Even if this was true (we don't have any evidence or a considerable sample of games to make any definite conclusions), the problem is that comp vs comp is not the same as comp vs human. A 2700 human maybe be as strong as a 2700 comp, but when it comes to Houdini 3300+ rated engines, a 2900 rated human maybe much weaker than the rating difference of 400 points may suggest; we simply don't know how strong Houdini really is compared to humans. In practical terms, I suspect it maybe close to 3500 rather than 3300.

I completely overlooked this possibility

That was not the case actually, the Rybka version they used was much weaker; the article says the following:

"We had tried to give FM John Meyer (FIDE 2284) knight odds in a four game match a while back with an intermediate Rybka version, but he was too strong for this and Rybka lost 0-4. So we tried a rematch at pawn and three moves, and this time Rybka won by 3-1"

It has been suggested that this format be retained but that the rapid tiebreaks, whatever the number, be played <first>. The idea behind this suggestion is that both players would know the tie break situation going into the classical games and it would motivate the player at the tie break disadvantage to play more aggressively from the beginning rather than the typical cautious play for the first few games.

Not only would that guarantee more games (always a good thing for the spectators) but it would be like a warm up for the classical time control portion and allow the first/early round nervousness (which Carlsen admitted having) to go away before the <serious> playing begins.

One problem I have with all schemes that try to determine accuracy by comparing moves against some engine reference is that, in order to save time and obtain a sufficiently large sample to yield statistically meaningful results, the reference engine's search depth is limited to what I consider an unacceptably low number of plies. In papers by Guid, Bratko, Regan, and others. Initial search depths used were as low as 12 for Rybka, and increased to 20 for Houdini, but based on personal experience I don't consider these search limits for either engine acceptable. The authors make a considerable effort based on non-changing move rankings as search depth increases to justify the accuracy of the results obtained at such low search depths but, frankly, I don't buy their arguments. I have often seen these two engines (as well as others) change their evaluation <and ranking> of moves as higher search plies are achieved. So I don't consider any results these authors have generated to be reliable.

And unfortunately Erik Varend did not indicate to what depth Rybka was allowed to search in order to establish the reference "best move" and its evaluation. So I have to withhold my acceptance of the results he presents until I find out the Rybka search depth used to establish the reference best move and evaluation.

I disagree. I am just a patzer and I repeatedly find endgame errors made by Houdini, Critter, etc. in endgame analyses run to considerable search depths, near 30-ply. I have confirmed these errors by sliding forward, using either the same engine or other engines. And, FWIW, I have found Rybka to be the best endgame engine, and I hypothesize that it's because it apparently prunes its search tree the least aggressive of all the major engines. Of course, none of this applies once the engines reach tablebase territory.

As far as the engines being substantially stronger than the top players, see my response to <nimh> above for a <possible> difference of opinion, subject to validation the CCRL/FIDE comparison tables and updating them to reflect current CCRL and FIDE ratings. But don't worry, even I don't believe my own arguments.

I don't think that it is any longer and the only reason I think that it keeps being brought up is for historical reasons. Chess at one point was the standard for artificial intelligence, with the thought that if a computer could play chess <using human principles> as well as a top human player, then artificial intelligence would have reached parity with human intelligence.

http://en.wikipedia.org/wiki/Human%...

Notice, this is mostly up to 2004-2005 when computer engine were some 400 points weaker than today.

Drawing somebody lot stronger than yourself on the level above 2700 is not so difficult as beating him. I believe that Carlsen would be able to survive quite a few games against say Houdini but that does not mean he would play on even terms.

Rybka engine that played last relevant human versus machine match against Milov lost close match to Zappa Mexico (another engine). I played Zappa in preparation on ICC, as a "cyborg" drawing. I used to beat him prior to that. The way how he got stronger over short time period was amazing to me.

The pressure on the file plus the tempo saved in getting the rook into play ended up being a significant consideration.

First of all i think that the color scheme WBWBWB - BWBWBW ( as was from Carlsen's perspective ) is not fair.

Since we know white has an advantage.
I would prefer it more like the tie breaks in a tennis match

W-BB-WW-BB-WW-BB-W

For competitors whose skills are close, and for anyone with white, lets say winning chance is > 50% this should prove to be fairer.

For competitors whose skills are close, and for anyone with white, lets say winning chance is > 50% this should prove to be fairer.

What do others think?>

If they didn't have this sequence, the same player would open with white after rest day.

wb wb wb wb wb wb

By reserving at midpoint, they give each player the same number of whites after rest day.

wb wb wb bw bw bw

Here's your suggestion - with rest days taken into account - you can do the comparison.