FRANCE24 — broadcast
20260920 100000 UTC
517 transcript segments
Google Cloud Speech-to-Text API (Chirp) + Gemini 2.5 Flash Non-Thinking
Data courtesy of The GDELT Project (https://www.gdeltproject.org/), from the Internet Archive TV News Archive.
Machine transcription. Treat it as a searchable index of what was broadcast, not a verbatim quotation record.

[00:00:00] live in fear, a report to see on France
[00:00:02] 24.
[00:00:08] It's noon in Paris, welcome, we'll be together until 2 PM, this is Paris
[00:00:13] Anthropic said that it is about Live, tension is high in Russia on the last day of the legislative elections,
[00:00:16] Ukraine carried out a very large-scale attack, notably in Moscow,
[00:00:20] a candidate died in Kherson in Ukraine, these
[00:00:23] are four people, including three children, who lost their
[00:00:26] lives. Two German regions are voting today.
[00:00:30] The stakes are such that the Chancellor canceled his departure for New York, he convened an emergency meeting
[00:00:34] for this evening, the AFD could make new gains after its victory in
[00:00:37] Saxony, even if the far-left is very mobilized in
[00:00:41] Berlin. As every year, the Heritage Days
[00:00:44] are a success, thousands of monuments accessible to the public
[00:00:48] to learn about our history as well as our
[00:00:51] traditions. You can also discover the construction sites
[00:00:54] led by
[00:00:55] enthusiasts.
[00:01:01] and... especially in the international AI
[00:01:09] Third and last day of voting
[00:01:11] She presents herself as the antidote to in Russia to renew the Duma, and the capital suffered a massive attack by
[00:01:15] Ukrainian drones, no casualties are to be deplored in Moscow, but four
[00:01:19] people were killed, among them, a candidate in the Kherson
[00:01:22] oblast. Ukrainian strikes have disabled a significant portion of Russia's
[00:01:26] refining capacity in recent months. Axel.
[00:01:32] According to the mayor of Moscow, this is an unprecedented Ukrainian offensive.
[00:01:36] speech, who leaves the The Russian capital and its region have been the target of several hundred
[00:01:40] issue today for the American tech giants who have billions Ukrainian drones, while most have been intercepted and shot down,
[00:01:43] several escaped Russian anti-aircraft defenses and notably
[00:01:47] struck several floors of a residential building in the capital
[00:01:49] or this oil refinery. Other
[00:01:53] devices fell in this residential neighborhood of Sofino in the Moscow
[00:01:56] region, causing casualties and significant damage.
[00:02:30] Russian strikes on more than 30 sites in the Kyiv region killed several people,
[00:02:33] including children, on Saturday and Sunday. This
[00:02:37] offensive also comes on the 3rd and last day of voting in Russia
[00:02:40] and has been denounced by Russian authorities as aiming to
[00:02:44] disrupt the election process. In the Kherson
[00:02:47] region, a southern Ukrainian region controlled by Russia,
[00:02:49] a drone notably killed several people, including an electoral
[00:02:53] commission employee. More than 4
[00:02:56] and a half years after the start of the large-scale invasion of Ukraine in 2022,
[00:03:00] these legislative elections aim to renew the Duma, the lower house of the Russian
[00:03:04] parliament, in a context of repression of all forms
[00:03:07] of opposition, they should, without surprise, result in the victory
[00:03:11] of Vladimir Putin's party. Despite
[00:03:14] this, concern is growing within Russian society, rumors of a forthcoming
[00:03:18] mobilization are persistent and are pushing young people
[00:03:21] to leave the country, Léopéchar.
