TheChek News Flow

Créativité et intelligence artificielle : «Pourquoi il faut (re)lire Marguerite Yourcenar»

FIGAROVOX/CHRONIQUE - Dans Paradoxe de l'écrivain, publié en 1983, Marguerite Yourcenar s’interrogeait sur la place de la créativité humaine dans le roman. Un livre plus que jamais d’actualité à l’heure de l’essor des intelligences artificielles génératives, explique notre chroniqueuse Aurélie Jean.

Aurélie Jean est docteur en sciences et entrepreneuse. Elle a notamment publié Les algorithmes font-ils la loi ? (L'Observatoire, 2022) et coécrit Résistance 2050 (L'Observatoire, 2023). Dernier livre paru : Le code a changé. Amour et sexualité au temps des algorithmes, aux éditions de l'Observatoire.

«Ils sont étrangement uniques car ils ont su tirer étrangement parti des circonstances et des faits et des émotions qui avaient été les leurs. Et c'est ce qui fait qu'ils sont reconnaissables.» Ces mots de l'écrivaine Marguerite Yourcenar résonnent chez quiconque a déjà créé dans sa vie… c’est-à-dire nous tous ! Et oui ! Même enfant on crée à travers un dessin, une maison faite en lego ou en pâte à modeler, ou encore une histoire entièrement imaginée dans une mise en scène qui incluent des dinosaures, des poupées, des voitures ou encore des peluches.

En pratique, quand on crée, on fait appel à ce qu'on nomme des ressorts et une ou plusieurs intentions. Les ressorts caractérisent le point de départ de l'acte de créer qui représente celui qui crée c’est-à-dire son histoire, et ses expériences. Les intentions caractérisant ce vers quoi on va, à savoir la raison concrète de l'acte de créer comme faire passer un message, rendre les choses belles, ou encore s'amuser. Peu importe la nature des ressorts et des intentions, leur existence est nécessaire à la création qui est alors le propre de l'être humain qui, contrairement à la machine (et donc à l'IA), maîtrise toutes les intelligences. En effet, la machine ne maîtrise que (le résultat de) l'intelligence analytique alors que nous - êtres humains - maîtrisons les intelligences émotionnelle, créative et pratique.

Certains aimeraient nous faire croire que l'humain deviendrait obsolète face à la machine toujours plus efficace, mais c'est oublier de préciser que cette même machine est déjà plus intelligente que nous analytiquement (et c'est tant mieux !). En revanche, elle ne maîtrisera jamais les autres composantes de l'intelligence bien humaine. En cela, une IA générative n'écrit pas ou ne crée pas un texte mais le génère comme son nom l'indique… et les écrivains le savent bien.

La pensée de Marguerite Yourcenar est intéressante car elle soutient d'une certaine manière l'idée que le style d'un auteur ne peut en aucun cas être utilisé sans en évaluer l'usage au sein d'un modèle d'IA. Le cas contraire, le plagiat en est la conséquence évidente. En pratique, une IA générative comme ChatGPT, entraînée sur un jeu de textes pantagruéliques, fabrique ce qu'on nomme des corrélations et des analogies statistiques sur ces textes en détectant des signaux faibles et forts, pour en retour générer un texte en réponse à une requête. À ceux qui affirment qu'une IA générative est capable d'écrire du Proust ou du Baudelaire, on leur répondra que ce n'est pas comprendre le besoin d'intentions et de ressorts chez celui ou celle qui écrit et qui n'existent aucunement chez la machine. Mais aussi qu'un style bien identifiable doit pouvoir être valorisé au sein des bases de données utilisées pour construire ces IAs génératives afin d'écarter le risque de plagiat, et la violence qui va avec. On comprend alors les discussions actuellement en cours où les écrivains, les scénaristes ou encore les journalistes revendiquent leur droit de protéger leurs créations… et leur style !

La pensée de Marguerite Yourcenar, il y a presque quarante ans, nous démontre également qu'il faut toujours contextualiser les grandes questions et préoccupations de notre époque dans les temps longs… On y trouve souvent des réponses qui nous éclairent, des interrogations qui nous bousculent ou encore des espérances qui nous font vivre. Marguerite Yourcenar avait raison, et on lui dit merci.

It’s been hard to browse tech headlines this week and not read something about billions of dollars being poured into data centers.

