Recreating intelligence in machines has been one of computer science’s grand challenges since Alan Turing. He proposed imitating the human mind and envisioned a symbolic approach. Today, AI has followed a different path, relying on vast datasets and compute in an attempt to brute-force cognition. This approach is based on a loose understanding of the brain, but does it represent the end point for machine thought or a stop on the journey?
In this talk, Prof. James Marshall will propose that the structural limitations of deep learning and technologies such as SLAM (simultaneous localization and mapping) prevent them from robustly solving the problem of “thinking” in a way that would satisfy Turing. Far from being “all about scale,” standard approaches are already presenting diminishing returns, he said. Marshall will present an alternative approach, based on understanding whole-brain function, not of humans to begin with, but simple organisms such as insects. He will explore how solving the problem of ‘solving’ thinking most likely has its roots in solving the problem of moving autonomously in the world.
In addition, Marshall will tie theory with practice by showing the short-term payoff of this approach: substantially more efficient and robust autonomy for robots in the real world.