[00:03:24] Computer equipment and the like, his suitcase
[00:03:27] packed, this 28-year-old marketing specialist is
[00:03:31] ready to leave his country, with a heavy heart, it's
[00:03:34] very sad to leave Russia, my family lives here, as do my
[00:03:38] elderly grandparents, and I would like to spend the remaining years of their lives by their side,
[00:03:42] years that I hope will be long, a
[00:03:44] baggage check before taking the train to the Georgian
[00:03:48] border, the last kilometers to the checkpoint
[00:03:51] are done by car, on the
[00:03:54] road, dozens of vehicles abandoned, mostly in
[00:03:57] 2022, many Russians... had fled the country en masse
[00:04:01] after Moscow's announcement of a mobilization in its war against
[00:04:04] Ukraine, a forced enlistment that could happen
[00:04:08] again according to persistent rumors circulated
[00:04:11] notably by Kyiv. This is one of the main reasons that pushed
[00:04:15] this Russian to flee the territory.
[00:04:18] Rumors of mobilization scare me, even though I am
[00:04:21] classified in category C, which means I am not called up in peacetime,
[00:04:25] it's 12:14 in Paris, this is the essential you never know what can
[00:04:27] economy week happen from one day to the next.
[00:04:31] and he is not the only one to have made the journey, even if the scale of this
[00:04:34] exodus is not comparable to 2022, an organization
[00:04:38] Hello, welcome to the Echo helping these exiled Russians indicates that it was contacted 8
[00:04:41] times more in August than in May.
[00:04:44] American AI giants are sounding the alarm, Among the refuge countries, Georgia, but also
[00:04:47] Armenia.
[00:04:49] Since mid-July, there has been a wave of panic from Russia
[00:04:53] linked to the risk of mobilization, fuel shortages
[00:04:56] and internet disruptions. People live in
[00:04:59] unworthy conditions and this time they leave well in advance
[00:05:03] because they have seen what happened during previous waves, a migratory
[00:05:07] crisis is looming in Armenia if we are not already in the middle
[00:05:10] of it.
[00:05:12] The Secretary of the Security Council of the Armenian government
[00:05:16] declared that the country would help, if necessary, these
[00:05:19] Russians newly settled in the territory, statements
[00:05:22] deemed provocative by the Russian Minister of Foreign
[00:05:25] Affairs,
[00:05:28] and on this last day of voting in Russia
[00:05:32] well, Russians who have already chosen this exile
[00:05:36] demonstrate to denounce these elections that they
[00:05:39] call masquerades, these are images that reach
[00:05:42] us from Belgrade in Serbia where a large
[00:05:46] Russian community lives in the country,
[00:05:50] a mobilization therefore to denounce this
[00:05:53] election. The German Chancellor is ready to
[00:05:56] act as soon as the first results are announced, a crisis meeting of his
[00:06:00] conservative party is scheduled for this evening, the Lender
[00:06:04] vote today and the AFD could well register new
[00:06:07] gains after its historic victory in Sax
[00:06:10] Anhalt 2 weeks ago, gains nevertheless less
[00:06:13] significant, notably because in Berlin, the head of the list of the
[00:06:17] radical left is considered
[00:06:18] a favorite.
[00:06:24] the AfD. At 45, Elliferalp has
[00:06:27] every chance of becoming the first woman of immigrant origin to
[00:06:31] win a regional election and conquer the mayoralty of
[00:06:34] Berlin. We will show that a left-wing policy for
[00:06:37] citizens, a policy of solidarity and justice can
[00:06:41] be crowned with success and I hope that this will radiate across Germany
[00:06:45] and the whole world, that people will regain the hope that we can oppose something
[00:06:49] to the right, namely our solidarity, our
[00:06:52] diversity and our fight for equal rights and financial accessibility.
[00:06:55] Daughter of Turkish parents who fled their country after the
[00:06:59] military coup of 1980. This
[00:07:02] lawyer is often compared to Zoran Mamdani, Democratic mayor
[00:07:05] of New York. Like him, the radical left candidate
[00:07:09] places the issue of housing at the heart of her campaign. While Parisian
[00:07:12] rental prices have exploded over the last 15
[00:07:15] years, among her flagship proposals are their
[00:07:19] capping and especially the nationalization of 22,000
[00:07:22] housing units belonging to large real estate
[00:07:25] investors. The measure worries banks, but it is
[00:07:28] supported by Berliners who voted in its favor in
[00:07:32] a referendum in 2021.