The world’s largest asset manager, Blackstone Group, reportedly plans to invest $8.2 billion to develop data centers in Spain. Energy and data center company Crusoe Energy Systems announced it raised $3.4 billion dollars in debt financing — and reportedly will raise an additional equity investment led by Founders Fund — for a data center in Texas that will be used for OpenAI. And pension fund AustralianSuper announced it has committed $1.5 billion to become a minority owner in DataBank, a provider of enterprise-class data centers across North America.

Of course, all of that is just this week and doesn’t even touch on the fact that Amazon plans to invest $100 billion in AI data centers in the next decade, nor the planned OpenAI and Microsoft joint data center project that is expected to cost $100 billion.

That heightened level of investment seems to be starting to bubble up to the startup realm. So far this year, $1.3 billion has been invested in database-related startups — those that provide connectivity, efficiency or other needed tools/solutions for the centers — per Crunchbase data.

The raises by data center-related startups this year already nearly matches the total raised in 2023 and could even reach the nearly $1.9 billion similar startups received in 2022 if the last two-and-a-half months of the year are strong.

Of course, this week helped out a lot, with Lightmatter locking up a $400 million Series D led by new investor T. Rowe Price at a $4.4 billion valuation. The startup uses light to link chips together and to do calculations for the deep learning necessary for AI.

The week also saw Xscape Photonics — a startup also using photonics technology to address the energy, performance and scalability challenges of AI data centers — raise a $44 million Series A led by IAG Capital Partners and with investment from the likes of Cisco Investments and Nvidia.

It seems like it should only be a matter of time before even more rounds into similar data center-related startups are announced as Big Tech pours billions into the new data centers needed to keep the promise of AI alive. Those centers will need new innovation — especially when it comes to tackling the energy consumption problem — and it is likely Big Tech and VCs will be there to provide the cash necessary to nurture those new technologies.

DAX:

SDAX:

Earlier today, we published our Generative AI Takeover List of 78 startups that could be acquisition targets. The list includes companies that develop their own models or robots, haven’t raised funding in at least two years, or both. These companies are most likely to face a cash crunch and be open to takeover offers.

My colleagues Stephanie and Rocket highlighted the conundrum for these startups: It’s expensive to build large language models and hardware, and customers are cautious of spending big on AI apps and tools with unproven returns.

A surprising company on the list is Mistral AI, the Paris-based developer of open-source models, founded early last year by alumni of Google DeepMind and Meta Platforms. The startup has found success among investors—and politicians—based in part on the notion of being a European alternative to OpenAI.

In June, Mistral raised $640 million in new funding, bringing its total funding to $1.2 billion, more than all but a handful of AI startups. That funding valued the company at about $6 billion.

Just because Mistral has raised a lot of funding doesn’t mean a takeover is unlikely: LLM developer Inflection, for instance, raised $1.6 billion but effectively sold itself to Microsoft in March—long before it had a chance to burn through that cash.

“We are an independent company and intend to remain so,” a spokesperson for Mistral told Stephanie and Rocket.

Mistral’s founders and researchers have AI expertise that big technology companies covet. Arthur Mensch, its CEO, worked on Gemini and Flamingo, a multimodal model, at Google. Timothée Lacroix and Guillaume Lample, its chief technology officer and chief scientist respectively, worked on the first generation of Meta’s flagship AI model Llama.

But Mistral’s path to revenue and profit is unclear. Its open-source approach to AI means it doesn’t sell access to many of its most popular models, while facing stiff competition from open-source model developers such as Meta.

Mistral generates some revenue by charging users of its application programming interface, but fewer than 10% of users pay for Mistral’s larger models at one of its partners, Stephanie previously reported. Mistral could also generate revenue from helping companies develop AI apps that use its models, the way Germany-based Aleph Alpha does after it slowed efforts to compete against OpenAI in developing advanced models. Mistral hasn’t indicated it is considering that type of business.

It’s also unclear how another major developer of open-source AI, Meta, will make a return on the tens of billions of dollars it is spending on its efforts. Meta, however, rakes in huge amounts of revenue from digital advertising and has bet that its investment in AI will improve its ads business.

