Of course you need some objective function and feedback about how well you're doing wrt that. In RL this is true too. The feedback needed for backpropagation might be, for each action you take in life, what your chance of death is right after that action