> A major advantage of employing machine intelligence early in the recruitment process is zero bias; candidates are judged purely on their merits. I am remind

> A major advantage of employing machine intelligence early in the recruitment process is zero bias; candidates are judged purely on their merits. I am reminded* of the AI Koan [1]: In the days when Sussman was a novice, Minsky once came to him as he sat hacking at the PDP-6. “What are you doing?”, asked Minsky. “I am training a randomly wired neural net to play Tic-Tac-Toe” Sussman replied. “Why is the net wired randomly?”, asked Minsky. “I do not want it to have any preconceptions of how to play”, Sussman said. Minsky then shut his eyes. “Why do you close your eyes?”, Sussman asked his teacher. “So that the room will be empty.” At that moment, Sussman was enlightened. These kinds of systems terrify me; there was the article yesterday about the mythos of model interpretability, and how ill-defined the term is. I am equally concerned about its complement, model auditability; how well can the system explain its decisions, or have them explained. How opaque is the reasoning of the system behind a given decisions. I agree that there's a fair argument to be made that useful models are often sufficiently complex such as to not be understandable in their whole, but I'm not convinced that that's sufficient reasoning to abandon the cause entirely; it just provides an upper limit on what can be accomplished. I know this is, functionally, just an ad, but there are an awful lot of ads lately with the same breathless enthusiasm for these kinds of poorly-understood systems. --- * Yes, I know this is tantamount to posting "relevant XKCD" or whatever; the default call/response. [1] http://catb.org/jargon/html/koans.html http://catb.org/jargon/html/koans.html