The list of companies that could acquire Mistral is short. Amazon seems to have an appetite for license-and-hire deals, having struck them with Adept, a developer of computer-using AI agents, and Covariant, a robotics software startup. Apple has shown interest in M&A too; my colleague Jon previously reported that its executives have met with at least three AI startups about potential acquisitions, and it quietly acquired French AI startup Datakalab in December, French business magazine Challenges reported.

Alternatively, Mistral’s founders could mimic Character.AI’s founders and return to one of their alma maters—Google or Meta—in return for a handsome payout. We could also imagine the startup following Inflection into the arms of Microsoft, especially since Microsoft invested in Mistral in February. A spokesperson for Microsoft said Mistral isn’t a candidate for acquisition. Amazon and Meta declined to comment. Apple and Google did not respond to requests for comment.

Mistral could also pursue a strategic partnership involving more financing, revenue sharing or joint research efforts, like Cohere has discussed with Snowflake or OpenAI has inked with Microsoft. In the leadup to its deal with Google, Character held research partnership talks with Google, Meta, OpenAI and xAI.

Mistral’s valuation may be a tough pill for potential acquirers or strategic investors to swallow, though. Inflection, Adept and Character were valued closer to $1 billion when they struck their deals. It could be challenging for a Mistral acquirer to pay a licensing fee large enough to please shareholders and small enough to avoid regulatory scrutiny.

But any deal still may face heat from regulators in both the U.S. and the EU. Mistral is one of Europe’s few AI leaders, even more so after three of the five founders of France’s H left the startup and Germany’s Aleph Alpha lowered its ambitions.

Europe’s top antitrust regulator said in February it would examine Microsoft’s investment in Mistral. And in an interview with CNBC in May, French president Emmanuel Macron said he “would prefer [Mistral] to become independent and grow on their own.”

Mistral on Tuesday released two models called les Ministraux, which are small enough to run on devices such as phones, instead of in the cloud. It said those models perform better on certain benchmarks than similarly sized ones from competitors. That performance, and the success of other models from Mistral, is another reason why it could be a desirable takeover target.

Bottom line: Don’t be surprised if Macron doesn’t get his wish.

Stellantis-owned brand’s painful EV transition has shaken home city of Turin and sparked acrimonious spat with Rome

At 21, Gianluca Rindone got his first job making bodies for Fiat’s Maserati cars alongside his father, a Sicilian who had been lured to Turin’s auto boom in the 1970s. He thought it would be a lifelong career but after three decades Rindone has been furloughed, falling victim to the carmaker’s decision to suspend production at its last Turin factory.

Rindone’s own sons — aged 15 and 8 — are unlikely to follow his path into Fiat’s workforce, he says.

“If there is not a job for the father, how can there be a job for the son?” frets the 48-year-old, who is relying on help from his retired parents to pay his mortgage and bills. “When you see a factory with nearly 100 years of history stop, the heart cries. If Stellantis goes, Turin dies. It’s as simple as that.”

Rindone’s lament is one heard across Europe, as the continent’s auto industry — and its 14mn jobs — faces an existential crisis, squeezed between the soaring costs of developing cleaner vehicles to meet the EU’s tough emissions standards and the cheaper models from Chinese rivals.

Fiat-owner Stellantis, which also makes the Peugeot, Citroën and Jeep brands, last month warned on profits, while Volkswagen, Europe’s largest automaker, is considering closing down plants in Germany for the first time in its history.

Once the pride of Italian manufacturing and known as “Mamma Fiat” for its extensive cradle-to-grave welfare schemes, Fiat was subsumed into Stellantis in 2021, after years of troubles.

The company last month temporarily halted car production at Fiat’s historic Mirafiori plant due to lack of consumer demand for the Fiat 500e — the electric version of the iconic vehicle that democratised Italian car ownership in the 1960s and 1970s. Initially scheduled to last four weeks, the plant’s shutdown has been extended until at least the end of October.

The crisis has far-reaching political consequences for Italy’s Prime Minister Giorgia Meloni, who is battling to boost economic growth and stabilise Italy’s fragile public finances, as well as for the whole of the EU.

The Italian car industry, including its vast network of parts suppliers, employs some 250,000 people and accounts for over 5 per cent of gross domestic product. But since 2018 Italy’s total car production has halved to 500,000 vehicles.

