WEBVTT - Google EMEA Boss Talks AI, Productivity and UK Riots 0:00:03.160 --> 0:00:07.480 Bloomberg Audio Studios, Podcasts, Radio News. 0:00:14.640 --> 0:00:17.400 Welcome to the City of London, the City of the City, 0:00:17.480 --> 0:00:23.360 the City of London. Please mind the gap between the 0:00:24.079 --> 0:00:33.080 and the financial hearts of the country, the city, the City. 0:00:33.479 --> 0:00:35.640 Welcome to in the City. 0:00:36.960 --> 0:00:42.360 Stand clear of the doors. Welcome to in the City. 0:00:42.360 --> 0:00:45.000 A podcast from Bloomberg about the story is important to 0:00:45.080 --> 0:00:47.839 the City of London. I'm Francin Laqua. This week a 0:00:47.920 --> 0:00:51.280 discussion of Europe's productivity problem and whether AI is perceived 0:00:51.400 --> 0:00:54.600 as a silver bullet or a threat. We're joined by 0:00:54.640 --> 0:00:57.360 Matt Britain, President of Google EMA. Matt, thank you so 0:00:57.440 --> 0:00:59.400 much for joining us. I mean, you've ever seen Google's 0:00:59.400 --> 0:01:01.760 business and opera in the region for ten years now, 0:01:02.240 --> 0:01:05.160 having joined Google in two thousand and seven. Before that, 0:01:05.200 --> 0:01:08.399 you worked in newspapers, technology and even road for TEAMGB 0:01:09.280 --> 0:01:12.200 at the nineteen ID eight Soul Olympics. So we need 0:01:12.600 --> 0:01:15.039 to talk about the Olympics. But you know, there's so 0:01:15.160 --> 0:01:19.640 much about AI and people understand it. People are excited, 0:01:19.720 --> 0:01:23.800 people are fearful. Where are we in this AI race? 0:01:24.080 --> 0:01:26.160 Thanks for having me, and I suppose it's been my 0:01:26.640 --> 0:01:29.960 joy and privilege to see how people harness technology over 0:01:29.959 --> 0:01:32.840 the last twenty to thirty years, and I think, you know, 0:01:33.080 --> 0:01:36.640 through that period, I think we're in the third big shift. 0:01:36.520 --> 0:01:38.319 At first, I think was the birth of the Internet. 0:01:38.360 --> 0:01:41.440 So I was at university and the Internet application you 0:01:41.440 --> 0:01:44.240 could use was email, and it wasn't mind blowing because 0:01:44.240 --> 0:01:46.479 there was somebody sitting next to me sending me the email. 0:01:46.520 --> 0:01:48.440 That was where it going to come from. And then 0:01:48.640 --> 0:01:50.240 when I started in the world of work, we saw 0:01:50.280 --> 0:01:52.560 the birth of the World Wide Web, which is actually 0:01:52.640 --> 0:01:56.320 invented as I was leaving university. And obviously I've worked 0:01:56.360 --> 0:01:59.160 for twenty years in a company in roles that have 0:01:59.240 --> 0:02:03.320 depended upon that innovation, but that wasn't earth shattering until 0:02:03.320 --> 0:02:05.840 it became accessible to everyone. And Google's played its part 0:02:05.920 --> 0:02:09.480 through Android and through Chrome and making it easy for 0:02:09.520 --> 0:02:11.240 people to use the web. And now we've got this 0:02:11.320 --> 0:02:13.440 AI revolution. And I suppose the way I think about 0:02:13.480 --> 0:02:18.840 AI is it's about spotting patterns, making predictions and learning. 0:02:19.080 --> 0:02:20.720 AI is not new. We've been working on it for 0:02:20.760 --> 0:02:23.720 about ten years. And Google Translate, my favorite product, is 0:02:23.760 --> 0:02:26.920 a great example of that, really understanding language well. But 0:02:27.000 --> 0:02:29.480 what is new, I think is the emergence of chatbots 0:02:29.520 --> 0:02:31.760 and people thinking, oh AI and chatbots are going to 0:02:31.800 --> 0:02:34.200 change everything, and I think it's a much broader based 0:02:34.200 --> 0:02:37.160 technology than that. So we don't yet know how it's 0:02:37.160 --> 0:02:39.160 going to be harnessed, but we can begin to see 0:02:39.440 --> 0:02:42.000 some of the opportunities and some of the risks of AI. 0:02:42.360 --> 0:02:45.040 So, Matt, one of the industries where actually, you know, 0:02:45.240 --> 0:02:48.119 AI has been applied for good is everything to do 0:02:48.880 --> 0:02:52.720 with health and protein folding, and that leads to many 0:02:52.720 --> 0:02:54.560 more discoveries like DeepMind. 