Visit our website:
http://www.france24.com
Like us on Facebook:
https://www.facebook.com/FRANCE24.English
Follow us on Twitter:
https://twitter.com/France24_en
Category
🗞
NewsTranscript
00:00This is Apropos.
00:04Global experts are here in Paris this week to debate the threats and promise of artificial
00:09intelligence.
00:10The summit is aimed at finding common ground on the rapidly evolving and disruptive technology,
00:16as well as examining how to keep France and Europe on the map as credible contenders in
00:21the AI race.
00:23Paris is expected to announce major investments in the fast-moving technology running into
00:27the billions.
00:29Peter O'Brien is there.
00:31He spoke earlier with Camille Knight.
00:34This is the first of two science days and boffins that have gathered here from around
00:38the world were very, very excited, particularly because of this recent breakthrough by Deep
00:44Seek.
00:45The prospect that we're going to be able to deploy advanced, powerful AI models at a fraction
00:51of the cost has got smaller players in France very excited.
00:57Now a lot of big names expected at the summit.
01:00Peter, you were able to speak to Yann LeCun, Meta's chief AI scientist.
01:04What did you learn from that conversation?
01:09Well I wanted to ask him whether he felt vindicated because he's French, of course, and he's been
01:15absolutely championing a open source approach to AI for years now.
01:20That's what Meta does.
01:21They have these Lama models that are very powerful.
01:24They get copied around the world because they are open source.
01:28And he's been banging the drum and criticizing open AI, for instance, for closing off access
01:34to its models.
01:35So when Deep Seek came along and blew a lot of the competition away in terms of efficiency,
01:40he sounded pretty happy.
01:42I also talked to him about his boss, Mark Zuckerberg.
01:46He said that more masculine energy was needed in corporations.
01:52So when it comes to AI and developing AI, I asked Yann LeCun whether more masculine
01:58energy was needed there as well.
02:00I had a big project called Zeta, whose goal was to basically build a large-scale LLM.
02:08And then for various reasons, a smaller group in France decided that they needed LLM for
02:15their project.
02:16And it turned out to work better than the big ones.
02:19It's not because of the quality of the people.
02:20They were both very talented, but the problem was that one of organization and management,
02:31the small group had a lot of autonomy, a long leash, was relatively small, and had a lot
02:37of freedom, essentially.
02:39The big group was kind of top-down managed, more directed, and in the end, a less efficient
02:46way of organizing research.
02:47And so we see the same thing with diptych.
02:49Nothing to do with masculine energy or lack thereof?
02:52No, no.
02:53Not at all.
02:54That doesn't come into it?
02:56Absolutely not.
02:58Peter O'Brien there reporting.
02:59Well, for more, we're joined now by Simon Coulton, Professor of Artificial Intelligence
03:04at Queen Mary University of London.
03:08Thank you so much, Simon, for being with us on the program.
03:11Let's start with the timing of this summit, coming as it does in the aftermath of the
03:16emergence of DeepSeek.
03:18How much of a game-changer do you think this Chinese technology is?
03:24I think it is a big game-changer.
03:27Of all the various predictions of how this is going to change AI and AI funding, I think
03:32the one which I think is most sensible is that we'll see this as a World Wide Web moment,
03:38where it became a lot easier to make, post, and find web pages through the World Wide
03:43Web.
03:44The Internet was not so usable before that.
03:47And now AI has become 50 times cheaper, if you believe the figures, to operate.
03:53And that can be just the beginning of it.
03:56If there's as much money poured into scaling down AI models as there was into scaling them
04:00up, then we can imagine hundreds of times cheaper.
04:03So that will make it available to organizations to use at scale a lot more often, a lot more
04:12down the tech chain, so that we will find that it's an explosion of usages of AI.
04:19And we might find that that will mean that money flows from the kind of data center area,
04:26where we're building up these huge banks of graphics processors, into more startups, into
04:33more thin-layer companies, which are able to run these models on their own machines.
04:39And then if the models get even smaller, which I think is quite unlikely, to be perfectly
04:43honest, but if they do get smaller and they can be crammed onto a laptop, the really good
04:48models, not the smaller distilled versions of them, if the really good ones can be crammed
04:52onto a laptop or a mobile phone, then that is a huge game changer, because we don't need
04:56quite so much of the GPUs available.
04:59So it could be a pivotal moment.
05:02And then, of course, there's a whole geopolitical element to it as well, as well as a financial
05:06element.
05:07So I think it will be seen, ultimately, as a big moment in the tech arena.
05:13A big moment as well.
05:14It also wiped a lot of money off the value of some of the biggest players in the sector
05:18initially.
05:19Does this suggest that American tech companies who pledged billions of dollars in AI investment,
05:25that perhaps that was a little short-sighted?
05:26Possibly.
05:27Certainly, the market thought that.
05:31The timing was just fabulous of the release of this.
05:35Not only was it just a day or so after the announcement by the American president of
05:40the Stargate project, $500 billion is being pumped into infrastructure like data centers,
05:46but it was also just before the weekend.
05:48So that meant that the market had time to enact a kind of epic overreaction to it.
05:55It wiped $600 billion off the market capital of NVIDIA and a trillion off the NASDAQ in
06:01total.
06:02However, I kind of believe the line that if you make something cheaper, people will buy
06:08it more.
06:09There's an economics theory about that.
06:12So all of this is still going to be run on GPUs, and I don't think it will mean at all
06:17that NVIDIA will suffer in the long term.
06:19I think there'll be equally as much compute required for all of the extra usages of GPUs
06:28that will be enabled by this leaner, slimmer technology.
