The new generation of neural networks finds its applications not only in content generation or translation. She is already capable of saving lives.
Generative AI embodies a new phase in the rise of artificial intelligence. With this new category of neural networks, it is no longer a question of predicting a new trend, but of creating it. Deep learning models designed to produce images of instructions have begun to democratize the trend. Among them are the now famous Dall-E, Stable Diffusion or MindJourney. Their results are impressive (read the article Image Generation AI: JDN Test Reveals Surprising Results). Another step has just been taken with the video generation models unveiled by Meta and Google consecutively in September and October.
The first applications of generative AI date back a few years. In 2018, the very first book written by an AI was published: “1 the Road”. Trained from classics of Anglo-Saxon literature, the underlying neural network is imbued with their vocabulary, structure and style. The last text tells of a road trip from New York to New Orleans. It is based on data captured by a car, equipped with a camera and a microphone, that traveled the more than 2,000 kilometers between the two cities. “In September 2020, the first image of a black hole will be generated by an AI (from billions of information from 8 radio telescopes around the world, editor’s note)”, recalls Lobna Karoui, head of the data & artificial intelligence center of excellence at Capgemini (read the article published in Nature).
From comics to cinema
Then, in late 2021, Steve Coulson, creative director of transmedia storytelling agency Campfire, produced the first comic based on images created by a generative AI. This odyssey combines fantasy, fantasy and science fiction and is divided into three parts: Summer Island, Exodus and The Lesson. At the same time realistic, precise, but also coherent in narrative and style, the drawings give the impression of being born from the imagination of a sensitive and seasoned artist. Finally, in August 2022, Jason Allen won first prize at the highly anticipated Colorado State Fair in the “digitally manipulated photography” category for a work produced thanks to Mindjourney. “Generative AI is the creative tool that allows the machine to express its imagination,” summarizes Lobna Karoui.
“Generative AI is the creative tool that allows the machine to express its imagination”
At the same time, numerous publishers are integrating generative AI into their offerings. Among the industry heavyweights, in October 2021, Microsoft unveiled GitHub Copilot, an intelligent assistant designed to translate technical specifications into lines of code. In collaboration with the French Hugging Face, ServiceNow will follow in September 2022 with the open source project BigCode. At the beginning of October it was Salesforce’s turn with CodGen. “Our technology not only promises to save experienced developers time, but also enables non-developers to build applications by exposing their needs,” promises Salesforce.
In mid-October, Adobe announced its intention to integrate generative AI into Photoshop at its global MAX event. “It will make it possible to break down an existing image into variations, or even add new objects or new sections to it by describing them in natural language,” the group points out.
In France, the Strasbourg start-up Imki has made generative AI its trademark. Among her accomplishments, she relied on a GAN-type AI (for Generative Hostile Network) to expand Monet’s canvas beyond the original frame, as shown in the video below.
Another realization of Imki: a compelling projection on the whole of the ancient theater of Orange. “By creating multiple variations based on a few basic images, generative AI allowed us to obtain a very large amount of customized content in record time. Thanks to the Unreal Engine 3D game engine, this content then adapted to the dimensions and relief of the ancient theater (according to technique video mappingpublisher’s comment)”, explains Frédéric Rose, director and founder of Imki. In the end, 25 synchronized video projectors draw a gigantic animated fresco in the heart of the Roman monument.
Even more impressive is that Imki made a docufiction on behalf of the municipality of Nîmes in which you are the hero. With the city’s Roman monuments in the background, it tells the story of a gladiator. “Using generative artificial intelligence, visitors can model their image on certain characters. The AI seamlessly integrates their face instead of the actor’s, then orchestrates the lip sync and facial expression based on the text and story,” explains Frédéric Rose from. .
Plan for the unexpected
In addition to creative professions, generative AI is used more prosaically in automatic text translation. “In the automotive or aerospace industry, it also optimizes predictive and preventive maintenance,” underlines Capgemini’s Lobna Karoui, before specifying: “It allows datasets to be generated to identify rare failures, but those that can occur. C This is with especially the case in the test of autonomous cars, where it will produce huge amounts of traffic configurations to simulate and anticipate the maximum number of dangerous situations, especially those that are not present in the data history.” The same applies to the physical security of public buildings. “Shopping malls or amusement parks, for example, use it to represent risky behavior that has not yet taken place,” Lobna Karoui illustrates.
“Generative AI is potentially applicable to any type of design, from clothing making to architecture to printed circuit boards”
And Frédéric Rose adds: “Generative AI is potentially applicable to any type of design, from clothing design to architecture, including printed circuits. It will simply have to integrate the constraints specific to each area: constructive feasibility, economics, robustness of materials. ..” In this logic, Imki starts working on new scenarios aimed at the luxury and network industry.
The pharmaceutical sector has also taken up the subject. In its latest chapter to date, Sanofi announced on November 8 that it had signed an agreement with Insilico Medicine, a start-up that puts generative AI at the service of new drug discovery. Thanks to this method, she has already developed treatments against cancer or fibrosis. Through this partnership, Sanofi plans to invest $21.5 million in Pharma.AI, the AI platform developed by Insilico Medicine. An environment that ingests millions of data samples to identify disease signatures and pinpoint the most promising molecules from billions to target them. Sanofi’s objective through this agreement: to shorten the development time of a new drug by several years.