Certainly not. Hence the qualifier.
The short answer is that the best programs will beat the best GMs in match play all the time now. When I posted the link to the 2002 Kramnik-Fritz match above (tied at 4-4), I didn’t realize Kramnik had actually lost a 2006 rematch 4-2.
This is relatively new…in the late 90’s/early 00’s results were mixed. Even some lower-level GMs could adopt a very stodgy, slow technique that confused computers; in particular a middling Dutch GM named John van der Wiel perfected that technique and for a while was called the ‘computer-killer’.
But the machines won out; what kills GMs now are the refined scoring techniques, the deeper plies (counts of moves forward) due to jumps in processing speed, and the inclusion of endgame databases.
Did the endgame databases really make a significant difference? It seems to me that, in any scenario that you could fit into a database, the computer’s standard tactic of “brute force look ahead a bunch of ply” would already work pretty darned well.
It the database makes it work smarter, not harder. Once your calculation gets a game state matching one in the database, you don’t have to think about it any more - you’ve already determined what will happen if you get there.
I’d like to believe that a large part of the increase in computer playing strength is a better evaluation function, not a deeper search, but the existence of cheap memory and computing has allowed for massive end-game tables and insane numbers of calculations looking for them.
For the game of Go there are simply waaaaay too many possible combinations for a computer to brute force easily. Depending on how you look at it the number of possible combinations of a Go game can reach 10[sup]360[/sup] (or 10[sup]700[/sup] or more depending how you look at it). That is way, way, way past the total number of atoms in the Universe. Big ass number.
In comparison chess has a far smaller number crunching task for a computer with 2*10[sup]46[/sup] possible combinations. Still ginormous but nowhere near the Go problem.
Of course computers don’t assess the whole possible game at once but rather go move to move looking deeper and deeper. One can see the task for the Go game goes off the rails pretty fast. In chess the problem gets hefty but not near as fast so is doable. Looking 4-5 moves ahead is achievable and makes for a suitably challenging opponent for most people.
Ah, wait, I see it now. If you’re already in one of the states listed in your endgame database, then you could probably take it from there by brute force, but if you’re five moves away from a database position, then you could look ahead five moves to the database position, and jump from there to the end of the game, whereas before you would still fall short.
I wonder how long even go will survive. the best go programs are rated about 1 kyu - I am a reasonable club player at 5 kyu so they will beat me (in practice I find positionally i can still just outplay them, but darn it they always beat me tactically as they can home in on an isolated groups weakness - the fact that there are so many possibilities in go also works against humans too) I reckon in another ten years we will have a computer world go champion -
I use to “cheat” in chess by setting traps. I wasn’t good enough to beat the better players so I would play what looked like an obviously bad game with a 2nd strategy in the background. It worked until they figured it out and then I got trounced.
In this case the programmers could see such a gambit and adjust the program around it, which means it took a human to see it.