If explainability is a problem (e.g. "We don't know why our algorithm recommends higher-interest rate loan products to African Americans"), then that's a rabbit hole that needs to have funding put towards digging up as well.
agreed and I was flippant to make a point. I did not mean to say there aren't smart people in government. But I would venture to say that people who understand ML/AI are _not_ working in the government. Or put more correctly, the _odds_ of a person working in government being up to date on how LSTM works today is such a low number, that I would be fair to say it is extremely unlikely that anyone on any board would be smarter than Sen Hatch _in this domain_.
> then that's a rabbit hole that needs to have funding put towards digging up as well.
Agreed and is the larger point I was trying to make: we currently, no matter how smart you are, have _any_ way of knowing what your "algorithm" is doing. Or if you do, it's a rule-based system and isn't AI/ML. We should though! And I think pursuing that will give us much higher rate of return than our black-box strategy everyone seems to be using.
Let's say we had this. We had some ML/AI team go up to Congress/DoJ and show them some tensors, some "back propagating LSTM" horseshit. How does that help the people that don't understand how the ad model works?
Hell, I bet there is _no one_ using AI or tensors or any of that shit that can tell you _why_ their model output Foo instead of Bar.