To improve this, we add to the evaluation a factor that takes in account the position of the pieces. The initial evaluation function is quite naive as we only count the material that is found on the board. With alpha-beta, we get a significant boost to the minimax algorithm, as is shown in the following example: The number of positions that are required to evaluate if we want to perform a search with depth of 4 and the “root” position is the one that is shown.įollow this link to try the alpha-beta improved version of the chess AI. The positions we do not need to explore if alpha-beta pruning isused and the tree is visited in the described order. The alpha-beta algorithm also is more efficient if we happen to visit first those paths that lead to good moves. The alpha-beta pruning does not influence the outcome of the minimax algorithm - it only makes it faster. The alpha-beta pruning is based on the situation where we can stop evaluating a part of the search tree if we find a move that leads to a worse situation than a previously discovered move. This helps us evaluate the minimax search tree much deeper, while using the same resources.

Step 4: Alpha-beta pruningĪlpha-beta pruning is an optimization method to the minimax algorithm that allows us to disregard some branches in the search tree. This is something we’ll improve in the following step. The effectiveness of the minimax algorithm is heavily based on the search depth we can achieve. With minimax in place, our algorithm is starting to understand some basic tactics of chess: Minimax with depth level 2. The best move for white is b2-c3, because we can guarantee that we can get to a position where the evaluation is -50 (That is, we try to either minimize or maximize the outcome at each level.) A visualization of the minimax algorithm in an artificial position. In this algorithm, the recursive tree of all possible moves is explored to a given depth, and the position is evaluated at the ending “leaves” of the tree.Īfter that, we return either the smallest or the largest value of the child to the parent node, depending on whether it’s a white or black to move. This is done by using the Minimax algorithm. Next we’re going to create a search tree from which the algorithm can chose the best move. Playable on Step 3: Search tree using Minimax Black plays with the aid of the simple evaluation function. The only tangible improvement is that our algorithm will now capture a piece if it can. With the evaluation function, we’re able to create an algorithm that chooses the move that gives the highest evaluation: The simplest way to achieve this is to count the relative strength of the pieces on the board using the following table: Now let’s try to understand which side is stronger in a certain position. We’ll start by creating a function that just returns a random move from all of the possible moves:Īlthough this algorithm isn’t a very solid chess player, it’s a good starting point, as we can actually play against it: Black plays random moves.

Using these libraries will help us focus only on the most interesting task: creating the algorithm that finds the best move. The starting position is used as input and the output is all the possible moves from that position. A visualization of the move generation function.

Based on this, we can calculate all legal moves for a given board state. The move generation library basically implements all the rules of chess. We’ll use the chess.js library for move generation, and chessboard.js for visualizing the board. Step 1: Move generation and board visualization You can view the final AI algorithm here on GitHub. I’ll demonstrate how each affects the algorithm’s playing style. Let’s explore some basic concepts that will help us create a simple chess AI:Īt each step, we’ll improve our algorithm with one of these time-tested chess-programming techniques. By Lauri Hartikka A step-by-step guide to building a simple chess AI

Chess move calculator