Building human like chess engine

How to build human-like chess engine

I’m not a software engineer and I don’t have enough time to learn C++  that is perquisite for me to build my own chess engine. So, I’m asking for help just to see if the idea of Master Alexander Shashin can be implemented to achieve new height of chess understanding. Some of the idea might be rubbish or hard to implement in the actual coding.

Ergo, here I go.

The life of a chess game could be described as the gradual movement of the system towards the 8th rank (meaning the eighth rank for White and the first rank for Black - M.S.). Even if you play randomly, this movement will take place because pawns can’t go backwards. The point of strategic play is the gradual movement of compact chess masses towards the 8th rank. The difference here is “compact.” This movement could be accomplished by moving either the rear guard or the advance guard of your army. In both cases the movement of chess masses affects the level of the system’s safety. I introduced the term “packing density,” which I hope would be understood intuitively. A higher packing density means more safety. Safety or survival of the system is dominant. Our human point of view is that the aim is to destroy the opponent’s pieces, the king in particular, but this is only a 2nd tier of the system: its desire for expansion.

So to find the best move in a given position we need to consider the following four factors:

1) Material (m). A very simple and well-known calculation 9-5-3-3-1. Divide your sum by the opponent’s sum. Usually m=1. It is analogous to the mass of a system.

2) Mobility (p). The number of legal moves in a given position. P1 is the mobility of our pieces, P2 is the mobility of the opponent’s pieces. P1/ P2 = p. Mobility is analogous to the kinetic energy of a system.

3) Expansion factor. It is defined by calculating the center of gravity of a given position. Take the starting position. White has 8 pieces on the 1st rank, and 8 on the 2nd rank: 8*1+8*2=24, divided by the number of pieces, 24/16=1.5, so the center of gravity lies “between” the first and second ranks. In every position there are two expansion factors, White’s and Black’s. An important number is the Greek Delta, which denotes a change or variation of a quantity, and the difference between them. The expansion factor is analogous to the potential energy of a system.

4) Packing density. There are several different types of packing density: by all pieces and pawns; by king and pawns; by king, pawns and knights (short range pieces); by pawns only; and local density at certain sections of the board. The most informative densities are those by short-range pieces. Imagine a position: Kg2, Nf3, and pawns on f2-g3-h2. The area of the smallest rectangle that contains these pieces is 6. There are 5 pieces, 5/6=0.83, very dense and therefore a safe construction! Notice that I only need four factors to describe any position. Two of them, the expansion factor in the form of a center of gravity; and packing density, are original, or, at least, I never seen them suggested by anyone else. In most positions it is enough to know m and p. If m=1, p is the main factor we need to identify, which algorithm of the three known should be applied. If p>1.25, it’s “Tal,” p<0.8, it’s “Petrosian” (note that 1.25*0.8=1), and if p belongs to (0.8, 1.25), it’s “Capablanca.”

“Tal” means that 1) first we look for an open attack, 2) optimal piece deployment, 3) sacrifice, 4) material gains. In an open attack; we attack material targets according to the hierarchy: king, queen, rook, bishop, knight, pawn, or an empty square. “Capablanca,” when we have no or little mobility advantage, suggests a different approach: 1) optimal piece deployment, 2) pawn advances, 3) exchanges. After the exchange, if the position still belongs to “Capablanca,” return to 1. The process is cyclic. “Petrosian” is defensive play, the “anti-Tal.” If your position is bad, give up territory, like the Russian army during the Napoleonic War in 1812. Giving up your territory increases your packing density and makes you more resilient. Then try to reduce the deficit in mobility and change the algorithm to “Capablanca.” There are, naturally, many subtleties in move-searching, but these are my general rules.

If m=1, p equal to 1.25 and 0.8 suggests that we are at a bifurcation point. How should we decide between whether it is “Tal” or “Capablanca?” If factors 3 and 4 don’t help either, then make a random decision. Another chaotic zone appears inside the “Capablanca” algorithm and is linked with the problem of exchanging pieces. In some positions the system is indifferent to exchanges; whether you swap pieces, or not, does not change the system status. So you are free to make a random decision.

All credit goes to Master Alexander Shashin for his magnificient Move-Search algorithm…

fantastic... thnx chessotic for sharing this. Let me give a try

The problem with getting a computer to play like a person isn't the move search algorithm, it's the problem of getting a computer to mis-apply criteria in a human-like way. 

I've been toying with the idea of creating a chess engine that is built using genetic algorithms or evolutionary programming. Instead of telling it how to play, just give it random instructions, and have it play tournaments against copies of itself that have small random mutations. the top few spots in the tournament are allowed to "mate" to produce the next generation, and the process continues. Let it play a few thousand tournaments, and eventually it will learn to play well on its own.

Every attempt to dumb down the basic depth-first-search, iterative deepening, alpha-beta pruning regime has resulted in programs that make stupid mistakes, totally unlike what a reasonable human player would make.  Brute force, augmented with opening books and endgame tables, beats grandmasters; algorithms based on "chess principles" (control the center, mobility, etc) are whimps.

There are no such thing as algorithm based on 'chess principles'...All chess engines coding are divided into two parts which are interconnected,move searching and evaluation criteria rubric...And would you call Tal chess(Ancestor of future Hiarcs whimp at 2100?)...

What i've trying to do is to build a sintetics chess master instead of an overlarge calculator which the general public has mistaken as AI...

A few of my friend is doing a research on genetic algorithm..interesting stuff...good luck on that anyway..

The all-important hyphen...

Believe it or not, that was my first thought too! First we must learn to build a human -- then we can worry about building one that is like a chess engine!

Surely all the prograamers out there must be able to create a simple program that outputs these five parametes every turn??? ShashChess on GITHUB is NOT IT. It is just modified Stickfish...

Why not train an AI with access to the best available engine in order to evaluate the quality of potential moves, augmented with a layer trained on the suboptimal moves played by actual humans referenced from the databases of matches?

Seems like as easy problem to solve.