Search is extrapolation. Learning is interpolation. Search+Learn is the formula used by AZ. Don't forget AZ taught us humans a thing or two about a game we had 2000 years head start in, and starting from scratch not from human supervision.
no, search is not extrapolation. Extrapolation means taking some data and projecting out beyond the limits of that data. For example, if my bank account had $10 today and $20 tomorrow, then I can extrapolate and say it might have $30 the day after tomorrow. Interpolation means taking some data and inferring the gaps of that data. For example, if I had $10 today and $30 the day after tomorrow, I can interpolate and say I probably had $20 tomorrow.
Search is different from either of those things, it's when you have a target and a collection of other things, and are trying to find the target in that collection.
Search can go from a random init model to beating humans at Go. That is not interpolation.
- Search allows exploration of the game tree, potentially finding novel strategies.
- Learning compresses the insights gained from search into a more efficient policy.
- This compressed policy then guides future searches more effectively.
Evolution is also a form of search, and it is open-ended. AlphaProof solved IMO problems, those are chosen to be out of distribution, simple imitation can't solve them. Scientists do (re)search, they find novel insights nobody else discovered before. What I want to say is that search is on a whole different level than what neural nets do, they can only interpolate their training data, search pushes outside of the known data distribution.
It's actually a combo of search+learning that is necessary, learning is just the little brother of search, it compresses novel insights into the model. You can think of training a neural net also as search - the best parameters that would fit the training set.