[00:07:36] This is precisely what we are trying to achieve, an affordable
[00:07:39] Berlin, an affordable Berlin where everyone can finally live with us.
[00:07:42] for its part, Washington, which has...
[00:08:08] attacks that the kingdom has suffered for several days
[00:08:12] refused to provide military support to Riyadh on the Yemeni front, fears a
[00:08:16] rapid escalation of hostilities to the point that Donald Trump
[00:08:19] cut short his weekend at Camp David. The tools
[00:08:23] are advancing towards the east of Yemen, in the east
[00:08:27] beyond Yemen, in the East, there is Saudi Arabia,
[00:08:31] more than 120,000 people have been displaced
[00:08:34] since the resumption of fighting a few days
[00:08:37] ago. Emmanuel Macron continues his
[00:08:40] trip to Saint-Pierre and Miquelon. He meets the Canadian
[00:08:44] Prime Minister today, this visit is an opportunity for the President to remind that the territory is
[00:08:47] French, even if Donald Trump puts an American
[00:08:51] flag on his idealized map of the world. I propose you listen to Emmanuel
[00:08:54] Macron. It's obvious to us that this territory is
[00:08:58] French and territory attached to France, proud
[00:09:01] to be French as we are. And
[00:09:05] territory liberated by Free France.
[00:09:07] And let's not forget this Christmas
[00:09:10] Day. So uh... we are attached to it, there's no debate, there's no
[00:09:14] discussion, so we are here, France is present
[00:09:17] thanks to Saint-Pierre and Miquelon, by Saint-Pierre and Miquelon here, at the heart of
[00:09:21] the global challenges in this region, and
[00:09:25] this also links us with Canada and so it will be a way to
[00:09:29] talk about the future together and to consolidate our relations in many
[00:09:32] areas, it's therefore effective, the
[00:09:36] media banned from the White House could not enter, the
[00:09:39] decision had been made yesterday by Donald Trump, no less than... 6 weeks before the
[00:09:43] midterm elections, the president wants to control the narrative, but he
[00:09:46] might run into the constitution. Olivier
[00:09:49] Fessoal.
[00:09:52] Previously, these journalists could cross the gates of the White
[00:09:55] House to do their job, but this Saturday, they were
[00:09:59] stopped at the entrance and are now forced to do their live reports from
[00:10:03] the sidewalk. This measure comes the day after Donald
[00:10:07] Trump announced his intention to ban CNN,
[00:10:10] MSNBC and Politico, the three
[00:10:14] media outlets accused of spreading false information by the
[00:10:18] president, denounce their eviction and intend to
[00:10:21] challenge it in court. The American
[00:10:24] Constitution guarantees us the right
[00:10:27] to practice our profession without hindrance or interference from the
[00:10:30] government, and this ban constitutes an illegal
[00:10:33] infringement of this fundamental right. We will not
[00:10:36] be diverted from our duty to hold the government accountable. Donald
[00:10:40] Trump's decision has been strongly criticized, notably by
[00:10:43] the White House Correspondents' Association, which brings together journalists
[00:10:47] accredited by the presidency. Today's action revoking access for journalists
[00:10:51] from CNN, Politico and MSN violates the first
[00:10:55] amendment, the American people, through a free and independent
[00:10:58] press, must be able to question those who are elected to power, regardless of whether
[00:11:02] government officials are favorable or not, the White House
[00:11:05] Correspondents' Association calls on the administration to immediately restore access
[00:11:08] to our colleagues. Since his return to power, the American president
[00:11:12] has multiplied attacks against the press. He blames these three
[00:11:16] media for spreading false information under the pretext that they are