0:02:54.880 --> 0:02:57.160 Yeah, so this is one of the things I think 0:02:57.200 --> 0:03:00.560 is most exciting is the scientific discovery impossibility. So there 0:03:00.560 --> 0:03:03.200 were one hundred and seventy five thousand proteins. They're the 0:03:03.240 --> 0:03:06.520 building blocks of life, they're essential for drug and disease research, 0:03:06.960 --> 0:03:09.480 one hundred and seventy five thousand of them, where researchers 0:03:09.560 --> 0:03:13.200 are painstakingly through observation come up with a three D structure. 0:03:13.919 --> 0:03:16.640 And deep Minds saw the opportunity to use AI that 0:03:16.680 --> 0:03:20.480 spots patterns, makes predictions and learns to uncover the three 0:03:20.560 --> 0:03:22.440 D structure of all the proteins, and in a matter 0:03:22.480 --> 0:03:24.200 of months they did just that. So we went from 0:03:24.240 --> 0:03:26.720 one hundred and seventy five thousand to two hundred and 0:03:26.720 --> 0:03:30.920 fourteen million and we chose to make that available for 0:03:30.919 --> 0:03:34.160 free in a database so that anyone researching a disease 0:03:34.240 --> 0:03:37.360 or looking at drug discovery can use it. And today 0:03:37.440 --> 0:03:42.400 two million people are using the database of proteins in 0:03:42.440 --> 0:03:43.880 their day to day work, and I think that means 0:03:43.920 --> 0:03:47.880 we'll see faster disease research, drug discovery, and breakthroughs that 0:03:47.920 --> 0:03:50.680 are for the general good. That's something we made available 0:03:50.720 --> 0:03:52.560 for free, and I feel proud of that because it's 0:03:52.600 --> 0:03:55.200 built on the back of all that painstaking research. We 0:03:55.240 --> 0:03:58.280 also have Isomorphic Labs, which is a commercial venture, and 0:03:58.320 --> 0:04:01.160 we're working in partnerships with Novarta and other drug companies 0:04:01.880 --> 0:04:05.600 to look at commercial opportunities to bring those discoveries to market. 0:04:06.680 --> 0:04:08.880 Mattter, how difficult is it to decide what it goes 0:04:08.920 --> 0:04:10.960 out in the open world so that everyone can use 0:04:11.360 --> 0:04:14.000 the underlying models, and how much you want to keep 0:04:14.040 --> 0:04:15.440 it at it for Google? 0:04:15.800 --> 0:04:17.480 Yeah, So what we did with Alpha fold and now 0:04:17.480 --> 0:04:19.520 I'm proud of this, you know, and Google's history is 0:04:19.560 --> 0:04:22.400 always about trying to be open about innovation. When you 0:04:22.400 --> 0:04:25.000 think about search, it's connecting to your website, not ours. 0:04:25.560 --> 0:04:27.200 What we did was we said, well, you know, we've 0:04:27.200 --> 0:04:29.239 built this on the hard work of all those people 0:04:29.520 --> 0:04:32.360 who got to the one hundred and seventy five thousand proteins, 0:04:32.680 --> 0:04:35.920 and it's too important a discovery to keep behind a paywall, 0:04:36.160 --> 0:04:38.560 so we made it available to everyone. Now. We also 0:04:38.600 --> 0:04:41.680 have a commercial operation called Isomorphic Labs that's looking, as 0:04:41.720 --> 0:04:45.359 others are to sort of exploit those those databases to 0:04:45.360 --> 0:04:48.320 try to find commercial opportunities, and we're working with Novartis 0:04:48.400 --> 0:04:51.000 and other farmer companies to do that because obviously you're 0:04:51.000 --> 0:04:53.240 going to need to commercialize these things too. So I 0:04:53.240 --> 0:04:56.560 think that's a smart but generous way of operating in 0:04:56.600 --> 0:04:57.600 this space, and that. 0:04:57.640 --> 0:05:01.000 Of course changes vaccines, that changes the way we do trials. 