Sherwood, Russell Friday, January 19, 2018
In modern CC one area which sorts out the “Adults from the Children” (I was going to say Men from the Boy’s but that is rather sexist!) is how players deal with the dreaded engine evaluation of 0.0
Firstly was does an evaluation of 0.0 actually mean? Generally its not drawn, although some players seem to think that this is the case considering the number of draw offers made at this point. It actually means that the Engine considers the position balanced, using its evaluation criteria and search method.
Now let us consider the three sections of this definition
A balanced position does not mean a drawn one! Simply put balanced means that the positives and negatives of each side balance out. What can be interesting to see is if you make a move the Engine now considers it to be -0.07, even though little has changed in the position.
What is vital is that we need to consider what the Positives and Negatives of the position are (for both sides) and look for moves which move the balance in our favour.
Engines evaluate positions and then link these positions into lines to give an evaluation of a position. A typical human player will analyse in a relatively haphazard manner during a game, the reason for this haphazardness is that we build much of our knowledge by pattern recognition and motif knowledge (This is a very fancy way of saying you “know” an attack is on as you have played this kind of position before and also know it tends to be a Bishop Sac on h7 that unlocks the position). Engines are different, they have fairly fixed criteria that they look for and also use proxy measurements – so, for example, how would you define space in the sense of a cramped position? Our definitions tend to be quite grey and lose but the engine will have a very strict one (probably something like the number of available moves)
This means that often the engine's evaluation is 0 because it cannot see anything or the scoring simply cancels out. For us, to progress, we need to add our positional knowledge and understanding of motifs (Small components of plans – e.g. a Rook Lift) to direct its tactical brilliance. Often simply showing it a move will be enough to steer it in the right direction.
We also need to consider the faults in the evaluation criteria of the engine(s), due to the fairly fixed nature of their evaluation criteria. Let me give a very simple example.
Let’s say we have a very simple engine that only uses a piece points based system for evaluation (1=pawn, 3 = Knight and so on). How would we beat it – traditionally the answer would have been with Gambits and/or Sacrifices later in the game. Why did this tend to work – because there is more to evaluating the position than material, so even though we might be -2 down in the very simple evaluation criteria, we might actually have a massive winning position!
So we might try and add some more knowledge to our engine and this will improve the situation but again there will be ways to exploit both the fixed and faulty nature of the evaluation criteria. This is very much the story of engines until early in the 21st Century. The gaps still exist but are much more subtle. For example, many engines struggle with a Queen vs Minor Pieces situation, unless the position if fairly easy to progress.
So how do we work in this area – simply we need to evaluate the position ourselves. Are those doubled pawns actually bad or are they a positive, in this position? and develop our plans accordingly
Different Engines has different ways that they decide which positions to analyse. In a 0 position, we should look how we can encourage them to look in a manner more conducive to our cause. I have covered the details in the past but the main method is to look at the Engines Parameters, for functions such as LMR, Null Move, Wide Search and similar. These all encourage the Engine to look harder at a position. The Evaluation faults will remain but even the dumb engine may stumble upon a plan!
Putting all this together to progress from 0 way may need to go into negative territory but as long as our own evaluation is clear this should not be a concern as this situation will right itself over the course of the game (hopefully). I know in playing the KID this is a mandatory path!
Updated Friday, January 19, 2018 by Russell Sherwood