Global AI Safety Summit, can AI be effectively regulated?
Neil Lawrence, DeepMind Professor of Machine Learning at the University of Cambridge talk with CGTN Europe on the Global AI Safety Summit.
Category
🎵
MusicTranscript
00:00 Neil Lawrence, Professor of Machine Learning at Cambridge University. Neil, talking shop
00:05 or useful summit, which is it?
00:10 I think it's incredibly significant to have the United States and China signing the same
00:15 document and I don't think the significance of that can be underestimated. It may not
00:21 be that there's an enormous amount to say in that document, but in order to get both
00:27 those great nations to sign such a document, that's always going to have to be the case.
00:31 So we'll look back on this moment, but we'll be looking back on it from whatever it leads
00:37 to and I think it will be seen as a significant turn where we get these two great countries
00:41 talking again.
00:42 I can't do artificial intelligence, but any intelligence observer would suggest that we
00:47 can't even regulate the internet. I mean, you can't regulate AI, can you?
00:52 I think you point really well made, Jamie. I think one of the things that is unfortunate
00:57 about the summit is it's not addressing the real issues that we're facing in our different
01:01 societies with our different approaches to regulating them. Artificial intelligence is
01:06 just an extension of internet. It's just an extension of this rapid form of communications,
01:11 which is massively disruptive to the way we run our societies. And there are benefits
01:16 accruing to certain people and it's the reactions that our societies take institutionally to
01:23 that take a long time and it's incredibly difficult to regulate that. But that's not
01:26 just AI, that's the whole of cyber because of the speed of deployment of these new capabilities.
01:31 As you say, some big players there, China's there, the United States is there, the UK
01:35 is there, the EU is there, but I mean, there are lots of people who are not there. Can
01:39 this gathering and the later ones make any progress if there isn't a complete buy-in
01:46 from everyone around the world?
01:49 I think it's an incredibly good point. I think that there's a real danger for the developed
01:54 world to be navel-gazing here. So the issues that we might feel in professionalized societies
02:00 where we have large numbers of lawyers and accountants whose roles they feel are threatened
02:03 by these technologies are extremely different to the issues that developing economies are
02:08 facing. And I think many of them are looking on at this quite bemused in terms of the nature
02:13 of the conversation we're having when these technologies prevent enormous opportunities,
02:17 but enormous dangers for their societies. The potential for misinformation to disrupt
02:22 delicate societies is absolutely tremendous. And many people have already died because
02:27 of that. And I think that it would be worthwhile having more of that conversation as well as
02:32 the conversation on these frontier risks.
02:35 Where and how is AI already being used safely, if I can put it like that?
02:41 Well, it's quite pervasively used. I mean, in terms of recommendations on the internet,
02:48 it's filtering a lot of what we see. Now, whether it's being used safely, I think, in
02:52 that context is now the question, because although those are quite inconsequential decisions,
02:56 what adverts we see, what social media posts we see, we see they have a downstream effect
03:00 on creating divisions in society where certain political groups will only look at their own
03:06 posts, whereas in the past, they may have been presented more information from a wider
03:10 spectrum of media.
03:11 I think one really interesting domain where AI is progressing slowly but has enormous
03:16 potential for benefit is in medicine, where, of course, we already have an enormous amount
03:19 of regulation about what a medical device consists of and what that should do before
03:24 it can be safely deployed. So there are lots of domains where we've got lots of good regulation,
03:29 and that's already being used in collaboration with people who are deploying these technologies.
03:33 But naturally, those things move a bit slower because to conform to the regulation requires
03:37 a lot of work.
03:38 Neil, I'm sure we'll talk again in coming weeks and months. For the moment, thank you
03:41 very much. Neil Lawrence, Professor of Machine Learning at Cambridge University.