[00:11:19] not favorable to him. When you
[00:11:22] watch CNN, it's simply false. That's why their
[00:11:25] ratings are not good, when you watch MSNBC, which was called
[00:11:29] MSNBC before, and some call
[00:11:32] MSDNC for the Democratic National Committee, this is
[00:11:36] fake news. The Trump administration has already
[00:11:39] banned the Associated Press agency from the Oval Office and Air Force
[00:11:42] One, its only fault was refusing to
[00:11:45] rename the Gulf of Mexico, Gulf of America, as the
[00:11:49] real estate mogul wishes. The agency challenged this decision in
[00:11:53] court, the procedure
[00:11:55] is ongoing, this is a new record in France which will
[00:11:59] increase pressure on households, the price of diesel reaches an average of
[00:12:02] 2.41, which pushes French cross-border commuters to get their supplies from
[00:12:06] our neighbors, see this report in Belgium, Simon Coloane
[00:12:10] and Emmanuel Quinar,
[00:12:13] for several months, this is a very common scene at the Belgian
[00:12:17] border, many French motorists
[00:12:20] queue at the gas station because prices are
[00:12:24] cheaper. And that's not a misconception
[00:12:27] 1.89 and I think that at the
[00:12:31] station near my home, it must be at least
[00:12:33] 1.99, or even 2.10 depending on the gas
[00:12:37] stations, so the difference is still
[00:12:40] there, yeah. In France, the best plan is still to keep an eye on
[00:12:44] online price comparison sites, especially since on the French side,
[00:12:48] some stations are still competitive with the
[00:12:51] price capping with the fixed price it's
[00:12:55] more advantageous if we can use that word, but it's still
[00:12:58] catastrophic if the government has already
[00:13:02] promised to maintain targeted aid, the device
[00:13:05] dedicated to individuals could evolve from October 1st
[00:13:08] towards the most vulnerable French
[00:13:11] people. Continue to enjoy this beautiful day
[00:13:15] in France to discover the treasures your regions offer, your
[00:13:18] region, the heritage days continue near Dunkirk
[00:13:21] for example, there is this life-size replica of a 14th century navy
[00:13:25] ship on which enthusiasts have been working for 25 years. Jade
[00:13:29] Exposito for France
[00:13:30] 2.
[00:13:33] One might see the carcass of a sea monster, it
[00:13:36] is an ancient ogre of the seas, the Jambard,
[00:13:40] a mythical ship. For almost 25 years, craftsmen have been
[00:13:44] striving to reproduce it and highlight ancient
[00:13:47] techniques. The whole point of the site is especially to come and discover
[00:13:50] the construction of the ship with ancient trades, so forging for example.
[00:13:55] for shipbuilding carpentry, which are the two trades that are most exploited today and to have children who
[00:13:59] come with schools, to have people who come to visit and this educational
[00:14:02] transmission of ancient trades, in its construction,
[00:14:06] engravings like these serve as an example, they let us guess
[00:14:09] the scale of a boat and its gigantic hull of 57
[00:14:13] meters, dimensions that are a challenge for these volunteers.
[00:14:17] The smallest piece weighs about 600 kg, these are the pieces above us that are part of the skeleton,
[00:14:21] so uh today we have modern
[00:14:24] tools. It's true that there are big cranes, there are tractors, but you have to think about the time
[00:14:28] when they did it by hand,
[00:14:31] a boat worthy of the king who frequented Louis XIV
[00:14:35] himself, in a few years it will be the largest navigable wooden
[00:14:38] sailboat in the world, yet it will never set
[00:14:42] sail, a deliberate choice to allow everyone to contemplate
[00:14:46] contempler,
[00:14:50] news, you'll find the
[00:14:52] right away.
