Can you think of a small piece of machine code that you could inject into arbitrary computers that would produce useful effects in some significant proportion of the computer population? (Don't think so...)
Computers and biological systems are so wildly different. Although perhaps you'd find better analogies comparing brains to machine learning systems. Eg. "Trying to find out how ADs work is like applying some simple numerical transformation on some particular part of a deep neural network and trying to find out what changes are produced".
Oddly enough.. I think ML researchers often do find little "tricks" that improve the performance of neural nets without being able to fully grasp how or why. It's somewhat similar to how we use and understand ADs.