Toward Intelligence on a Chip

The common theme of traditional approaches to machine intelligence is to devise mechanistic explanations of brain function based on available understanding and translate them into algorithms that can be implemented in... [ view full abstract ]

The common theme of traditional approaches to machine intelligence is to devise mechanistic explanations of brain function based on available understanding and translate them into algorithms that can be implemented in conventional computing machines.I propose that, instead of attempting to understand and emulate the brain's learning functionality, we should regard the brain’s tremendous learning ability as a byproduct of its self-organized structure and dynamics and redirect our efforts toward emulation of the self-organization processes that produce this structure. Let us identify materials with the capacity for memory, heterogeneity, plasticity, connectivity, power efficiency, etc., and employ these self-organizing processes to grow a brain-inspired structure on a chip, using the brain’s features as markers and guidelines.