And if the artificial intelligence does not mean the decline and the negation of the human being, but a way of thinking about our freedom ? It is in any case one of the tracks selected by Thomas Solignac, 28 years old, who has created in 2016 in Paris the software publisher Golem.’ve. Indeed, this graduate of Epitech who is trained in the philosophy of the social sciences at Paris-Nanterre takes his distances with the ” deep learning “, a learning technology very much in vogue at this moment, which relies on a large number of data. This “deep learning” was primarily popularized by the French Yann Le Cun, and the Canadian Yoshua Bengio and Geoff Hinton in the course of the last forty years.
before being ? “It is architectures of artificial neural networks that learn to represent data in a hierarchical manner,” says in effect to the Point Yann Le Cun, a researcher at the university of New York. “The machines are learning to represent the world with multiple levels of abstraction. “To do this, they swallow a large number of data that pass through the sieve of algorithms of discriminant analysis. Interesting but sometimes tedious : in fact, the “deep learning” is based on a large number of examples that must be the most often characterize manually, which can be repetitive, and therefore not very long.
also Read Yoshua Bengio, the master to think of the machines
However, there is no one best way to teach a know the machines, ” Thomas Solignac, who cites the example of the debate that opened in 1975 at the abbey of Royaumont Jean Piaget and Noam Chomsky around the learning of the language. For the epistemologist switzerland, the innate is almost non-existent, so that, for the american linguist, there are a large number of cognitive structures common to all individuals. “The first approach characterizes the deep learning which considers that a machine can learn it all from scratch in “chowing down” a large number of statistics, when the second is based on a vision of the world that is more qualitative than quantitative, ” says Thomas Solignac.
go the distance compared to the deep learning
This last approach, which is inspired by Thomas Solignac, allows the machines to have a better understanding of the modular world. “Take a company that uses an AI to respond quickly to e-mails. When the environment changes radically, as was the case with the onset of a pandemic, the second approach allows programs to be more responsive. “The passion of Jacques Ellul offers a technology that does not require training, she already knows the functioning of human language, and this, in an innate way.
The installation time is shortened, because it adds only the knowledge specific to the enterprise : the products, methods of manufacture… It is possible, because the artificial intelligence of the Golem.’ve already built a few bases of the functioning of language. Where the learning machine tries to reproduce the average behavior observed in the examples, the approach of Golem.’ve, the more close to the AI, so-called symbolic, embeds elements of modeling human reasoning.
Orange : the difference between the fruit and the color
“This allows, for example, to resolve an ambiguity between an orange, the fruit and the color, without ever having ingested sample,” explains Thomas Solignac, who adds : “To understand a text, it is necessary to take into account the context performative, that is to say the possible actions in a given context. “(1) It is also a way for him to reduce the number of through algorithmic that go hand in hand with the digestion of biased data by machines, when it has recourse only to the ” deep learning “. Wishful thinking for a company that does not account for the time that the 26 employees ? Of Course, Golem.I do is that David and Goliath compared to the giants in american Google and chinese Baidu, but its approach could strongly influence them in the future.
read also : Yann Le Cun, Stanislas Dehane, the clash of the brains
(1) Read about “When to say is to do” John Langshaw Austin (Threshold)
writing will advise you
artificial Intelligence : “It is necessary to distinguish what is possible from the fabrication “