0:05:01.480 --> 0:05:03.400 Is there a part Is there an industry that's not 0:05:03.440 --> 0:05:06.160 been disrupted yet by AI that you think? Well? 0:05:06.600 --> 0:05:08.680 I mean from sine, I think we're at an incredibly 0:05:08.800 --> 0:05:11.520 early stage here, And you know, I talk a lot 0:05:11.560 --> 0:05:13.880 of business leaders as you do, and to governments, and 0:05:14.760 --> 0:05:17.120 I think last year was like the wow, you know, 0:05:17.240 --> 0:05:19.200 oh my god, this is amazing. Have you tried this yet? 0:05:19.200 --> 0:05:20.960 Et cetera. People could see for the first time in 0:05:21.000 --> 0:05:24.360 their practical ways, this is quite capable. But now we're 0:05:24.400 --> 0:05:27.599 into the how, and we're relatively short of concrete case 0:05:27.640 --> 0:05:32.000 studies yet of the genuine positive impact all the risks 0:05:32.000 --> 0:05:34.200 that can come right, but we're starting to see them 0:05:34.279 --> 0:05:37.279 so in an interesting area. So you know, one example 0:05:37.320 --> 0:05:39.560 that many of your listeners will sort of be aware of. 0:05:39.880 --> 0:05:41.960 If you think about fraud and money laundering in the 0:05:41.960 --> 0:05:44.920 banking sector, it's something like a two trillion dollar problem globally. 0:05:45.440 --> 0:05:49.520 And our team worked with HSBC spotting patterns, making predictions 0:05:49.520 --> 0:05:54.560 and learning to identify patents that could be fraudulent transactions, 0:05:54.800 --> 0:05:58.120 and they were able to massively reduce false positives but 0:05:58.160 --> 0:06:02.600 massively increase the number of potential transactions to look at. 0:06:02.600 --> 0:06:05.040 And so that's a good example of applying this technology 0:06:05.040 --> 0:06:07.560 in a very practical way that can create value. 0:06:07.720 --> 0:06:09.600 When you speak to a lot of the banking chief executives, 0:06:09.600 --> 0:06:12.440 they'll say units thanks to us that we spotted a pattern, 0:06:12.480 --> 0:06:15.760 for example, on the algorithm that led to a stopping 0:06:15.800 --> 0:06:17.880 a fraudulent case trying to get access money in a 0:06:17.920 --> 0:06:22.880 central bank, and the other that is great. It also 0:06:22.960 --> 0:06:26.200 means that you can replicate a voice pattern to access 0:06:26.240 --> 0:06:27.320 money when it's not needed. 0:06:28.000 --> 0:06:30.760 Technology is a tool and it can be used for 0:06:30.839 --> 0:06:33.160 good and it can be used for ill and I 0:06:33.160 --> 0:06:36.240 think we need to be really cognizant of that when 0:06:36.279 --> 0:06:38.200 we're designing and building things, to try to sort of 0:06:38.240 --> 0:06:40.200 factor in things that make it harder to use it 0:06:40.240 --> 0:06:43.240 for ill. But also we need rules and regulations, and 0:06:43.279 --> 0:06:46.159 that's why things like the AI Act in Europe are important. 0:06:46.160 --> 0:06:48.840 It's too important not to regulate AI and it needs 0:06:48.880 --> 0:06:51.440 to be regulated well, and that means sort of being dynamic, 0:06:51.760 --> 0:06:53.480 you know. The way to think about regulating AI is 0:06:53.520 --> 0:06:55.560 to think about the benefits that are possible and then 0:06:55.600 --> 0:06:58.960 how do we protect against the risks. And another example 0:06:59.000 --> 0:07:01.840 we're working on is breast cancer screening. Right, so one 0:07:01.839 --> 0:07:03.280 in seven women in their lives are going to be 0:07:03.320 --> 0:07:07.719 affected by breast cancer, and yet access to mammograms and 0:07:07.920 --> 0:07:13.000 simple diagnostic knostic technology is limited and so spotting patterns, 0:07:13.040 --> 0:07:16.520 making predictions learning. We work with the NHS here in 0:07:16.560 --> 0:07:20.000 the UK to look at thousands of mammograms, train and AI. 0:07:20.240 --> 0:07:23.119 Now the AI can diagnose the science of breast cancer 0:07:23.440 --> 0:07:25.640 as well as a clinician. It does not mean that 0:07:25.680 --> 0:07:28.880 clinicians are irrelevant. Actually, what it means is you come 0:07:28.880 --> 0:07:31.160 in and you get an answer quicker. It means that 0:07:31.200 --> 0:07:33.520 clinicians can help on the edge cases and the people 0:07:33.560 --> 0:07:36.680 need treatment, and so the outcomes for patients and clinicians 0:07:36.680 --> 0:07:39.160 are both better. And I think that's a good and 0:07:39.200 --> 