Quote from: tee-rex on September 19, 2016, 09:48:29 AM

Quote from: actmyname on September 18, 2016, 02:39:36 PM

Quote from: tee-rex on September 18, 2016, 07:12:34 AM

I'm still pro-human, as it were. Algorithms, even the newest ones, still need hardware to run on. If I remember correctly, every time a reigning world champion had been playing against a machine (for example Kasparov versus Deep Blue), the machine had been either a general-purpose top-of-the-notch supercomputer or a custom built device with massively parallel processing technology crafted specifically for playing chess and nothing else. And even in these cases the outcome was far from predetermined. I understand that today's regular computers are a lot faster and have a lot more memory than they were and had some twenty years ago. But I still think that they are not quite there yet to crush humiliatingly a grandmaster level chess player.

Besides, what is an estimated elo of a chess algorithm? Any chess match is time restricted, and would the algo show the same elo on some decrepit hardware given the time limitations for making a move? In my view, an estimated elo of such an algorithm is more hype than reality.

Check out this video: Hikaru Nakamura vs Stockfish minus b-pawn

Here, a grandmaster was playing against Stockfish (back in 2014, so it must have been an earlier version) where he had a pawn advantage. And then proceeded to lose.

An estimated elo of 3447 doesn't mean that in reality it would be ~3000. No. For Stockfish 7, it's elo is 3339 +-16. Surely, they predetermined older algorithms' elo and then started basing newer ones off that, and it kept going until what we have now. The only way grandmasters are able to compete with the best chess algorithms is if the algorithms have a handicap. This is the reality of it. Do a little bit of research and you'll understand how powerful they truly are. No, we're not anywhere close to "solving" chess, but the analysis of these programs is pretty top-notch

Anyway, you can always employ the old trick of passing on the moves between two chess programs. And if we have the same program playing against itself, it still boils down to how fast the underlying hardware is, right? Apparently, the faster instance of such a program will most certainly smash the slower one. Applying this logic further to programs with lower estimated elo running on faster equipment and winning against programs with higher elo running on slower equipment (as slow as required to guarantee a loss for the latter), we have no other option left but to conclude that an estimated elo of a chess program should necessarily be a function of how fast hardware is in respect to playing chess. In other words, the algorithmic strengths of a chess program can be totally negated by using scrap facilities.

Out of curiosity, could a chess program running on a slower computer even theoretically win a match against a faster instance of itself, provided all other conditions being equal? I think it is still possible to end the match in a tie, for example, due to insufficient material on the side of the faster instance of the program, but what about a victory? Or, is the outcome completely predetermined?

I would expect that the programs were run on systems with the same stats and configuration. I don't personally know the process but I'm sure they list it there or you may find out somewhere.

I think that inquiry is based purely on the difference in processing power. Since the technology has been advanced exponentially, many older computing processors would not be able to compete with newer ones provided there is a time limit - some may not be able to handle the algorithm, even! Maybe they'll rust before a game is finished Wink