I learned by myself Neural Network because I thought it would solve my issue. I know nothing about ML or AI in general.

My problem is complex but can be easily translated to: I need two outputs where one is bad, one is useless and one is good.

Example: a Mortal Kombat style game where I need to choose: should I attack or not and if I did attack, have I won or lost. So:

• output_1: Should I attack (bool: yes/no)
• output_2: if I attacked, I inflicted or took damage (bool: inflicted/took)

It’s obvious that in this case if I keep only choosing not to attack that would be useless, so I can’t choose that output as ideal.
So I need to attack but only if my inputs tells me it’s a good idea (example: distance is ok and opponents isn’t attacking).

Is there a way to do this only with Neural Network? Am I missing something?
I asked “that” famous AI engine how to do this and it told me to use Loss Functions in NN, but I can’t find any example of Loss Functions in Neural Network, only in ML.

So if I want to solve this issue do I need to change from NN to any other ML algorithm? Which one, please?