An interesting tidbit is that while the difference in rating between Carlsen and Anand is 95 points, the difference between Carlsen and Houdini 3 is 376 points.
More big problems in comparing these ratings: intransitivity of chess strength, variation of strength for different time controls, volatility/time variation of human strength, and the much lower statistical significance of human rating lists. There's probably more that don't come to mind.
But anyways, yeah, it's pretty indisputable that computers absolutely massacre humans at chess these days.
Open book essentially represents a very extensive analysis of one position (the starting position) that has been carried out by humans over centuries. We aren't as fast as computers, but we've been working on this one problem longer and with massive parallelism, and so we've got the main lines covered quite deep.
It used to be that most of book went deeper than a bookless computer could analyze in the time available for the opening in a timed game. This meant that a booked up human could steer the game into lines that appeared favorable for the computer when analyzed to the depth the computer could analyze, but were actually favorable to the human when analyzed to book depth.
Computers are now fast enough that they can reach or surpass book depth in most lines, making it much much harder for a booked up human to steer the game into a favorable line.
As an example of how fast computers are now, I just let Stockfish 4 64bit have a go at the start position on my 2008 Mac Pro (dual 2.8 GHz quad core Xeon processors, without hyper threading). I gave it 3 minutes (3 minutes per move is common in world class chess events).
These were the settings: 8 threads, idle threads do not sleep, 8 GB hash table.
It reached a depth of 30 ply (44 ply in selective search), and examined 940 million nodes, averaging around 5.3 million nodes per second. (It picked d4 as its move, and expects a reply of d5).
The endgame tables also aren't as important as you might think at first. They only cover endgames without pawns, and so it is not that often that positions from these tables appear in a game. If one of these positions DOES show up in a game, and the computer misplays it without access to the tables, that isn't likely to give the human the win--it most likely will just let the human draw in a lost position. That's because the problem computers have with these endings without using tables is that they aren't able to see how to make progress from a won position so they end up hitting the 50 move draw rule. (Note that this only applies to games with a lot of pieces left. Even fairly old chess programs could do the basic piece-only endgames without needing tables).
Thus, if the tables were taken away, the overall result would be that in the games where pawns remain (most games) computers would win a lot more than humans, and in games where all the pawns go away while there are still several pieces left humans would get some draws in lost positions.
Overall, take away both opening book and endgame tables, and computers would still dominate, just not by quite as much because of the rare pieces only endgame that humans would sometimes draw from a lost position, and the rare case where the human could steer the opening into one of the few lines where book is still deeper than the computer can reach and that line's evaluation changes between the depth the computer reaches and the depth the book reaches.
What's the alternative? Harrison Bergeron-style handicapping of computers, to soothe our human egos?
But whilst the fact that computers are now better than humans at chess is interesting and impressive, it doesn't really matter in terms of things like the world championship.
Machines can beat humans at all manner of competitive events - plenty of vehicles, and probably some robots, could go faster than Usain Bolt for example - but that's irrelevant when it comes to discussing the Olympics.
Not really. Ratings inflation means you can't compare the absolute values, but is much less noticeable on ratings differences.
So 2660 + 4.7% = 2785 (125 points) in 1972.
The equivalent today would be something like;
2750 + 4.7% = 2879 (129 points)
The same gap in ability translates to only 4 more points.
Actually even that modest difference should not arise from ratings inflation. That is because the Elo system is based on the idea that a given "gap in ability" is supposed to yield a constant Elo delta. An Elo delta of 100 points is supposed to translate to a 64% score for the stronger player, a delta of 200 points to 76% etc.
1. Is it certain that rating inflation is linear? 2. Is there actually any evidence that the constant Elo delta property actually holds even with inflation? It's well known that Elo is flawed, which is why people have been working on alternate rating systems...
Why do I say ratings inflation may not exist ? It is true that in Fischer's time there was one 2700+ player. Now there are 30 or so. That's the evidence for ratings inflation. But it seems likely that the top players are just better than they used to me. Maybe there really are 30 players today who can play at 1972 Bobby Fischer level.