0:07:40.920 we're in the process of looking at how to scale 0:07:40.960 --> 0:07:43.200 and roll that out. That's a good example of how 0:07:43.200 --> 0:07:47.480 I think AI will work. It's humans and technology working 0:07:47.520 --> 0:07:50.280 to get better outcomes as a result of the collaboration 0:07:50.400 --> 0:07:52.760 and partnerships. I think the brand AI isn't a very 0:07:52.760 --> 0:07:55.360 helpful thing. It's a set of tools that are smarter 0:07:55.440 --> 0:07:57.080 and different from some of the tools we've had. 0:07:57.240 --> 0:08:00.480 But is there a danger in Europe regulation at the 0:08:00.520 --> 0:08:04.520 stifles innovation because if you regulate too quickly then you 0:08:04.560 --> 0:08:05.800 don't know where it could go. 0:08:06.320 --> 0:08:09.080 Yeah, I think there's a real danger of that. We 0:08:09.120 --> 0:08:11.600 do need regulation, and the regulation needs to be clear 0:08:11.640 --> 0:08:13.760 on the sort of rules of the road. But if 0:08:13.760 --> 0:08:16.680 you look in Europe. In the EU, over the last 0:08:16.800 --> 0:08:19.640 five years, there have been over one hundred pieces of 0:08:19.680 --> 0:08:23.440 regulation around digital technology one hundred and those pieces of 0:08:23.480 --> 0:08:27.160 regulation typically are implemented in different ways at different speeds 0:08:27.200 --> 0:08:29.920 in all the twenty seven member states. Now think about 0:08:29.920 --> 0:08:32.200 that as a barrier to innovation for a moment, that's 0:08:32.320 --> 0:08:35.800 really hard to contend with and you might say, actually, 0:08:36.640 --> 0:08:40.040 it's something which makes it particularly hard for startups and 0:08:40.120 --> 0:08:44.720 scale ups or European single country companies, whereas a large 0:08:44.800 --> 0:08:48.560 organization like a Google or a Microsoft can absorb some 0:08:48.640 --> 0:08:50.480 of the costs of doing that. But even for us, 0:08:50.559 --> 0:08:54.240 you know, and our competitors, we've had to slow down 0:08:54.240 --> 0:08:56.600 the launchers of some of our AI based product products 0:08:56.800 --> 0:08:59.880 in Europe. So what's the right thing for the European consumer? 0:09:00.280 --> 0:09:02.280 You know, it's interesting Mario Draugi, who's I think really 0:09:02.320 --> 0:09:05.880 well respected, is working on a proposal for the EU 0:09:05.960 --> 0:09:08.840 around competitiveness. So one of the big opportunities of AI, 0:09:08.920 --> 0:09:11.199 I think is can it help us get more people 0:09:11.200 --> 0:09:14.400 into work, help people be more productive and drive growth 0:09:14.400 --> 0:09:15.160 which we all need. 0:09:15.520 --> 0:09:18.080 Is it difficult and too complex to understand what the 0:09:18.200 --> 0:09:20.400 roles of tomorrow will be with the jobs of tomorrow. 0:09:20.600 --> 0:09:23.040 We're in a moment where the headlines about you know, 0:09:23.520 --> 0:09:25.839 what could happen to work almost write themselves, and you know, 0:09:25.920 --> 0:09:27.880 jokingly you might say they can write themselves and you 0:09:27.920 --> 0:09:31.320 just put a prompt in, but nobody really knows. And actually, 0:09:31.440 --> 0:09:33.920 again it's an opportunity for us to seize the moment, 0:09:33.960 --> 0:09:35.400 and so how do we want the world of work 0:09:35.400 --> 0:09:38.160 to be? You know, I think there's lots of drudgery 0:09:38.200 --> 0:09:40.079 work that none of us would really like to do it, 0:09:40.080 --> 0:09:43.199 and our jobs that can go away, you know. And 0:09:43.240 --> 0:09:45.320 there are many technologies that we use today that people 0:09:45.400 --> 0:09:47.200 feared when they came in but actually just made us 0:09:47.280 --> 0:09:50.160 much more productive. So I think for most jobs, the 0:09:50.200 --> 0:09:52.520 most likely outcome is you'll be more productive. You'll be 0:09:52.520 --> 0:09:55.040 able to spend more time on things that only humans 0:09:55.160 --> 0:09:57.480 can do, like what So I think, you know, think 0:09:57.480 --> 0:09:59.960 about your work, you know, think about your work before 0:10:00.120 --> 0:10:02.280 the advent of the internet, and how easy it is 0:10:02.320 --> 0:10:04.760 now for you to research stories, to find leads, to 0:10:04.800 --> 0:10:07.319 reach out to contacts. How do you think about your 0:10:07.320 --> 0:10:09.600 work tomorrow? Well, you know, we have a product called 0:10:09.679 --> 0:10:12.200 Notebook LM where you can put in a whole bunch 0:10:12.280 --> 0:10:16.120 of files that you might be researching, videos, audio, transcripts, 0:10:16.360 --> 0:10:19.120 and then you can ask questions of the AI about 0:10:19.200 --> 0:10:22.720 specifically those those documents and artifacts that you feed in 0:10:22.760 --> 0:10:25.880 and that can really help you synthesize, get summaries and 0:10:25.920 --> 0:10:28.520 so on. So I think, you know, that's a small example, 0:10:28.679 --> 0:10:31.800 but an example of how new, more powerful tools will 0:10:31.840 --> 0:10:34.559 help you to do you do your work differently. 0:10:34.240 --> 0:10:36.240 Matt when you look at, you know, some of the 0:10:36.240 --> 0:10:38.480 public services. I keep on being told by heads of 0:10:38.480 --> 0:10:41.200 the departments that actually their computers don't even match, you know, 0:10:41.200 --> 0:10:43.959 and this is not necessarily NHS, this is other parts 0:10:44.000 --> 0:10:46.600 where they should match. So is there a danger that 0:10:46.679 --> 0:10:50.080 in an already divided society where some people have a 0:10:50.080 --> 0:10:53.680 lot and others don't, that this you know, social divide 0:10:53.760 --> 0:10:56.480 will grow even bigger with AI if we don't think 0:10:56.520 --> 0:10:58.360 about it holistically. I guess yeah. 0:10:58.360 --> 0:11:00.640 One of the things that most exercises is, you know, 0:11:00.840 --> 0:11:04.960 over the long term, do the divisions within countries and 0:11:05.000 --> 0:11:08.760 across countries widen? And I think technology historically has actually 0:11:08.760 --> 0:11:10.719 helped us if you look at the long view, has 0:11:10.760 --> 0:11:13.959 helped us to sort of see millions, billions of people 0:11:13.960 --> 0:11:17.120 lifted out of poverty. We're at an all time high, 0:11:17.160 --> 0:11:19.120 I think in terms of women's education and so on 0:11:19.160 --> 0:11:21.800 and so forth. So sometimes the big picture misses the small. 0:11:21.800 --> 0:11:24.440 But you're absolutely right, we need to avoid it haves 0:11:24.480 --> 0:11:27.840 and haves, not situation with technology. What's kept me working 0:11:27.840 --> 0:11:29.480 at Google for so long is, you know, we're all 0:11:29.520 --> 0:11:32.800 about making technology work for everyone. You know, whether you 0:11:32.840 --> 0:11:36.200 can type or spell, doesn't matter, whatever language you speak 0:11:36.240 --> 0:11:38.760 doesn't matter. We want it to work for you. And 0:11:38.800 --> 0:11:40.680 I think with AI it's the same. We need to 0:11:40.679 --> 0:11:42.920 really think about how do we harness technology for everyone. 0:11:43.360 --> 0:11:45.960 What the opportunity is. Well, you know, today somebody without 0:11:45.960 --> 0:11:49.160 a formal education can now write almost as well as you, 0:11:49.200 --> 0:11:51.480 and I could probably not quite as well as you, 0:11:51.520 --> 0:11:53.000 but as well as I could much better. That's a 0:11:53.040 --> 0:11:55.880 huge that's a huge opportunity. I can communicate with somebody, 0:11:55.920 --> 0:11:57.640 you know in Thailand and I don't speak type, but 0:11:57.679 --> 0:12:00.160 I can communicate in pretty good tie now thanks to 0:12:00.200 --> 0:12:02.800 these tools. So you think about that. I actually think 0:12:02.800 --> 0:12:05.360 we're going to see an opportunity for many people who 0:12:05.400 --> 0:12:08.880 don't have access to the formal education sector to participate 0:12:08.960 --> 0:12:10.640 much more fully in the world of work. 0:12:10.960 --> 0:12:13.840 Whose job is it to think about retraining? Is it 0:12:13.880 --> 0:12:19.079 private companies, is it actors like you and other AI proponents, 0:12:19.200 --> 0:12:20.120 or is it governments? 