Brain-Machine convergent evolution: a window into the functional role of neuronal selectivity

Central nervous system neurons manifest a rich diversity of specializations and selectivity profiles. However, the precise roles these properties play in brain function has remained the domain of intuition and educated guesswork. Here we propose a different approach to address this central question: searching for insights that can be gained from identifying parallel functional properties that emerge in both brain and artificial, engineered, systems. The central rational for this approach is that functional similarities observed in systems whose origins are so disparate are unlikely to be mere coincidences. To the contrary, these parallel properties point to critical bottlenecks that force the convergent evolution of widely diverse systems towards similar functional properties. In this review, we will focus on the mammalian cerebral cortex and present examples derived from three levels along the cortical hierarchy of visual processing: oriented receptive fields in primary visual cortex, relational geometries in high order, face selective, cortical areas and place-related entorhinal grid-cells at the top of the cortical processing hierarchy. In these three cases, intriguing parallels appear between the functional properties of cortical neurons and successful engineered systems. Such coincident properties highlight their significant functional role. Thus, contrary to the common intuition that artificial brain models are more informative as their basic elements are made more brain-like, here we propose that searching for parallel convergences with artificial systems could provide insights into brain function particularly in those cases that are far removed from realistic brain biology.