On Friday the country’s auto workers are descending on Rome to demand Brussels, which has banned the sale of new combustion engine cars after 2035, provide more financial support for the green transition. The protesters also want Meloni’s government to work with Stellantis, the country’s only large carmaker, to tackle the sector’s challenges to save their jobs.

But Rome and Stellantis disagree on how to do this. Meloni has remonstrated against the production of Fiat models outside Italy, exacerbating tensions.

Italian industry minister Adolfo Urso has been pushing the EU to reconsider the looming curbs on combustion engine sales but Stellantis’ CEO Carlos Tavares disagrees on whether this is the right approach.

“Instead of arguing about the regulations, it’s better to work hard to comply in the most efficient manner,” he told Italian lawmakers at a parliamentary hearing last week.

Tavares also complained about Italy’s high energy prices and Rome’s paltry spending on incentives to support purchases of electric vehicles by the middle class — which he said was just a fifth of what other European countries had spent.

Yet he insisted Stellantis was committed to making cars in Italy, calling Fiat’s factories a “strong asset” for the company.

“We love our plants and believe they are the right answer to the challenges we have in the future,” he said. “That’s why we don’t sell them to the Chinese.” 

Such blandishments ring hollow to autoworkers in Turin, where Fiat was founded in 1899, and which prospered during Italy’s postwar car boom. “Turin was constructed for Fiat, but today we don’t have Fiat. Stellantis is a multinational company, and we know [they] go where costs are cheapest,” Rindone said

Fiat’s Turin workforce has already been steadily declining since the late 1990s, with four other factories closing in recent decades. While Stellantis denies it plans to close Mirafiori, many workers suspect the company wants to quietly wind down the plant by waiting for its labour force — whose average age is now 57 — to retire.

Many workers have already accepted generous buyout offers.

“We are the last generation hired,” said furloughed worker Fabbio Mosesso, 53, who joined Fiat nearly 36 years ago. “The business is not spending money to incentivise EV purchases, but to push people to leave their jobs.”

Autoworker and union representative Giacomo Zulianello, 58, accused Stellantis of “bleeding us”, adding that “the complete killing of Mirafiori is too much even for Tavares. But in reality, Mirafiori is already closed.”

Michaela San Filippo, 51, a second-generation Fiat assembly line worker also on furlough, said she and fellow workers were struggling financially, even skipping on medical expenses. “You have to renounce everything that is not necessary,” she said. “You don’t live; you just survive.”

Stellantis has put new businesses into Mirafiori, including an EV battery testing centre, a unit recovering and recycling car parts, and a new assembly line for transmissions for Fiat 500 hybrids, due to start production in 2026.

Part of the factory also houses an exhibition of vintage Fiat cars. Yet marking the 125th anniversary of Fiat’s founding this summer, Urso expressed anguish at the factory’s diminishing output.

“Fiat is Turin,” said the minister, a member of Meloni’s rightwing Brothers of Italy party. “It was the greatest industrial plant in Europe, and we can’t accept that it is becoming a mere industrial museum.”

Not everyone is pessimistic. Stefano Lo Russo, Turin’s mayor, said the city was undergoing a transformation not dissimilar to the era when cars replaced horse-drawn carriages.

“When we moved from horses to cars, we stopped having a horse supply chain in the city and we started to have engine mechanics,” Lo Russo said. “It was progress. New technologies mean change.”

He said Turin’s carmaking history and its engineering colleges were supporting a vibrant new aerospace industry, underpinned by activities of France’s Thales and Italian defence champion Leonardo.

“The real vocation of the city is not only automotive — Turin is the city of engineering and manufacturing,” Lo Russo said. “We can work on cars, satellites, aeroplanes. More than a city of cars, is a city of engineers.” 

David Avino, a former Italian military officer and software engineer, chose Turin to launch his space venture Argotech in 2008, offering space services and astronaut training as well as making small satellites. Employing around 170 people in the city, Argotech is actively recruiting, although Avino said the space sector needed at least another decade to grow to scale.

“If everyone keeps investing . . . it will be effective in 10 or 15 years. But it’s important to start,” he said.  

Andrea Giordano, 36, joined Argotech after his previous employer, a leading car components company, furloughed its Turin labour force earlier this year.  

“My hope is that like the auto boom of the 1970s, this city will have the aerospace boom,” said Giordano, a warehouse operator. “In the meantime it’s going to be hard. We are going to have to tighten our belts.”