0:12:20.280 --> 0:12:22.280 I always think about three words when I think about this. 0:12:22.480 --> 0:12:25.680 It's bold, responsible, and together. So bold is about making 0:12:25.720 --> 0:12:28.200 sure we pursue the biggest benefits of innovations, and we've 0:12:28.200 --> 0:12:30.680 talked to some of them. Responsible is about from the 0:12:30.720 --> 0:12:34.400 start thinking about misuse mistakes and how we kind of 0:12:34.400 --> 0:12:36.760 design that out and create the rules of the road. 0:12:37.240 --> 0:12:39.640 But then together, and I think that's the most important bit. 0:12:39.679 --> 0:12:41.640 You know, when you look at these sort of scientific breakthroughs, 0:12:41.679 --> 0:12:44.400 it's not our job, and it can't be fully government's job, 0:12:44.440 --> 0:12:47.120 and it's not the individual's job. It's a collaboration. So 0:12:47.160 --> 0:12:50.680 you need government and companies and communities to work together 0:12:51.000 --> 0:12:53.600 to harness technology for good, to decide what the rules 0:12:53.600 --> 0:12:55.760 of the road are. You know, governments have to regulate. 0:12:55.880 --> 0:12:58.640 That tends to be a slower process, but principles and 0:12:58.679 --> 0:13:01.720 responsibilities can be and there companies come up with codes 0:13:01.720 --> 0:13:04.040 of practice and can be scrutinized, and that's an important 0:13:04.080 --> 0:13:07.160 thing there. But then communities actually are the places where 0:13:07.160 --> 0:13:09.480 you really want to define what is and isn't acceptable, 0:13:09.480 --> 0:13:11.520 and those things can vary across countries. 0:13:11.559 --> 0:13:14.360 As we know, we've seen some pretty extraordinary pictures on 0:13:14.520 --> 0:13:17.880 the riots in the UK. How far does the responsibility 0:13:17.920 --> 0:13:20.079 go for some of the tech platforms, some of the 0:13:20.120 --> 0:13:23.960 social media actors to keep that in check with disinformation. 0:13:24.559 --> 0:13:26.280 I mean, at first I want to say that Hart 0:13:26.360 --> 0:13:28.960 goes out to families in the community in Southport who 0:13:29.000 --> 0:13:32.800 are affected by that horrible attack at the start of this, 0:13:33.240 --> 0:13:36.400 and then we've seen this series of riots across the countries, 0:13:36.640 --> 0:13:39.880 sort of somewhat disconnected from the reality of that. Of course, 0:13:39.960 --> 0:13:41.720 you know, I think the real world and the internet, 0:13:42.000 --> 0:13:43.560 or the real world and the digital world are the 0:13:43.600 --> 0:13:48.000 same place and the same rules should apply, and the 0:13:48.080 --> 0:13:51.000 role of media, both social and traditional media in this 0:13:51.040 --> 0:13:53.680 is something we should be looking at. How do we 0:13:53.720 --> 0:13:57.199 deal with this? Google, I suppose you know two things. One, 0:13:57.600 --> 0:13:59.839 Search is often the place that people come to try 0:13:59.880 --> 0:14:01.839 and find out the truth, and that's why we work 0:14:01.920 --> 0:14:05.440 with news organizations around the world on search and on YouTube, 0:14:05.440 --> 0:14:07.640 so that when you ask about news, you're pointed to 0:14:07.679 --> 0:14:10.280 credible news sources. Now there's a difference of views across 0:14:10.280 --> 0:14:12.560 those news sources, but you can find a selection and 0:14:12.600 --> 0:14:15.080 that's a key role that we play, and that's why 0:14:15.080 --> 0:14:18.800 we work so closely with news organizations. Second thing for us, though, 0:14:18.840 --> 0:14:21.480 is to have policies that are clear about what we 0:14:21.720 --> 0:14:23.800 will and will not allow on our platforms. On something 0:14:23.840 --> 0:14:26.800 like hate speech. It's not easy, so people would say, 0:14:26.840 --> 0:14:28.960 you know, on YouTube, you know, you seem to be 0:14:28.960 --> 0:14:31.640 pretty good at keeping it clear of pornography, but actually 0:14:31.680 --> 0:14:33.720 if you ask people in the street, you have pretty 0:14:33.760 --> 0:14:36.320 clear agreement on what was and was not pornographic. But 