In this article, we describe a new search technique that we call "Strategic Search": Instead of developing a game tree by considering every legal moves in a given position, we consider only the goals that can be achieved from that position. As achieving a goal can take tens to hundreds moves in complex games like Go, this means that the search will run much deeper, than with usual search techniques. Of course, to do this, one has to be able to predict and assess the future position once the goal will be achieved. This is done by applying modeling rules that are at the time being, hand-coded in the system, but should be learned in the future by a machine learning module