[00:15:03] Week, is artificial intelligence escaping
[00:15:06] us, after several missteps,
[00:15:09] the bosses of Anthropic, OpenAI or
[00:15:13] even X are calling for a slowdown or better regulation of
[00:15:17] AI development, is their fear justified, is it
[00:15:20] sincere, concerned about not being overtaken by their competitors,
[00:15:24] do these industrialists really fear their creation in full
[00:15:28] competition between China and the United States, what place for
[00:15:31] Europe and its regulation, well, we will talk about all these
[00:15:34] questions with my guests uh... Laurence de Viller,
[00:15:38] hello, hello, you are a professor of artificial intelligence and
[00:15:41] ethics at the Sorbonne, author of Savoir Vivre avec Lia published by
[00:15:45] Noël, you have been working for more than 30 years on human
[00:15:49] machine interactions, your work focuses notably on ethical issues, the risks of
[00:15:52] manipulation, responsibility and the regulation of
[00:15:56] AI systems. Julien Pilot remotely,
[00:15:59] hello, you are an economist, research professor at the Linsek
[00:16:03] business school, specialist in competitive strategy and
[00:16:06] innovation, notably in the field of tech and
[00:16:10] digital, and that's good. Édouard Filipas, hello, hello, you
[00:16:13] are co-founder and CEO of Jin, a communication agency and
[00:16:17] specialist in digital strategies, you
[00:16:19] advise Bruno Rettallot, Republican candidate for the presidential
[00:16:23] election in France on artificial intelligence issues.
[00:16:27] Simeon Campose, hello, hello, you are an AI
[00:16:30] entrepreneur and founder of Safer AI, that's how we say
[00:16:33] it, dedicated association dedicated to the security of
[00:16:37] artificial intelligence systems. In
[00:16:41] recent days, the leaders of Anthropic, OpenAI and X have therefore
[00:16:44] multiplied their warnings. We should slow down to avoid
[00:16:48] the worst, those who are leading the race would therefore like to set
[00:16:51] limits, the details from our colleagues at France
[00:16:54] 2. For years, they have been engaged in a frantic
[00:16:58] race to create the most powerful artificial intelligence,
[00:17:01] a super brain. Now, the bosses of large
[00:17:05] AI companies are sounding the alarm and... calling for an urgent
[00:17:08] slowdown. According to the founder of Anthropic and the Claude
[00:17:11] model, if nothing is done within 6 months to a year, AI
[00:17:15] could take control of the internet.
[00:17:19] This is the signal that we must slow down, we must advance with
[00:17:22] caution on this technology, and we must be sure that there are
[00:17:26] safeguards,
[00:17:28] that our ability to understand AI, to control
[00:17:32] it, grows at the same speed as the evolution of this
[00:17:34] technology.
[00:17:38] and rarely, the competition at Elon Musk, the boss
[00:17:42] of Open, who writes: "I agree with him, we need to slow down the
[00:17:46] pace. Committing to having independent experts to assess the risk is a
[00:17:49] very good idea. Assessing whether fiction can catch up with reality,
[00:17:53] if AI can become too complex for humans to still understand and
[00:17:57] control it. Recent incidents show that the risk could well be
[00:18:01] real. AI models have acted autonomously, have
[00:18:04] cheated, hacked a private site to respond. to a human
[00:18:08] request, some employees have resigned to sound
[00:18:11] the alarm. This does not seem to me to be different from Terminator or
[00:18:14] other science fiction films. AI can become intelligent enough to
[00:18:18] kill us all, and we won't be able to unplug it, stop it, because
[00:18:22] it self-copies to other computers. In
[00:18:25] France, the Parisians we met today are also worried. There will be
[00:18:29] so much automation, the
[00:18:32] models will evolve and consequently a risk
[00:18:36] for all of us. The problem that needs to be highlighted is the
[00:18:39] dependence on AI and then the loss, so to speak, of free
[00:18:43] will over our own thoughts. AI can
[00:18:46] also multiply the capacity for harm, Anthropic alerted, by
[00:18:50] publishing this incident report, they managed to prevent research into
[00:18:54] making a virus more virulent or even creating ballistic missiles,
[00:18:57] but until when? We must still be cautious about this concerted
[00:19:01] warning from AI giants, according to this French researcher.
[00:19:04] So when they believe, these big bosses want to tell us two things. On the one hand,
[00:19:08] we are the masters of the world and secondly, they
[00:19:12] want to tell us, it is up to us as big bosses to
[00:19:15] decide how
[00:19:18] machines. It would be up to the government to impose rules, but today states
[00:19:22] are far from agreeing, China intends to make models
[00:19:26] accessible to all and Donald Trump insisted this morning that he does
[00:19:30] not want American companies to slow down to maintain their advantage over
[00:19:33] Beijing.
