Can a Neuroscientist Understand a Microprocessor?


Today’s Wired asks a question about what neuroscience techniques can actually tell us, referencing a 2017 paper by Eric Jonas1 and Konrad Paul Kording titled “Could a Neuroscientist Understand a Microprocessor?”

According to Wired:

This January, a video game chip started a scientific reckoning. It all began when some “microchip archaeologists” photographed the chip—the MOS 6502 microprocessor that lived inside Atari—and built a digital model of its interconnections. Then some neuroscientists put it to the test. One by one, they knocked out the transistors in their map, trying to get at what the circuit was for. It’s similar to what neuroscientists do when they lesion a part of the brain, or silence single neurons. Their project was simple: Could they use the arsenal of neuroscience methods to get at the function of a simple circuit?

They failed. Miserably. The scientists’ experiments didn’t produce much information about Donkey Kong, Space Invaders, or Pitfall—just which transistors you could knock out and turn the game off. The result was damning for researchers pursuing the connectome, a bottom-up recreation of all the brain’s interconnections. To the neuroscience community, the message was clear: Brain scientists may have plenty of bottom-up data about the brain, but they’re far from using that data to understand how the organ works.

This reminds me of Valentino Braitenberg’s Vehicles: Experiments in Synthetic Biology. John Wiseman gives a great summary:

Valentino Braitenberg describes a series of thought experiments in which “vehicles” with simple internal structure behave in unexpectedly complex ways. He describes simple control mechanisms that generate behaviors that, if we did not already know the principles behind the vehicles’ operation, we might call aggression, love, foresight and even optimism. Braitenberg gives this as evidence for the “law of uphill analysis and downhill invention,” meaning that it is much more difficult to try to guess internal structure just from the observation of behavior than it is to create the structure that gives the behavior.