Gauging China's AI strength

0:00This BBC podcast is supported by ads outside the UK.

0:30at Whole Foods Market. Rubrik is the security and AI operations company built for what happens after an attack hits, not just the moments before. AI has turned the threat landscape into quicksand, moving too fast for any human to fully predict. That's why an agentic cyber resilience platform matters. Automated recovery, clean data, a business that keeps moving no matter what hits. One platform, not a patchwork of stitched together tools and gaps. Don't wait for the next attack. Secure and accelerate your business at rubric.com. Again, rubric.com.

1:27and legs. I mean the unassuming ones which quietly clean floors and move stock around warehouses. And this technology, in my opinion, can't come soon enough in order to help us as a society get over that demographic challenge that we're all going to face. And the sport that's more googly than Google. Do I hear the sound of leather on willow? Well, he struck that one right past silly point. We look at how AI is deciphering the confusing and mysterious language of cricket.

2:22Relations between China and the United States will be at the top of the news agenda this week, with the leaders of both countries due to meet in Washington, D.C. The meeting follows a run of alarming headlines about the risk AI may pose, from swarms of autonomous agents from the big AI companies escaping test environments, to hack online services, to dire warnings from insiders about the threat AI poses to security, the financial system and maybe even to humanity itself. Much of the discussion about what should be done has considered the need for multinational agreements. So what do we know about China's AI industry?

3:03Rebecca Archisati is co-head of the Science, Technology and Innovation Program at the Mercator Institute for China Studies, and she's based in Brussels. I began by asking her if China was really behind the United States in AI development, and if so, would they catch up? China is still behind by several metrics. However, I also think that they're catching up quite quickly. remarkably fast. I think where the US really holds an advantage is that they have the most computing power. And that has to do with several things, mainly the fact that China indigenously still cannot produce the most cutting edge fabrication equipment to make AI chips.

3:49And a lot of those chips are therefore designed by US companies, produced mainly in Taiwan and subject to expert controls, meaning that Chinese AI labs don't have access to the same amount of high-end computing power as their US competitors. That remains the case, even though the Chinese semiconductor industry has made a lot of progress. Is there a different flavor to Chinese AI? I hear that said sometimes, that they're less concerned about large language models, more concerned about things like robotics? I think they're doing both. I think China is interested in exploring many different types of AI and many different applications.

4:34And it's also looking at different pathways to achieving what, you know, in the West we call artificial general intelligence. China also has a term for that. I think overall, scientists in China aren't necessarily convinced that the only way of achieving artificial general intelligence is by, you know, scaling these large language models. There are many different paradigms that they're pursuing. But if you look at what they're saying, it's very, very clear that they're also aiming at, you know, human-like intelligence. They're also looking in that direction. And the government is very much supporting advanced research in that direction.

5:19How long will it be until China does catch up? I mean, how fragile is that American lead? It's difficult to say, but I think if you look at the recent report, the threat intelligence report that Anthropic released, I think that maybe that's where we may be able to find part of the answer. What that report says, I found quite interesting, is that Chinese labs apparently haven't only been distilling U.S. models, they've also been rerouting their own user requests towards U.S. models like Claude. To me, if that is true, it tells me that Chinese labs have been able to piggyback the compute constraints by relying on US models.

6:13And when we talk about distilling a model, we basically say you sort of learn how it responds to questions, to how it interacts, and you use that to build a version of it, essentially. Yes, that's correct. Rerouting is a different practice altogether, right? By rerouting, you are essentially taking your user's input, you know, prompt, request, and you are redirecting it towards a model developed by one of your competitors unbeknownst to your user, which is a very different practice. Is China in a good place to develop advanced AI? I mean, does it have resources and potential assets which the US maybe doesn't have?

6:58China is in a very comfortable position when it comes to energy, for example. It has a very resilient energy mix, and energy is very important when it comes to both training and deploying AI models. Sure, because these data centers just hoover it up, don't they? I mean, you know, big data centers are like aluminium smelters in terms of electricity demand. Yes, and they have plenty of renewables. They have still a lot of fossil fuels in their energy mix. They're going nuclear as well. Chinese AI companies are able to access the electricity they need for a much lower cost. And that is really an advantage when you're trying to build this, you know, large scale infrastructure to power a society's or an economy's transition towards the AI era.