0:14:36.400 --> 0:14:39.120 on something like hate speech or incitement of violence, there 0:14:39.120 --> 0:14:41.640 are lots and lots of shades of gray. And about 0:14:41.640 --> 0:14:44.120 ten years ago we had a real challenge on YouTube 0:14:44.120 --> 0:14:46.840 with videos and across the web, but YouTube as well, 0:14:47.200 --> 0:14:51.440 videos that were about sort of inciting extremist behavior, hate 0:14:51.440 --> 0:14:53.760 speech and so on. And we work with one hundred 0:14:53.760 --> 0:14:57.480 and fifty expert organizations to come up with policies which 0:14:57.480 --> 0:15:01.240 we then had people classify video. We then use the 0:15:01.320 --> 0:15:05.600 classified videos to train AI and then we published transparently 0:15:05.640 --> 0:15:08.080 how successful we are. So I can say that today 0:15:08.240 --> 0:15:12.040 over ninety percent of videos that violate our violent extremism 0:15:12.120 --> 0:15:15.080 policies never get seen by a single human, and then 0:15:15.080 --> 0:15:16.920 the rest of them we can take action very quickly. 0:15:16.960 --> 0:15:19.880 So in the case of the UK, our teams are 0:15:20.160 --> 0:15:23.840 able to move quickly to remove violative videos and channels 0:15:23.880 --> 0:15:24.360 than they've done. 0:15:24.400 --> 0:15:27.120 So how difficult is it to actually know where this 0:15:27.160 --> 0:15:29.960 piece of disinformation comes from? If it's from a state actor, 0:15:30.520 --> 0:15:33.760 is it more difficult for Google to channel than if 0:15:33.760 --> 0:15:34.640 it's from an individual? 0:15:34.760 --> 0:15:36.600 Yeah, So if a zoom out from the UK situation 0:15:36.680 --> 0:15:39.000 for a minute. What we see is, yes, there are 0:15:39.400 --> 0:15:41.960 an array of actors who try to intervene in the 0:15:42.000 --> 0:15:45.560 democratic process for example. So you know, we have a 0:15:45.760 --> 0:15:49.800 team that looks after election integrity across our platforms and 0:15:49.880 --> 0:15:53.480 in the recent EU elections, they're absolutely working closely with 0:15:53.520 --> 0:15:56.200 the election authorities training people on how to use the tools, 0:15:56.480 --> 0:15:59.040 et cetera. And we have a threat analysis group that 0:15:59.080 --> 0:16:02.760 shares with governments and key agencies. What we're seeing because 0:16:02.960 --> 0:16:04.880 we've got lots of popular products, so we see lots 0:16:04.920 --> 0:16:07.160 of attacks. Well, I can say is in the EU 0:16:07.240 --> 0:16:11.280 elections actually it was sort of more old style attacks. 0:16:11.400 --> 0:16:14.200 I think Microsoft suffered some attacks on accounts for example, 0:16:14.800 --> 0:16:16.920 rather than that was feared. You know a lot of 0:16:17.080 --> 0:16:20.920 fake AI generated imagery. But just as we have tools 0:16:20.920 --> 0:16:25.280 that people can use to generate with AI fakery, the 0:16:25.360 --> 0:16:28.480 technology we've got to be able to detect it is accelerating. 0:16:28.520 --> 0:16:30.240 And so we've been using AI on YouTube, as I 0:16:30.320 --> 0:16:35.160 just mentioned, extensively for years to classify and address videos. 0:16:36.080 --> 0:16:39.840 Are we over estimating how quickly some of the AI 0:16:39.920 --> 0:16:42.720 products will take over? We had this amazing interview with 0:16:42.760 --> 0:16:45.440 the Bumble founder and she was saying, you know, for dating, 0:16:45.480 --> 0:16:47.880 for example, you're going to have a personal assistant that 0:16:47.960 --> 0:16:50.840 kind of dates for your online and then reports back 0:16:50.880 --> 0:16:53.960 on who's a super suitable match. I mean that feels 0:16:54.000 --> 0:16:54.640 like science fiction. 