What a great idea. This paper asks this question: Would or would not the same methods neuroscientists use to probe that black box we call the brain be useful for probing that different black box we call the MOS 6502?
I don’t like the way they tried to find the answer. They should have actually asked neuroscientists to document the behavior and nature of the 6502. What they actually did was disingenuous: they directly applied neuroscientific methods to the task. Take the Lesion a single transistor at a time section as an example.
That said.. I bet the Visual 6502 person would LOVE to see the raw data (and the ML model) behind this beautiful figure. UPDATE: apparently, the raw data (at least) may be found in Amazon S3 buckets and this repo contains links or addresses or whatever.
Note that Konrad is a neuroscientist. This is not a computer scientist making fun of neuroscientists, it is a neuroscientist making fun of themselves.
In case you missed, this is an obvious homage to Can a biologist fix a radio?, which I also highly recommend.
Note that we do have a radio-level understanding of some biology, biological clock of fruit fly for example. But it took 30 years. A well-deserved Nobel prize was awarded for the work.
Reminds me of an essay I read a while back, The Empty Brain:
Your brain does not process information, retrieve knowledge or store memories. In short: your brain is not a computer…
Also, check out the website for this paper: http://ericmjonas.github.io/neuroproc/