This is backwards: Theory is machine-learning free, because we havent begun to systematically analyze the machine learning boxes to figure out where the theory arises out of it. If our brain can do it, then we can do it with the machine learning systems, but the relevant field is very underdeveloped. We still call ANN systems "black boxes" but eventually we 'll have to open them and start naming the parts inside. Much like how thermodynamics arise from statistical mechanics, theory will arise from connectionist dynamics
This is backwards: Theory is machine-learning free, because we havent begun to systematically analyze the machine learning boxes to figure out where the theory arises out of it. If our brain can do it, then we can do it with the machine learning systems, but the relevant field is very underdeveloped. We still call ANN systems "black boxes" but eventually we 'll have to open them and start naming the parts inside. Much like how thermodynamics arise from statistical mechanics, theory will arise from connectionist dynamics