countries like China and Russia. What do you have in mind?
PM SUNAK. I think look I think in the first instance a mainly what we’re trying to do is acknowledge that there is a global dimension to this challenge and AI doesn’t respect National borders countries are at varying stages of their thinking about how we can put guard rails in place to make sure that we can realize the benefits of AI in a way that’s safe and secure and it seems a sensible and reasonable step to just try and bring some of those countries together to talk about these problems you know in a collective Forum because ultimately there will need to be some coordination so I think it’s quite early thinking as I said it’s more just about bringing people together – who are countries who’re together – well it’s just bringing people together who are thinking about these things in a in a similar way to exchange ideas share information and as I said because AI doesn’t respect National borders and I think we will all benefit from hearing and talking to each other in a conversation with the businesses themselves and I think that that is really what this is about I think we’re a lot we’re a long way from anyone establishing a IAEA equivalent for AI those things are you know along into the distance but in the first instance just talking through this with like-minded countries.
CLARKE. So you’ve got the summit you published a white paper in March um the the last session of this Parliament begins in November will there be do you expect there to be an AI bill in The King’s Speech?
PM SUNAK. You know I’m not going to comment now on Fourth session legislation all those conversations are happening and no final decisions of being made I think actually what we need to do and I think probably we can do lots of this without legislation is sitting down and figuring out what are the safety features in guardrails that we’d like to put in place and I think we have a sense of what some of those might be whether it’s watermarking whether it’s pre-release information sharing whether it’s reporting a licensing of large training runs um whether it’s pre-training or pre-deployment disclosure transparency on the data sets that are used I think that’s the kind of category of things that we need to do to spend time on which is why the foundation model task force has been set up and funded specifically in its early phase to focus on Safety Research I think it’s an area where the UK can lead because we already have good capability in it so if we can develop that capability further because it’s a growing field it will it will mean that we will know what to do and it may well be and actually so far the conversations with the three leading Foundation model companies have been incredibly Cooperative.
CLARKE. But they’re calling for legislation.
PM SUNAK. Actually well I think what they want are for governments to put in place the the uh the guard rails and they’re very open about that and they’re very I say thus far they have been incredibly constructive and open and transparent with government and want to try and find a solution together which I said it’s too early to prevent what all that might look like but you can imagine a world where at least the initial stage that don’t require legislation necessarily but actually just require us to get in there and do safety evaluation on the models or have access to them and I’m I said we’ve had announcements already from some of those Foundation model companies that they will provide that access to the government to be able to do that on a on a priority basis which I’m grateful for and I think will help us get this right.