7:51I think this is a big advantage that China has. How does China see the AI race? I mean, does it see it as a race it needs to win? as a race it should be in? Yes, China very much sees it as a race. If by China we're talking about the Chinese Communist Party or the government of China, of course, they very much believe that China should be the leading force in AI. They see AI as sort of a power maximiser for both national competitiveness in economic terms, but also in the military domain and in the strategic domain. They want to be in a position where they dominate this technology and therefore they want to surpass the US, which is currently the incumbent.

8:43Do you think we really know what's going on in China's AI industry and the development of its AI models? It can be quite secretive. We shouldn't assume that we know everything. And I suspect that going forward, we may have probably less and less access to certain information. The Chinese government has been trying, for example, to limit the liberty of AI talents to move abroad. And this tells you that the Chinese government is increasingly concerned about know-how and technology flows from China, also data flows. It's concerned about any form of know-how falling in the hands of U.S. companies and the U.S.

9:33government in particular. And with this growing scrutiny, I think we may expect information to also disappear more quickly and perhaps some Chinese labs to obscure more of their frontier R &D activities. And that was Rebecca Archisati from the Mercator Institute for China Studies speaking to me from Brussels.

9:58September's always busy. Whole Foods Market can help. Their September Stock Up event makes it easy to load your pantry and freezer with flavorful, nourishing food. Even better, there are hundreds of sales. Get dinner going with canned soups and veggies. Lean on frozen pizzas, pastas, and seafood everyone loves. Their Build Your Own Family Meals feeds four for just$35. Stock Up at Whole Foods Market. John Deere is putting in work and putting up highlights. Oh, man, look at those passes. Tillage, planting, harvesting. Three years in, and these pre-owned machines are still at the top of their game.

10:37We're talking reliability, performance, and long-term potential. And get this, these veteran machines can pick up the latest precision tech with ease. You know it, and fans ready to add John Deere-certified pre-owned equipment and more options to their lineup can shop at machinefinder.com. Booyah!

11:07This is Tech Life on the BBC World Service with me, Chris Valance, and with you, our listeners. Last week, Shona McCallum spoke to an academic and author about what children should be learning in a world transformed by AI. Regular listener Nukombi WhatsAppped us from Cameroon. He told us, rather than simply trying to keep AI away from them, I believe we should teach young people how to use it responsibly and creatively. If they're properly guided, they will be better prepared to understand both its risks and its enormous opportunities. Well, that's kind of our business too, Nukombi, trying to understand the fast-changing world of tech.

11:48And if you'd like to help us out with an idea for a future programme or a reaction to this one, do get in touch. You can WhatsApp us a text or voice note to plus 44 330 1230 320 or email us at techlife at bbc.co.uk. Do remember to say who you are and where in the world you are.

12:15Now, some breaking news. Robots have taken over the world. Don't worry, I'm not talking about the robot apocalypse, but the fleets of autonomous robots dutifully performing their mundane tasks in supermarkets, airports and factories around the globe. Designing these robots poses lots of hard technical questions, legs or wheels, for example, but the component that really, really matters is the robot's brain. Brain Corp is a US company which has made the operating systems, or the brains of more than 50 ,000 robots across 48 countries. David Pinn is the chief executive of BrainCorp. I asked him how you turn something like a floor cleaning machine operated by a human into a floor cleaning machine with a brain of its own.

13:03We work with our partner, you know, a company that's already providing manually driven machines. We work with them to design a version of the machine that's got the right computer, that's got the right sensors in order for it to understand its environment. And then we, BrainCorp, provide the software that is taking in the input from those sensors, understanding where the robot is in the environment, making decisions about how to navigate in a way that's going to most effectively clean the floor. In this example, think about a grocery store, which typically likes to clean the floor around eight o 'clock in the morning after all the stocking overnight is done.

13:44And while customers are starting to come in for their morning shopping, we're very comfortable and really pay a lot of attention, not just to how we navigate to get the job done, but also how we navigate and how we interact with the customer going around in the store. And so there's a lot of work in the software algorithms, in the behavior of the robot to make sure that it's both doing its job well, but also being courteous to the people around it. Not very far away from where I'm talking to you, there is a university cafeteria and the university cafeteria has a robotic dish collector and it's basically a set of trays on wheels.

14:21But it's a very basic sort of robot. It just wanders up next to the tables and sort of flashes until you feel guilty and put a cup on it. How far away are we from those kind of robots that can really interact with people in the way, say, a human cleaner would interact with people? You see robots today that are very confidently driving around, but you don't see an arm and a hand on the robot that can pick up a cup, for example. And the reason why you don't see that yet is because the complexity of the human arm and hand cannot be overstated. We, as an industry, have solutions around what we call mobile manipulation in labs, in demos.