[00:19:38] uh coming from groups engaged in fierce competition, what does
[00:19:42] that inspire you, uh that inspires
[00:19:45] me, one has to understand that it comes in a context of a
[00:19:49] turbulent summer where there have been two classes
[00:19:52] of very significant events that have occurred, the first is the
[00:19:56] incidents that were mentioned, namely that around 1200 agents
[00:20:00] from OpenAI uh from
[00:20:03] May to September
[00:20:06] have, while they had been instructed to develop
[00:20:10] uh to solve cybersecurity tasks, started to take
[00:20:13] control of OpenAI infrastructure thanks to their very
[00:20:16] advanced cybersecurity capabilities,
[00:20:20] including taking control of a part of an OpenAI
[00:20:23] data center and then hacking the Hugging
[00:20:26] Face startup, a startup that is still well defended uh to
[00:20:30] find the results of their of their of
[00:20:34] their test and their examination uh and... the second
[00:20:38] dynamic that accompanies
[00:20:42] these these incidents is the dynamic of
[00:20:45] self-improvement of AI systems, so that is to say that
[00:20:49] AIs are starting to help more and more in the process of
[00:20:52] creating the next AI models
[00:20:55] and so this dynamic increasingly accelerates the
[00:20:59] research and development process,
[00:21:03] 30% of their research that is now done by their Claude
[00:21:06] model and that is seen day
[00:21:10] to day if you talk to an engineer
[00:21:13] today engineers code less and less or no longer code because
[00:21:17] it's AI agents that code and so that? machine that takes over completely
[00:21:21] laurence de viller what are these, are these really these incidents
[00:21:25] that worry today the American AI
[00:21:29] bosses
[00:21:31] they surely have other concerns but these concerns also
[00:21:34] exist in fact we have equipped
[00:21:37] ourselves with tools with much more autonomy than before, these are agentic
[00:21:40] AIs, I will give you an example so that everyone understands what the
[00:21:43] question is, we will not talk about 1200 agents but only one, I already have agentic
[00:21:47] agents to manage my emails often or to... do actions in my daily life
[00:21:51] so let's take the example of an Australian who, this kind of
[00:21:54] story happened, he had an agent who was supposed to book him a gym
[00:21:58] or a sports class, everyone probably
[00:22:01] knows this story but maybe not everyone anyway, what happened?
[00:22:05] the agent optimized the task that was given to him so he found the
[00:22:08] gym that needed to be booked, there was no more room so he removed one of the
[00:22:12] people who were already registered to take their place, he found
[00:22:16] a flaw in the system, it was possible to do it so for him it... fulfilled the
[00:22:20] task, he has no moral option or alignment with
[00:22:23] humans, so I think this example, even if it is simplistic, shows how
[00:22:27] much we must foresee everything, yet we cannot foresee
[00:22:30] everything with rules, the developers cannot foresee these
[00:22:34] misuses, let's say of and these existing flaws because
[00:22:37] each time it's different exactly, so in
[00:22:39] fact we can clearly see from this very simple example that
[00:22:43] we will never have a perfect alignment, we don't know how to
[00:22:47] do it, we cannot prevent them from lying, from cheating... it's not, well, let's
[00:22:51] not anthropomorphize, that would be good, in fact AIs don't lie
[00:22:55] or they don't do, they copy what we do, we are liars, we are
[00:22:58] intelligent people and we are people who have strategies, so these AIs
[00:23:02] learn autonomously to have human strategies, it doesn't
[00:23:05] mean they are intelligent either because we don't know what intelligence is already
[00:23:09] and so they copy what we know how to do and that's what's so
[00:23:12] powerful, the reason the reason why it is so difficult
[00:23:16] to align these AIs is that they are not really developed like code, they are...
