Computer stock pickers generally don't view balance sheets, P/L statements and other fundamentals data in isolation - they match them up temporally to historical market data and try to model what the effect of future financials releases will be on the price of the relevant security. If markets are historically irrational w.r.t. fundamentals, due to human falliability, or whatever else, our machines seem destined to repeat our mistakes.

Is a mistake still a mistake when it is a profitable action?

If fitting to human irrationality increases generalization performance, then it does not matter if the "machines seem destined to repeat our mistakes", it is still a useful signal. If fitting to human irrationality decreases generalization performance, your algorithm is overfit to noise (and you have bigger fish to fry than human irrationality).

Overfitting to noise is perfectly avoidable, not pre-destined when part of your data is noisy (noisy data is the rule not the exception).

Why? If the robot sees a historical parallel to a popped bubble---the robot might anticipate the popping short-sell now?

Algorithms look at the very same things humans look at. They will try to time the market and speculate as much as now or even more.