How do you know when to stop training your AI?

Could be a discussion around ethics or simply knowing where to stop.
Also about unconscious biases designed by atypical categories of Data Scientists (young, white, male, well educated, never poor…)