Could a neuroscientist understand a microprocessor?

The Visual6502 team reverse-engineered one of the chips used in the early Atari video game system:

…we exposed the silicon die, photographed its surface at high resolution and also photographed its substrate.  Using these two highly detailed aligned photographs, we created vector polygon models of each of the chip’s physical components – about 20,000 of them in total for the 6502.  These components form circuits in a few simple ways according to how they contact each other, so by intersecting our polygons, we were able to create a complete digital model and transistor-level simulation of the chip.

This model is very accurate and can run classic 6502 programs, including Atari games.

By the way, this is the same idea that Robin Hanson argues will be used to create Ems of human brains.

Eric Jonas and Konrad Kording then applied the same types of techniques which neuroscientists use to try to understand the human brain to the simulation–including lesion studies, analysis of spike trains, and correlation studies. Could the tools of neuroscience be used to understand the much simpler Atari brain? The answer is mostly no. The authors, for example, looked at three “behaviors”, Donkey Kong, Space Invaders and Pitfall (!) and they are able to find transistors which uniquely crash one of the games but not the others.

We might thus conclude they are uniquely responsible for the game – perhaps there is a Donkey Kong transistor or a Space Invaders transistor.

Of course, this conclusion would be very misleading but what are we then to make of similar brain lesion studies? The authors conclude:

…we take a simulated classical microprocessor as a model organism, and use our ability to perform arbitrary experiments on it to see if popular data analysis methods from neuroscience can elucidate the way it processes information. We show that the approaches reveal interesting structure in the data but do not meaningfully describe the hierarchy of information processing in the processor. This suggests that current approaches in neuroscience may fall short of producing meaningful models of the brain.

I was surprised to read this:

Granger causality [37] has emerged as a method of assessing putative causal relationships between brain regions based on LFP data.


Comments for this post are closed