How one can model physiological reward mechanisms occuring in the brain using artificial neural networks? E.g. are there efforts to use the notion of dopamine or similar substances in the artificial neural networks. Maybe introduction of the physiological reward mechanism can lead to the emergence of consciousness or at least enhance the effectiveness of reinforcement learning?
Essentially - how neural network models reward? People's brain perceive money as the ultimate reward because almost everything other can be bought by this. So - mental perception of owning money gives reward. But how this notion of reward is modeled in artificial neural networks? How networks know that some money is assigned to the network's account and so, the network should feel happy and rewarded and should strive to repeat successful behavior?
I am reading https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5293493/pdf/elife-21492.pdf and I hope that it will move me in the right direction.
It is quite confusing. The old-school neural networks expect that there are 2 separate phases: training and inference. So, the network receives all the feedback (let it be called reward) in the training phase and network receives nothing in the inference phase. But maybe network should receive some reward during acting-inference phase as well, kind a lifelong learning.
