what is the % of chances of grandmasrer winning chess game against computer?

This would have to be a very limited excercise: only in the elite group at very prestigious tournaments ( like the current one at Wijk aan Zee ) and only one engine game per tournament per player. A marginal bit of inflation at the highest level ( which inflation is happening in a greater extent anyway ) would not be that a high price to measure true engine strength.

By the way I proposed a solution to the inflation problem in this thread:

http://www.chess.com/forum/view/general/possibility-of-a-normalized-rating-system

According to WGM Natalia Pogonina, in a match between any GM and Houdini , a human being is expected to draw 1 out of 10 games and lose the rest. It's not exactly what she said but I couldn't find the thread.

Playing against chess engines reminds me of the movie "Real Steel". For the general public, fighting machines is science fiction, for chess players/aficionados, it's a reality. The times when a human could beat an engine in a match are over, in my opinion, in the same way a human cannot fight a robot in the movie, and the gap is just getting bigger and bigger.

I'd like to see Carlsen-Houdini in a 10 or 12 match game. Kasparov did it in his day, after all. The computer would almost definitely win, they are just too good these days, but it'd still be a good spectacle.

I think the very strongest humans would still have a chance in a match, if certain conditions were used.  Eg the human gets a sample of 1000 games against strong opposition - computer and human - and across a range of openings, played by the exact engine he will face, 3 months in advance, employs a team to help prepare, and no modifications whatsoever are allowed to the engine either in those 3 months prior to the match or once it starts.  This means the engine has to have self learning or intelligent randomisation imlemented so that if the GM finds a draw/winning line, he cannot safely play it again every game with the exact same result.

I would be very interested in such a match. However with all the preparation and effort that would go into such a match it still would be a sample of one which is statistically speaking not significant. The result would be debated ( not without any basis ) and we would not get any closer to establishing the real strength of the engine.

Only relatively large number of games played under tournament conditions on a regular basis could once and for all decide the debate whether engines have really reached super human strength of 3300-3400 or they are just a little bit better than the best humans ( if they are at all better ).

Sample of one again.

Just a quick calculation:

At the heart of the both the Elo and the Glicko system there is this formula:

This is the expected result on a scale of 0 to 1 of player A against player B if their respective ratings are RA and RB. In the case of the Glicko system this is corrected for the rating deviation but the basic structure of the formula remains the same. The choice of the constant 400 is common in both systems ( this is an arbitrary value and Elo chose this for a mysterious reason ).

You can plot this formula if you assume x = RB - RA using Wolfram alpha:

It is visible from the diagram that for |x| > 500 the expected value of the result are close to a complete victory for one side.

In fact you can solve the equation E ( x ) = 0.05 for x ( which would mean one draw out of 10 games and all the other games lost ):

This means that if the rating difference between two players is greater than 511 points than chances are that the weaker player would not score even half a point in a 10 game match.

Since claims of 3300-3400 engine strength would mean just this big a difference between the engine and Magnus Carlsen it would mean that the engine would beat him 10 - 0 which I can't possibly imagine.

Engines are weaker at that kind of time control in closed games, not stronger.  A highly (even completely) closed and strategic position can become one where a good human player can almost instantly play a move that's "good enough."

While at the same time, the very nature of computers is such that their huge advantage gets neutralized, as they have to start computing the near-infinite number of possibilities from scratch at each opportunity.

What makes such positions winnable for a Nakamura, is the fact that the short time control forces a situation where the computer is quickly forced to make sub-optimal moves, as it doesn't have time to process the variation trees properly and at length.

In other words, it's little more than a parlor trick.  Forced to play out that same position against Houdini in a situation where Houdini got several minutes -- or God help him, several days -- to compute each position would be a hopeless task.

There are also situations where even with many minutes in a very powerful machine, the engines give you an edge of around 1.00,  but they are only shuffling pieces from one place to the other.

Sometimes the time doesn`t really matter for the engine, they don´t change it´s opinion or change it for another useless move (I`m talking about middlegame positions, not about endgames in engines without tablebases).

A computer would never lose a correspondence game. Ever. Carlsen might be able to draw for sure, but he will NEVER beat the computer. The computer always plays perfect, always. If carlsen manages a perfect game its a draw. No human right now can beat a computer. I also have no idea where you possibly got that carlsen plays BETTER than the computer because this is not true, he can play on par but never better, because the computer is always perfect. However in a shorter game like blitz or bullet or rapid the computer will win 9/10 and draw 1/10 because all humans make mistakes. Carlsen and Hikaru make mistakes, the computer never does

slim to none. and slim just left the building.