06:32And Simon, there's a global AI summit taking place in Paris this week.
06:38We've been referring to it earlier.
06:40The UAE has just announced that it'll be investing up to 50 billion euros in a data center here
06:46in France.
06:47What does Europe need to do at this point?
06:50Is it those kind of investments that we need to be securing in order to become really a
06:55global player when it comes to artificial intelligence?
06:59Yes, I still think there's very much a requirement for compute.
07:04Ask any of us in academia where we don't have access to these huge tens of thousands of
07:09GPU units and we'll tell you that, yeah, compute is key.
07:13All of my PhD students are always desperate to find GPUs in order to train their models.
07:19But there are other factors at stake here that will need to change as well.
07:24I try and take a balanced view.
07:25I understand the value of AI safety and ethics, but there are reasons to think that the
07:34European Union is being overly careful and that might stifle innovation and that
07:42ultimately it means that companies can't take risks.
07:46There's going to be battles over copyright, so over copyrighted data enacted in various
07:53industries. So there are governmental requirements as well as just building compute
07:59data centers, which will have to make it more amenable for startups to take advantage of
08:05them and really flourish in a less risky environment.
08:11And when it comes to investing into this technology, what then should be the priority?
08:16Is it funding that's needed for the actual tech?
08:19Is it funding that's needed for ethics?
08:21Is it funding that's needed to ensure the safety of these AI models?
08:25What should be the priority here?
08:29Well, all of the above, really, and I might add to your list fundamental research like the
08:35kind of thing that we carry out in universities.
08:38I'm hoping that like Demis Hassabis, the DeepMind co-founder and recent Nobel Prize
08:45winner, he's come out on record recently in the Financial Times saying that there needs to
08:49be way more artificial intelligence PhDs funded in the UK.
08:54And that is surely true of France and other countries, if not all countries.
08:58PhDs can undertake really groundbreaking, risky research, but there just isn't enough
09:05funding for them.
09:06And often they go straight into industry to these big companies without giving back to the
09:14society that funded them.
09:17PhD ceases and results.
09:19So I think compute data centers absolutely required technology in order to get these off
09:23the ground. The fact that deep learning requires a huge amount of compute, a huge
09:27amount of data and a huge amount of time is not going to go away, even with deep seeks
09:32revelations. But also there needs to be more money into AI ethics regulations, but how we
09:40can all come to a good agreement about that so that technology can move on for the good of
09:44society without being hamstrung too much by regulations.
09:48If you ask various people in various startups, they will point out that it is quite
09:53stifling to be within EU regulations right now in the AI sector.
09:57And Simon, you say that we're actually looking at a battle royale between capitalism and
10:03AI and then government forces, on the other hand.
10:06So where is all of this evolving to, do you think?
10:10My guess is that capitalism will win.
10:12It's depressing.
10:13But ultimately on things like jobs, white collar jobs are now under threat in a way that
10:20they weren't previously.
10:22And in many respects, you could imagine companies saying, you know, this is great.
10:27We can deploy our human resources in new, more interesting ways rather than doing
10:32drudgery. But the reality is that if an AI system can take their job or can do their job in
10:39a much cheaper way, then company executives are going to take that option.
10:43And we'll have to start seeing real discussion about four day weeks, about universal basic
10:49income, about ways in which society can cope if the changes in the job market, as are
10:57predicted by a lot of people, really happen.
11:00Now, I'm glad that we'll take away a lot of the drudgery, even from creative work and
11:05freeing people up to be expressive in the arts and music.
11:09But I don't think companies are really going to take the lead in making this a nice
11:16transition for their workers.
11:17And I think that's where government will have to pick up the slack and really legislate
11:21so that not just that people keep their jobs, but also that they are replaced in a
11:27human friendly way if AI systems can indeed take their jobs.
11:32And just finally, Simon, everyone had assumed, I think, up until the last few weeks,
11:35that the US sector was leagues ahead, really, when it came to artificial intelligence and
11:40all the technology surrounding it.
11:41But does the emergence of these Chinese competitors suggest that there really is room
11:46at the top in terms of global leadership here?
11:49Yes, I mean, that's clearly the message that has come out from China.
11:54Those of us in the business kind of knew about DeepSeek for a while beforehand, but it
12:01did come as a shock to most of us at how good this technology was, how competitive, how
12:06fast. And of course, that it was done entirely by Chinese nationals in a Chinese
12:11company operating openly under state intervention rules.
12:16So they're quite clear that they are subject to Chinese regulation and therefore their
12:20data is available to Chinese government and they are censored in various ways.
12:26So it does, people are calling this a Sputnik moment for good reason.
12:31During the space race, there was that moment when Sputnik was launched and the Americans
12:37realized that they were in a competition and therefore that sparked the race to the moon.
12:43I think history will record this time as being the AI race.
12:46And if they didn't already, the Americans now know that they are in a proper race for AI
12:52dominance. My hope is that it's a collaborative race.
12:55I'm a utopian in that respect.
12:57And like people like Yan Likun, I'm hoping this will be for the benefit of mankind
13:02across the planet.
13:04But it may all get nasty.
13:07Let's see. If you control AI, you control a lot of other aspects of human culture.
13:13Simon, we'll have to leave it there for now.
13:14Thank you so much for being with us on the programme this evening.
13:17That is Simon Colton, Professor of Artificial Intelligence at the Queen Mary University of
13:23London. Well, that's it from us.
13:25For now, just stay with us, though. We'll be back shortly with more world news.