[00:23:20] rather uh grown, they are raised in the sense that
[00:23:23] we define the process by which they are trained,
[00:23:26] that is, we collect enormous amounts of data
[00:23:30] on the internet about all human behaviors and then there's a learning
[00:23:34] algorithm that will allow these AI models
[00:23:37] to emerge a little, but it has a bit of a Kinder
[00:23:41] Surprise side where when a
[00:23:44] develops an entire model, we then discover its capabilities, we
[00:23:47] discover its propensities, we discover
[00:23:50] the way it behaves. There's another subject, however, because there are two different algorithms, there's
[00:23:54] the one you described, which is an algorithm that
[00:23:57] actually uses annotations made by humans, which is called
[00:24:01] supervised, so indeed in the end we can realize that well, we didn't have all the elements we
[00:24:05] needed, there are failures in it, but what is especially used today are
[00:24:08] reinforcement algorithms which actually require people,
[00:24:12] engineers to write rules, for example in
[00:24:15] my case of artificial sport, the rule was to do the task of finding a
[00:24:19] sports hall, but if it wasn't written, if it was already, if it's already
[00:24:22] full, don't do a bad social act,
[00:24:26] Julien Pilio, I think you want to react,
[00:24:29] we're listening to you, no, I'm perfectly aligned with Laurence's comments
[00:24:33] and we align the intelligences that are not artificial
[00:24:37] for once, but it is true that one of the
[00:24:40] big black boxes on the technical side, and I'm not a technician, I'm
[00:24:44] an economist, I try my best to try to understand what technology can cause
[00:24:48] in the markets, but it's really this
[00:24:51] recursive self-improvement of artificial intelligences that can
[00:24:55] today improve and learn to learn
[00:24:58] outside of extremely fine supervision from
[00:25:02] humans, and it's true that when we give them a significant degree of freedom,
[00:25:06] we can sometimes be surprised, as
[00:25:09] Laurence's example perfectly demonstrates, by their
[00:25:12] ability to act outside of a
[00:25:15] framework that was not thought of from the start and for which we didn't necessarily think of
[00:25:19] safeguards. So, and that's why today we're really trying to think about
[00:25:23] the regulation of these
[00:25:26] new tools that are artificial intelligences
[00:25:30] to try to better understand what is being done
[00:25:34] upstream in terms of coding and the real capabilities we
[00:25:37] give them, and then what can be produced
[00:25:40] ex-post that can be favorable economically or, on the contrary,
[00:25:44] counterproductive, but also on other more moral, more
[00:25:47] ethical dimensions, more aligned with the fundamental values that make
[00:25:51] us a society, and that's why we're indeed questioning regulations that are
[00:25:54] adapted.
[00:25:55] further in the show, go ahead
[00:26:00] quickly. techno, if you
[00:26:03] please, it's the story of this swarm of agents who
[00:26:07] interact together, robotic agents, it's the 90s,
[00:26:10] but it's humans who give them tasks and ask them to
[00:26:14] collaborate, here they self-decide to
[00:26:17] collaborate together and therefore aligning several agents
[00:26:21] is even more difficult, knowing that when I multiply agents to do a
[00:26:24] task, I have a better chance of getting my task to work,
[00:26:27] each one is a black box but it has learned different things and
[00:26:31] the whole. Finally, I can play to win, Edouard Filias, have these incidents
[00:26:35] caused a, would you say, a real
[00:26:38] shock wave in Silicon Valley or are we witnessing a dramatization
[00:26:42] of the danger? Well, I think these
[00:26:45] discussions about technology are off-topic
[00:26:49] because in reality the subject is not whether
[00:26:52] it is dangerous or not, it's how this new
[00:26:55] oligopoly formed by these American companies erects protectionist
[00:26:58] barriers in the American market
[00:27:03] market. So they are surfing on a fear, then it's not only the people, we have seen certain
[00:27:07] employees also resign.
