In this number
Deciphering ancient Greek, understanding the vocalizations of pigs, controlling plasmas for nuclear fusion, inspecting bridges, offering perfumers new olfactory leads, talking about sexual health… These are the new applications of artificial intelligence (AI) , and more precisely of its most recent avatar, deep learning, which have been announced in the last few weeks alone… Not a corner of our lives escapes the surge, to the point that one wonders where it will stop. Will artificial intelligence one day equal ours? This is probably not for now, because despite undeniable successes, and assuming that we agree on a definition of intelligence, difficulties still arise.
For example, the inner workings of algorithms are poorly understood. At the microphone of Xavier de La Porte, in the podcast “The Code has changed”, Milad Doueihi, holder of the digital humanism chair at Sorbonne University, offered some ideas for improving both the machines and our cohabitation with them. First tip, introduce forgetting mechanisms so that the programs focus on the essentials. Then, take inspiration from children, their endless learning and their acquisition of sociability to take into account the context. Finally, he called for reintroducing a little “stupidity” into these machines so that they leave us a margin of freedom. A next Special Edition will therefore be devoted to artificial stupidity!
by Loic Mangin
p. 10 Great witness: Serge Abiteboul
“I have a problem: I don’t really know what artificial intelligence is”
I. Prowess and promises
p.18 The AI fairy at the bedside of the sick
Algorithms shine at interpreting medical imaging.
p.24 Machine translation owes its success to statistics
Interview with Thierry Poibeau
p.28 Proteins, the new playground for AI
Their 3D shape is calculated better and better.
p.34 Agents play collectively
Philippe Mathieu, Sébastien Picault and Jean-Christophe Routier
There is collective intelligence in multi-agent systems.
p.42 Communication, the two faces of AI
It promotes misinformation, and offers a way to fight against it.
II Work in progress
p.54 Written quiz for AI
How do you make sure the machines understand?
p.60 A single pixel is missing and everything is unlearned
Jean Paul Delahaye
Deep learning mostly lacks… depth.
p.68 Theory under construction
Between architecture and function, we do not know how to make the link.
p.76 Will the computer one day be the boss of maths?
Jean Paul Delahaye
A program does math. From there to becoming a mathematician…
III Get closer to the human
p. 86 Alternative route for the autonomous car
Give the wheel to your car? It’s not yet for tomorrow.
p.94 Living well with machines
Between the machines and us, there is emotion. And there, stop!
p.102 Drawing inspiration from the brain to better imitate it
Julie Grollier and Damien Querlioz
On the energy saving side, our gray matter is a model.
p.112 The unlikely artificial consciousness
A conscious machine, is it possible? We can doubt it.
p.120 The essentials
Leaf through an excerpt from this issue:
We want to give thanks to the writer of this article for this incredible content
Special issue For Science n°115 – May 2022 – How far will artificial intelligence go?
Check out our social media profiles , as well as other pages that are related to them.https://www.ai-magazine.com/related-pages/