“I have a love-hate relationship with AI” is how Stanislas Dehaene from the College de France chose to open his speech at The Brain Debate, a panel discussion at the 2022 FENS Forum in Paris, France. The event centred on how recent AI advances could inform and improve neuroscience research and featured an international panel of AI development and theoretical neuroscience experts.
The discussion touched on learning theories that could apply to AI and human brains. Dehaene, who is now a lead scientist at the Human Brain Project with a project dedicated to “the singularity of the human brain”, stressed that, while impressive in some regards, AI is still far from achieving the same type of learning biological brains, in particular human ones, are capable of. “So far, brains learn better: they can reconfigure themselves, form abstract representations and make compositions. And they can do it very fast, inferring general rules from very little initial data. We learn through formulas, models, symbols, and syntax. AI doesn’t work like that - it has a different learning style”.
Throughout the discussion, Dehaene often tied back these unique abilities the brain possesses to its biological structure, connectivity, physical limitations and the ability to share with others, stressing the necessity for AI to mimic these physical characteristics as much as possible to improve its learning capability. “An AI starts with random connectivity, while actual brains rely on initial architecture that informs the learning processes”, he elaborates. “I am impressed with current AI, but they still struggle with syntax and languages in a way our brains don’t. I look forward to AI models that integrate more recent neuroscience discoveries into them”.
Dehaene sees promise in the work of HBP to develop neuro-inspired technology. One of the HBP work packages, WP3, is specifically dedicated to such developments. “I’m expecting the Human Brain Project to achieve better integration of actual data about the human brain, its cellular and connectivity architecture with artificial intelligence models”. The key, according to Dehaene, is all about the actual brain architecture that could map AI networks. “The HBP could achieve an atlas usable as a constraint on artificial intelligence; that would be a huge bonus.”
HBP and EBRAINS at FENS Forum 2022
The FENS Forum 2022 officially opened this Friday in Paris, France. Jean-Antoine Girault, president of the Federation of European Neuroscience Societies (FENS), celebrated that this year's FENS Forum is being held in person, with over 8000 participants. "I am proud to say that this is the largest FENS Forum ever and the largest neuroscience event in Europe. That's thrilling", he said during the opening ceremony.
The Human Brain Project and EBRAINS have a booth at FENS Forum 2022 and prepared a series of activities throughout the event from 9-13 July. Visitors will have the opportunity to learn more about Europe's largest neuroscience project and the first dedicated European Research Infrastructure for the brain.
Text by Roberto Inchingolo