[3a2d29]

To add to this, I wonder how "explaining the algorithm" will work with algorithms that are trained with ML. Essentially they are black boxes, right? So as a tech company, would I have to just say what my best guess is on how it works?

[deadlocked]

As a tech company you know what outcome you want your ML algorithm to produce and you presumably have some way of figuring out whether or not it’s producing that outcome. Presumably you also know what’s being fed to the ML algorithm as training material.

[poisonborz]

I guess you always have a set of weighs that the system tries to work towards to.

[jdrc]

"gradient descent" should be enough