Blog #2 Rewrite: Why Could Human Chess Grand Masters Could Win Against AI

Chess has been in the market for many years now. Chess was known to be founded around 1500 years ago in India, and it is since the invention of digital computers that playing chess has changed its platform. traditionally only human can play against human, but nowadays more and more modern computer chess AI comes into play. Although computers has so much computational power that it can foresee hundreds of thousands of moves each time, human chess grand master can still wins. How is this possible?

To answer that question, let’s first discuss how computer chess AI worked. AI is a form of automated instructions that does not take any external human input parameters to perform tasks. In this automated process, computer programmer needs to consider three board questions: 1. Board representation, 2. search techniques, 3. leaf evaluation. The first question solves the chess game into data structure and the second question determines the possible potential moves that waiting to be further examined, lastly the third question which solves the possible best move at each position.

Search Techniques: (I used to think that computer considers every possible combination at each given move up to a certain fixed steps, but in reality, it is very computationally costly. Ex: for just 3 moves ahead in a typical 30 moves possible position, it would take computers to approximate 10^9 positions, and potentially takes about 40 years for an extremely fast computer to run.) In fact there are two types of search strategies involved, Type A which is what I thought before and Type B which uses a “quiescence search” and filter out to only look at good moves for each positions. To improve on this filtering, computer used pattern recognition to filter out some of the potential moves, to further saving computational time. Just like human players, computer using known grand master games as a way to learn about certain patterns. If some moves are rarely played, computer would classify it as a bad move. After solving search techniques question, computer now yield some potential good moves to be considered.

Leaf Evaluation: As the search techniques leaves several good moves to be considered, now instead of evaluation all the possible positions, computer only choose to evaluate several good moves using evaluation function, and compare the final position to draw a conclusion on what move is better than others. For example, an end position of checkmate is better than any other moves; a better pawn structure should also have higher evaluation score. In modern chess algorithms, this step varies the strength of the chess AI.

After analyzing ho chess algorithm work, comparing with human grand masters, AI can have more advantages in terms of searching techniques. Since AI was trained under known matches that are way more than hat human player could experience or memorize throughout their life time, the pattern recognition of AI would be a lot stronger than human players. However, that statement is only true in early or mid-game. In late game chess, pattern recognition become less important as number of pieces down numbered, there will be less information for chess algorithm to analyze and to think of the potential patterns for a good move, whereas human players can firmly relay on their own judgement rather than existing patterns. As a result, although computers can in parallel analyzing many more moves, occasionally human grand master can still win against AI.

Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s