I’ve read an article in Chess Life a while back, about a correspondence grandmaster, and how he uses chess programs and databases to play his games. He noted that in his games, he’d find nuances, in which he could find a better line than the computer. I presume the nuances Corr. GM’s find has to be in the late middle game, where its less prone to outright middle game tactics, but too soon to be considered an endgame. I find that’s the part of the game that’s actually trickiest, since your trying to both find winning tactics, while setting up a positions that gives you a long term advantage. So it can be easy to overlook some nicer tactics.
That got me wondering, since computer programs have gotten even stronger over the years, if correspondence grandmaster can still find better lines than a computer.
The article wasn’t really all that long ago, maybe a couple years, but I still wonder if chess engines like Houdini 3 pro, still make enough subtle mistakes that at least a Corr. GM can find weaknesses.
Even if engines didn’t get any stronger than Houdini 3 (or whatever flavor is preferred nowadays), you still have to contend with computers getting ever faster, thus slightly improving its play on newer hardware.
In the ICCF, many top players use computers, which are allowed. Computers still make positional errors, even early in a game, sometimes misjudging something for many moves, so many that impact of the depth of misjudgment is down a side branch beyond the analysis horizon of the computer. Additionally, there are time limits and multiple games. Unless someone had the resources to own at least 6, and perhaps as many as 12 high end computers, it would be logistically impossible to play in high level events and to let a computer sufficiently analyze to find these flaws (given enough time they might eventually as they increase their depth and breadth of analysis.) As a result, a hybrid approach, where a strong player steers the computer to analyze certain lines tends to work better than pure computer analysis.
Chess has a very large number of positions available to it. I don’t know if, with modern technology available, the following statement is still true, but I had read many years ago that the number of possible positions in the first 20 moves of a game exceeded the number of atoms in the observable universe. As a result, computers cannot analyze all possibilities. Computers can’t currently “predict” whether a move or idea will be good, they must consider it first and then analyze it “in hindsight.” Thus, for a modern computer to hold all variations/positions in its “mind” and cross evaluate them, even if position/variation storage were at an atomic level (and its not), it would take a computer the size of the universe (or larger) to accomplish that. Assuming the information to evaluate the position traveled through the computer at the speed of light, we’d be talking about billions of years for the evaluation to make it once ACROSS the computer. And that’s not the time for an entire game, which would be considerably longer.
So, while computers continue to improve in strength, for the foreseeable future, a hybrid approach to analysis can be expected. Perhaps quantum computers would eventually change this.