15:07But we, as an industry, have not yet cracked the case on how to do mobile manipulation in a way that's reliable enough to deploy at scale. We talk about the kind of AI, the kind of physical AI that will be in robots. I mean, do you think in the end that might lead to AI that is more like us. I mean, we're physical I, if you like, we're the product of millions and millions of years of evolution, learning to encounter the world. And that's what robots are having to do. So I wonder how far you can take robot AI. Yeah, there doesn't seem to be a limit. I think you've got it exactly right. We're training them with human action.

15:51You can see companies that are outfitting people with cameras near where their eyes are, with sensors, kind of like motion capture sensors for their limbs. And these companies are capturing all of that data, the vision data and the motion data, and they're feeding them into these deep learning algorithms. These algorithms are able to learn in kind of an organic way, which is fascinating. And that's the big breakthrough that we saw a few years ago with ChatGPT that's now migrating into robotics. And it's an amazing time. And we as an industry are figuring out how to solve really hard problems in jobs that are typically dull, dirty, or dangerous.

16:38The kinds of tasks that it's really hard to find labor to do it today. And the problem's only getting more acute with demographic changes, as we all know, aging population, declining birth rates. There's just going to be fewer and fewer people working to support the aging. And this technology, in my opinion, can't come soon enough in order to help us as a society get over that demographic challenge that we're all going to face. Should we be worried about whether the world of work is going to be displaced by robots? I mean, I know it must be a question you get asked all the time, but are jobs going to be taken by robots?

17:24Is BrainCorp going to put us all out of a job? I really don't think so. So the kind of automation that we do today is the kind of automation that really the labor just doesn't exist for it. We're not putting people out of work. We're helping people do jobs that just aren't being done. in floor care, there just aren't enough people to do janitorial work in grocery stores in general. And so our solution is showing up as a way not to displace labor, but to make the floors cleaner. Point number two is there are a bunch of jobs that could have been automated decades ago that aren't. And they aren't because people prefer working with people in a lot of jobs.

18:10My favorite example of this is bartender. Like you can imagine just putting bottles in some sort of piping system and just having like a vending machine. And you just have a cup underneath, right? I mean, it's like a vending machine. This technology has existed, I don't know, since the 70s. And yet bartenders exist. Why? Because people like ordering a drink from a person, as I do, right? And so even though a job can be automated, it doesn't mean it's going to be automated. What are the robots of the future going to look like? Sci-Fi is filled with humanoid robots and there are quite a few companies developing humanoid robots.

18:47What's your take on it? So I think that there is probably a place for humanoids, likely in residential environments, in my sense. But when I look at commercial and industrial environments, I really don't see the need for legs. Wheeled equipment is really what rules an industrial environment today. In the case of commercial environments, you know, we've, as a society, we've made tremendous strides in making them accessible for people with disabilities. And so you see ramps everywhere, you see elevators everywhere. These environments are also made to cater to wheeled motion. And so from my point of view, legs are expensive, they're complicated, and they're unneeded.

19:35And so I fail to see why a buyer of automation in an industrial or in a commercial space would choose to add the expense of legs when you can solve the same problem with a robot on wheels. And that was David Pinn, chief executive of US-based BrainCorp, discussing the future of commercial robots.

20:09Now, how about a bit of Soul Limbo by Booker T and the MGs?

20:25Now I'm not sure what Booker T or his MGs intended the song to be about, but for most people in the UK it summons memories of just one thing, cricket. And it's tempting to think of cricket as something that never changes, a soundtrack to an English summer. But of course it does change. The gravitational centre of the sport, many would argue, is now in India, and its millions of fans are spread around the globe. And as cricket fans have moved into new countries, so has the sport and the International Cricket Council, or ICC, which governs it, is keen to grow new audiences in new countries. And to do it, they're embracing AI.

