Outthinking Stockfish: Five Positions That Confuse Engines

Those who say they understand chess understand nothing.

Robert Hubner

Introduction

Nowadays the relevance of chess engines is unquestionable. Whether you are a beginner or a professional chess player, you have surely used chess engines for game analysis or opening preparation. The evaluation of engines is sometimes extremely complex and sometimes you understand nothing of them. Nevertheless they sometimes make mistakes. Today I will show you five positions where engines are or have been wrong.

#1 - Fortress Positions

First of all, let's clear up the meaning of fortresses. Generally speaking, in chess a fortress is an endgame draw technique in which the side behind in material sets up a protective zone which the opponent cannot penetrate. Although humans have been analysing these positions for decades, engines still believe that we can have a winner in these positions.

Certain defensive setups allow the defending side to hold a draw, but the engines can give the attacking side a decisive advantage because they're unable to prove progress within their search depth. Let's take as an example the position with king and bishop against queen. The bishop and knight's fortress is a kind of corner fortress. If necessary, the king can move to one of the squares adjacent to the corner and the bishop can retreat to the corner. This gives the losing side enough tempo moves to avoid being pushed. For example:

#2 - When Material Wins Isn't Enough

In the 2008 World Chess Championship, Viswanathan Anand played a great game in Game 3 against Vladimir Kramnik. Kramnik was under a lot of pressure after Anand sacrificed a piece to attack his king. The position was difficult to calculate, and computers often suggested that it was a draw, even though White had the advantage. This showed that engines are limited in positions that need quick thinking and long-term planning.

As the game went on, Anand kept pressuring Kramnik, who made mistakes because he was defending. Anand's pieces coordinated perfectly to deliver the final blow. This victory demonstrated Anand's mastery of dynamic positions and shifted the match's momentum. Game 3 remains a classic example of human creativity outshining computer evaluations in complex positions.

#3 - Pawn Breakthroughs or Zugzwang

Engines might have trouble with positions that require a lot of calculation or strategic recognition of pawn breaks or zugzwang. For example, can you find a way to win in this position?

Humans can intuitively recognize the breakthrough idea, but engines might have trouble without sufficient depth. Until concrete variations were shown by humans, engines couldn't evaluate this position without sufficient depth. This highlighted another potential issue with engines.

#4 - Material Sacrifices for Long-Term Strategic Compensation

It's often the case that engines don't fully appreciate the value of sacrifices in positions where compensation is based on long-term activity or pawn structures. A great example of this is game 3 of the AlphaZero vs Stockfish match. 

I'm not going to annotate this game as it's full of ideas that are extremely difficult for humans to understand. However, I will highlight one position. However, you can find the complete analysis of the game here. In this game, AlphaZero's sacrifices were not given enough credit by Stockfish in positions where there was a lot of strategic compensation.

#5 - The Horizon Effect

The horizon effect is when engines can't see far enough ahead to judge a critical move. They initially misjudged Kasparov's sacrifice because they couldn't see the full sequence leading to a winning attack.

As a final point, let's look at what is often called the 'Immortal Game' between Garry Kasparov, arguably the best player of all time, and Veselin Topalov, a grandmaster and world-class chess player, in 1999. The game annotations are from none other than the great Garry Kasparov himself.

What a great game! It was another example that showed us where chess engines are weak and where we need to improve them.

Conclusion

Wow, did you really make it to the end? Congrats, if you did, and thanks for reading the post! I hope you enjoyed this blog! I don't know if I was successful in actually making you understand how do engines evaluate some positions, so do let me know your feelings after reading this post in the comments!

This will be the end of this blog, any feedback will be appreciated!  Again, thank you for reading this post, and until next time, I am outta here.