0:16:55.240 --> 0:16:57.480 I don't know, bumble, I should just be clear, But 0:16:57.880 --> 0:16:59.800 I think it's up to us, Like you know, none 0:16:59.800 --> 0:17:01.640 of them, this is inevitable. And this is my point 0:17:01.640 --> 0:17:03.560 about like this is a moment where we have to 0:17:03.560 --> 0:17:05.959 sort of engage into this stuff and take control. So 0:17:06.000 --> 0:17:08.000 some people might want that for dating. I'm not sure 0:17:08.040 --> 0:17:10.119 I would, but you know, you'd have a choice. Or 0:17:10.160 --> 0:17:11.639 maybe you've got an advisor who said, here at some 0:17:11.720 --> 0:17:14.439 better suggestions or whatever. That's great. Same as you know, 0:17:14.480 --> 0:17:17.840 reading the news. Actually, I might want to have a 0:17:17.840 --> 0:17:21.040 little bit more dissonance in my consumption of news, and 0:17:21.119 --> 0:17:22.800 maybe an a I could suggest that. Or I might 0:17:22.840 --> 0:17:26.000 want to have an expert summary of a long running story, 0:17:26.000 --> 0:17:27.680 and if I could come to Bloomberg and your AI 0:17:27.720 --> 0:17:30.240 assistant would give me a brilliant summary of some of 0:17:30.240 --> 0:17:33.480 your best journalism around a complicated issue, that would be useful. 0:17:33.480 --> 0:17:35.199 But I think we have a choice, and it's up 0:17:35.240 --> 0:17:37.080 to us to make those choices. 0:17:37.320 --> 0:17:39.199 Who's it up to the consumer, or is it actually 0:17:39.560 --> 0:17:42.880 you know companies like Google to decide. I mean, if 0:17:42.880 --> 0:17:45.480 we were all to be given you an assistance that 0:17:45.520 --> 0:17:48.000 can help you in everyday life, is it too soon 0:17:48.080 --> 0:17:49.440 to say whether that's a good or a bad thing? 0:17:49.840 --> 0:17:51.639 Let me step back. I think where we are is 0:17:51.680 --> 0:17:53.920 we've got this broad based technology. We're beginning to see 0:17:53.920 --> 0:17:57.639 the potential, but most of the applications of that technology 0:17:57.680 --> 0:18:00.000 have yet to be invented. So we can get lots 0:18:00.240 --> 0:18:02.919 in generalizations. But you know, a specific example about how 0:18:02.920 --> 0:18:05.040 it might work with dating we could talk about, or 0:18:05.160 --> 0:18:08.120 how it might work with newsroom productivity we could talk about, 0:18:08.240 --> 0:18:10.920 or how it helps call center agents. Gives you better examples, 0:18:10.960 --> 0:18:13.359 So I think we need to look at the applications. 0:18:13.680 --> 0:18:16.680 How do you hire it at Google? I've heard from 0:18:16.760 --> 0:18:19.679 some tech companies that it's all about morals and values 0:18:19.720 --> 0:18:21.919 and it doesn't really matter if the candidate knows that 0:18:22.040 --> 0:18:23.560 much about technology to start with. 0:18:23.800 --> 0:18:26.159 Yes, absolutely, we try to hire people of good character, 0:18:26.600 --> 0:18:31.320 but also people who are collaborative, curious, creative, good communicators. 0:18:31.440 --> 0:18:33.399 I think the most important things when we're hiring are 0:18:33.400 --> 0:18:38.280 those skills of curiosity, creativity, collaboration, communication, Because we're asking 0:18:38.280 --> 0:18:41.120 people to invent technology that didn't exist before, and we're 0:18:41.119 --> 0:18:43.840 asking people to sort of explain that technology to the world, 0:18:44.160 --> 0:18:46.679 to listen and engage with governments and companies and partners 0:18:46.760 --> 0:18:49.119 on how we do that. What I'd say to people 0:18:49.160 --> 0:18:51.679 thinking about, you know, maybe choices for their children today 0:18:51.680 --> 0:18:54.000 with their children, I think it's really important we have 0:18:54.040 --> 0:18:59.760 a broad education. You know, innovation comes from interdisciplinary collisions, right, 0:19:00.040 --> 0:19:04.160 and I don't think we should be overly focused on STEM. 0:19:04.520 --> 0:19:08.000 I think we do need representation so that people building 0:19:08.000 --> 0:19:11.160 the technology have to be representative of everyone. And that's 0:19:11.200 --> 0:19:13.320 been a challenge and one of the reasons we've published 0:19:13.640 --> 0:19:16.920 our own data on diversity and trying to improve particularly 0:19:16.920 --> 0:19:20.119 representation representation of women and minority ethnic groups in the 0:19:20.200 --> 0:19:22.920 engineering side of Google. We've made progress, but there's more 0:19:23.400 --> 0:19:25.920