Monday, May 30, 2016

Cool paper: "Could a neuroscientist understand a microprocessor?"

Eric Jonas and Konrad Kording recently put out a paper that (in my opinion) successfully argues a point I've heard a lot of people try and fail to make over the years. As I read it, the argument is that the common tools of cognitive neuroscience are often misused and misinterpreted. And that we have a far worse "understanding" of how the brain works than we often suppose.

The way they make this argument is by more precisely defining understanding (roughly 'the ability to explain why something is broken when it breaks and possibly how to fix it') and then using a known system---the MOS 6502 microprocessor---to compare the understanding gleaned by neuro tools with what is known about the human-made system.

They basically perform a lot of neuro tests on a simulated version of this processor while it boots different old-school video games. In and of itself, this is pretty fun. There are several important insights and perhaps I'll go over all of them but just one example is the use of transistor lesioning. I've copied figure 4 from the paper to give an idea. What they found was that there are a bunch of transistors which, when lesioned (i.e. removed or made nonfunctional), prevent the processor from loading any game. There's a similar number of transistors which don't seem to affect the processor's ability to boot games at all. The interesting thing is the finding that several transistors seem specific to an individual game, and those are highlighted in the image. For instance, there are 98 transistors which, when lesioned, prevent Donkey Kong from booting but don't seem to affect the other two games. Jonas and Kording point out that in neuroscience, results like these are often used to make causal and functional inferences, such as "these transistors cause Donkey Kong to boot" or "these transistors are Donkey Kong transistors".

Obviously, these inferences are not exactly merited by the results in either case but it's curious that this fact is more obvious when the system being analyzed is one that is known (the microprocessor) rather than mysterious (the brain).

Anyway, the entire paper is well worth reading. As is the inspiration from Biology, Lazebnik's Can a biologist fix a radio?

1 comment:

  1. Cool commentary on a cool paper ... which has a prelude:

    Neuroscientists cannot understand a Microprocessor, but, staying in the metaphor, they can “Read what Machines Think”.

    Some papers on my home page (when I’m allowed by the copyright form):
    * 2009, Brain Informatics, Reading what machines “think”
    * 2010, Brain Informatics, Comparing EEG/ERP-like and fMRI-like Techniques for Reading Machine Thoughts
    * 2012, Brain Informatics,
    Parallels between Machine and Brain Decoding
    * 2011, Journal of Computational and Theoretical Nanoscience, Reading what machines “think”: a challenge for nanotechnology (locked),