The idea that anything that can be done with a narrow AI that you can quantify up a loss function for a reward mechanism is not true, since the brain is much more complex than a convolutional neural network or transformer.