A few years ago, the human brain served as the inspiration for building the first synthetic neural networks, one of the types of artificial intelligence (AI). Now AI is generating new knowledge that is consolidating our understanding of how the brain works—knowledge that in turn is proving useful for improving AI. Dr. Karim Jerbi, Associate Professor in the Department of Psychology at Université de Montréal and one of our collaborators, is among those exploring the links between neuroscience and AI.
In neuroscience, electroencephalography (EEG), magnetoencephalography (MEG) and other techniques generate thousands of data points that can be fed to machine-learning algorithms. Dr. Jerbi’s team is working with, among others, a database that stores EEG recordings along with evaluations of depression from thousands of subjects. Using these data, a classification algorithm can “learn” to distinguish between profiles of depressed and non-depressed individuals, and then perform a diagnosis. Dr. Jerbi’s interest lies not so much in diagnoses, however, as in “opening up the algorithm’s black box” to understand how it classifies data: what properties of the EEG recordings does it use to construct its classification process? This provides clues to better understanding depression. Another advantage of his approach is that it is data-driven; it therefore does without the a priori considerations involved when working with a hypothesis. “Rather than make hypotheses about the involvement of a particular brain structure and its functional role, we can explore the entire space of cerebral activity and benefit from a more exhaustive approach,” Dr. Jerbi explains. AI is thus becoming a tool for, and an accelerator of, basic research in neuroscience.
Artificial neural networks and machine learning can serve as models for understanding the human brain, but at the same time, the biological brain can serve as a model or an inspiration for artificial intelligence.
Associate Professor, Université de Montréal
In a similar vein, the artificial neural networks used in deep-learning algorithms can be harnessed to perform visual recognition tasks. Since these networks are inspired by our human brain structures, they can serve as models for understanding how the visual cortex processes images. “By looking at what’s happening in the various layers of these artificial neural networks, we can learn how the human brain performs categorization,” Dr. Jerbi adds. “This is a new source of knowledge emerging from artificial intelligence that we can go and verify by looking at the human brain.”
More generally, AI can therefore improve our understanding of information processing and the foundations of human intelligence, which in return will open up new avenues for the development of AI.
Because specialists in both fields are interested in learning, information processing, memory, intelligence, and so on, AI and neuroscience enrich each other. With Montréal being a global leader in both disciplines, it is only natural that it should be a focus of synergy between them, as exemplified by the UNIQUE (Union neuroscience et intelligence artificielle – Québec) research cluster, led by Dr. Jerbi, among others, and supported by IVADO. This grouping of international-calibre researchers is dedicated to the idea of merging the two fields to create a new discipline, neuro-AI, of which UNIQUE would be a world-leading exponent.