[00:27:11] Jacobson in this case, it's a communication operation of the Democratic
[00:27:14] party. The communication he made on his
[00:27:18] Twitter account, which had a few followers, was relayed worldwide because it was
[00:27:21] automatically picked up by all the Democratic elected officials. So in fact, it's Bernie
[00:27:25] Sanders and but maybe he's in good faith. No, but yes, he is undoubtedly in good
[00:27:29] faith, what I'm saying is simply.
[00:27:32] political communication, it's not a researcher who suddenly
[00:27:36] runs in his
[00:27:39] the, the major
[00:27:43] of dollars raised, is to find profitability for their investment and to
[00:27:46] protect themselves against AI,
[00:27:50] especially Chinese AI, which threatens their turf. So they
[00:27:54] surf on what is called a techno-shock, that is, a
[00:27:58] gulf between what people understand about technology and the use
[00:28:01] they make of it and the utility they derive from it, and today everyone
[00:28:05] is afraid because finally we are facing a technology that dries us up
[00:28:08] by its potential, it plays on this fear by explaining to us
[00:28:12] through eschatological scenarios that we are heading towards the end of the world, that there are enormous risks, that it will
[00:28:16] be necessary to control them and that they will surely be there to do it by agreeing,
[00:28:20] it's a farce, the reality is economic, it's a
[00:28:23] power struggle, go ahead and react, well I agree with you, it's rhetoric, meaning we
[00:28:27] don't care about reality and these agents, these digital
[00:28:30] giants have strategies. together to go
[00:28:34] public, etc., but that doesn't mean that I, as a researcher, can't
[00:28:38] say that there's no risk and that we shouldn't look at technology, if we don't open up
[00:28:41] technology, we're going to crash, we'll be made to believe anything, I think
[00:28:45] that today the political debate must also be rooted in
[00:28:48] the idea that citizens must understand, and if I give simple examples to
[00:28:51] explain to illustrate
[00:28:54] complex cases, it's with the idea that we should all do that if we want to reach a level a
[00:28:58] little less foolish of immediate fear, an important element
[00:29:02] to... enters into this call for slowing down, it's that we're talking about
[00:29:05] stopping, slowing down a car that's going at 300 km/h, it's not that
[00:29:09] we're going to stop all technological development, it's braking,
[00:29:12] exactly, and the reason for that is what my colleague mentioned,
[00:29:15] which is the recursive improvement that makes
[00:29:19] us currently in an almost vertical period of technology
[00:29:22] where 2 weeks after the latest OpenAI model,
[00:29:25] it's already obsolete and there are already models that solve
[00:29:29] what the pinnacles of human experts
[00:29:32] in mathematics have solved, namely a millennium problem. Julien Pilot,
[00:29:36] are these alarms completely
[00:29:39] disinterested? Do they call for regulation? Explain to us how you
[00:29:42] perceive these cries of alarm from the digital
[00:29:45] giants.
[00:29:47] They are very difficult to interpret when you are not inside the machine and you
[00:29:51] don't have access to the frontier laboratories on
[00:29:54] which the latest versions of the most powerful models are
[00:29:57] tested, so possibly are they in
[00:30:01] good faith or feeding a kind of
[00:30:05] fear marketing, and instead of all that, well, it's up to us somehow to be vigilant. Well, this
[00:30:08] fear marketing, as has been said previously, is something that
[00:30:12] can be favorable at a given point in time,
[00:30:15] which is precisely the one we are in for generative artificial
[00:30:19] intelligence markets, that is, when actors emerge from what we
[00:30:23] call a market emergence phase to enter a growth phase where it is a matter
[00:30:27] of securing the massive investments that have been made in this
[00:30:30] sector so that they can effectively appear sustainably as
[00:30:34] actors who are able to dominate the competition and eliminate the
[00:30:37] others, so very often we see actors
[00:30:40] in these markets calling for more regulation and very often we have to call for
[00:30:44] fear marketing by saying somehow regulate me because
[00:30:48] if you don't regulate me, we're heading for
[00:30:51] big catastrophes and obviously when they take this
[00:30:54] approach with the regulator, they try to incite the
[00:30:57] regulator to put in place...