21:04But cricket has always been keen to embrace innovation, as Finn Bradshaw, head of digital at the ICC based in Dubai, told me. Well, I mean, I think just in terms of innovation generally, I think cricket's been quite extraordinary. You know, it's gone from five days to one day to three and a half hours. And a lot of other sports have tried that with little success. But from a technology point of view, I think probably the greatest example is what we call DRS, which is the system that umpires use to validate their decision making. it's been in cricket for 20 plus years and it's just accepted as part of the game now you think of how much in soccer they're still arguing over VAR but there's many other examples whether that's bales that light up when they get hit by the ball or whether that's the way that broadcast pivots around the the stadium to give people 360 degree views it's we're a sport that pride ourselves on constant innovation one of the things you're trying to do is you try and promote cricket to the world is to do commentary in the language that people speak in the countries you're aiming to popularize the game to and you you're exploring ai generated commentary how does that work so we have been experimenting with this for at least over the last 12 months or so and it came about because we started to just put you know we obviously put our highlight clips out globally but then we started running our own live streams into certain countries and countries where we didn't think there was much cricket fandom at all suddenly started to we started to see them light up and so there are countries like Japan, Germany, France, Italy, Korea so initially we just did a straight literal AI translation of the normal English broadcast.

23:04And, you know, we did it with various services. And, you know, we got, obviously, I don't speak Japanese. So, we would get someone from the Japanese Cricket Association to listen to the commentary and say what it's like. And they would say, well, it sounds good. Like, it sounds, you know, the accent is right and it's pronouncing the words correctly. But there's so much nuance to cricket, it doesn't really make sense and and so obviously there's like very specific terms like yorker or a duck or whatever that if you do a literal translation either the word doesn't exist or it makes no sense but we deliberately keep create a bit of content that is quite simple in it in the way it talks to people and and over explains things and then we put that into a piece of software that then puts that out in 14 different languages.

24:08What's been the hardest bit to translate? It's the terminology. Yeah. I've got family members who aren't from a country where cricketing is popular, and I think they look at it like some kind of obscure dance. You know, they just don't get it. What do you find the hardest thing is? The trickiest one is a wicket because a wicket can be the pitch. A wicket can be getting out. It can be hitting the stumps. And so you sort of, again, that's where we work with the local cricket associations. Let's talk about the way people can watch the game. And another thing you've been looking at are 360-degree replays, which in normal circumstances might require a lot of cameras, but you're doing something different.

24:58Yeah, that's right. Because what we want to be able to do, because cricket is played in a large oval, is when something's happened, be able to move people through space to understand how fast someone had to jump or how quickly the ball was going. in a way that you can only sort of convey that by putting the camera, you know, almost next to a fielder. And obviously you can't stick a camera right next to a fielder. And so we've worked with technology companies now where they have a certain number of cameras around the ground, but then they use technology to essentially recreate the frames in between where a fielder was when they jumped here into the...

25:44And it almost like makes an assumption about where they ended up, if that makes sense. But it means that then we can provide this 360 degree view. What's the future of tech in cricket? It's really exciting to sort of see how we can make it more personal for everyone. You know, being able to do this in language is just the start. You know, imagine if we can take it to the next phase of, you know, who's your favourite, maybe who's your favourite commentator? Maybe you wish that Richie Benno was still commentating. Imagine if we could give you your own commentary on the match in the voice of Richie Benno that talks specifically to the kind of things you're interested in.

26:29Maybe you're really interested in the very ultra nerdy stats or maybe you're someone who just wants to sort of hear great stories from the past. You'll be able to create that sort of personalised experience for you I mean, when people hear AI-generated commentary, they'll worry that AI is going to replace commentators. And of course, you know, people are terribly attached to some of the wonderful voices that there have been. And cricket, particularly test match cricket, gives you so much space to improvise, to philosophise. You're not going to replace any of that, are you? No, because I actually don't think there's yet an AI model that can do a turn of phrase or capture a moment or even decide not to talk.

27:24But actually the best commentators often just shut up and don't say anything. Lesson to all us broadcasters. A bit different to radio. And on that note, we will end the interview. Thank you very much. It's been a pleasure talking to you. Thank you, guys. And that was Finn Bradshaw, Head of Digital at the ICC, based in Dubai.

27:51Well, we've reached the end of our innings on Tech Life for today. So with the light fading, I'm heading back to the pavilion. There's just time to remind you that you can contact us about any tech you're interested in. TechLife at bbc.co.uk is our email address. or WhatsApp us, a text message or a voice memo. The number is plus 44 330 1230 320. Please let us know your name and the country where you live. Today's Tech Life was produced by Tom Quinn and presented by me, Chris Vallis.

28:39Welcome to the Haunted Library. Have you come in search of a tale? This podcast invites you to browse the shelves and find stories that will linger long after you've left this place. From gothic masterpieces to pulpy penny dreadfuls, join me, Colin Morgan, with new episodes twice a week. Find the Haunted Library wherever you get your podcasts. Occ风-ee geez support

29:35a