IIT Bombay is organising Institute Lecture on 'The embodied self: the ultimate Turing test for AI? How sensory and motor predictions shape human perception and cognition' on Monday, December 19, 2022.

Details of the lecture are given below:

Title: The embodied self: the ultimate Turing test for AI? How sensory and motor predictions shape human perception and cognition.

Speaker: Prof. Sylvain Baillet, Professor, Neurology & Neurosurgery, Biomedical Engineering and Computer Science, Canada Research Chair (Tier 1, CIHR) of Neural Dynamics of Brain Systems, Montreal Neurological Institute (MNI), McGill University, Montreal, Canada

Abstract: Am I myself because my brain circuits integrate and compute information in a unique way? How does this explain the way I behave? These are possibly the hardest, and largely unresolved questions for neuroscience. How we approach these questions has changed in recent years, with the conjunction of theoretical constructs, large data ensembles, powerful experimental techniques combined, and the tremendous growth of data science and learning algorithms.

In this talk, I will review how these approaches have recently contributed to clarify the nature of the tension between our sensory environment and its mental representations. I will explain how some key neurobiological mechanisms can be studied via the measurement of polyrhythmic patterns of brain activity that develop in hierarchical brain networks. I further propose a mechanistic framework of brain activity that implements a generic form of contextual predictive inference of input signals to brain networks. This view is aligned with the principles of active inference, which predict that spontaneous brain activity during wakefulness constantly implements the self’s representation of its environment and the possible consequences of its actions. Inspired by this framework, I will review more specifically a series of recent findings that account for this hypothesis in a diversity of brain functions. In particular, we recently proposed to train artificial neural networks on naturalistic stimuli to produce encoding models of neural activity that account for contextual uncertainty and prediction errors in perception. I will show how we used this approach to reveal the brain signaling pathways for natural speech processing.

Overall, I believe that the time is now to inform new computational models for AI inspired by the most current developments in systems neuroscience.

About the speaker: Prof. Sylvain Baillet is a global leader in multimodal brain imaging and large-scale electrophysiology for systems neuroscience. He has developed pivotal methods for brain imaging and quantitative electrophysiology, with an emphasis on time-resolved brain imaging techniques, with strong interest in practical transfers to healthcare. He has published 145 articles, 9 book chapters (the majority as primary or senior author), >200 conference proceedings, 3 international patents, 200 invited lectures (~100 international). Prof. Sylvain Baillet's publications have been in leading journals, from Neuroscience to Computer Science, and top-tier outlets e.g., Nature Communications, Nature Neuroscience, Nature Methods, Science Advances, Neuron, PLoS Biology and PNAS that have earned a high level of citations: ~15,000 (27 articles over 100 cites each; Google-Scholar), with 4 articles in the Top 1% of most-cited publications in Neuroscience & Behavior (Web-Of-Knowledge). He has served on 13 journal Editorial Boards, and is presently Associate Editor of Science Advances.

As a mentor and educator, Prof. Baillet has trained >100 international students and postdocs with diverse backgrounds from Neuroscience to Computer Science. He has taught across the globe in Europe, North America, Australia, Japan, mainland China, Taiwan, Chile.

Prof. Baillet founded and led research and clinical imaging multimillion-dollar platforms: the MEG program he created in 2008 was recognized by US health-insurance partners and is now part of the routine pre-surgical workup of dozens of epilepsy and brain-tumor patients annually. The MEG core he founded at McGill in 2011 has become the second busiest scanner on campus, after 3T MRI. He has led the McConnell Brain Imaging Centre, the largest of its kind in Canada, for which he recruited 6 new PI leaders and 9 HQPs and raised $6.4M in infrastructure grants in 4 years.

He also has a track record of patented intellectual property and successful industrial transfers (co-invention of 3 patents, Scientific Advisory Board membership). For example, his patented MRI work for the clinical evaluation of acute stroke was transferred to the biomedical industry, used in national clinical trials, obtained CE and FDA certifications, and is now marketed worldwide in a clinical radiology commercial software suite.

Contribution to Open Science: Prof. Baillet co-created the Open MEG Archives (OMEGA, 2015) the first open-access data repository for magnetoencephalography (MEG; 800 international users, 60 journal articles). His team also hosts the software integration of Brainstorm, an international open-source neuroimaging software project recurrently sponsored by the NIH (4 consecutive R01 grants), with a community of  more than 32,000 users.