After thinking about this whole Alpha-Zero business for a couple days I started to ask myself a question:would it be fair if one player would be able to use Alpha for prep and another player not?
This is an important question because a program like Alpha needs much more advanced hardware to run than any other engines on the marked.And with more advanced hardware comes a much higher price.-just for the hardware,and then comes the program itself!
I doubt that most-if not all players could not afford hardware to run Alpha properly.
How are we going to deal with this dilemma in the future???
My personal opinion is that all engine use is wrong in competitive chess(I know-I'm a purist).But since everyone else seems to be fine with it I dont see it as a problem.Mostly because all players are able to use whatever engine they want-they are all affordable.
But with a program like Alpha-Zero???-not so much...
I think this is a discussion we MUST have sooner or later and would like to hear what other chess players have to say about it.
Please leave a comment and stay factual and sivillized!
Cheers.
Google doesn't want to create a machine for chess. The match between AlphaZero and Stockfish was just to demostrate the power of machine learning by itself.
Alpha Zero only needs the expensive hardware for learning, not for playing. Even before the time of engines, players that could afford to pay more or better seconds had some advantage. However, Fischer did most analysis on his own and beat the Soviet players who had more resources at their disposal.
You cannot forbid something you cannot control. So we have to live with engine preparation. In ICCF correspondence chess use of engines are allowed as it is impossible to control.
Faulty premise on several levels.
Massive hardware was needed for DeepMind to create a self learning logarithm. The program can run on commercially available PC's.
The chess program is not available, nor will it ever be available for general use. It is not a "program" in the usual sense. It can only make chess, go or shogi moves based on a logarithm it has self learned. There are no evaluation functions. Most probably AZ will never make another gaming move.
Nobody but a very few know the full extent of Google's hardware, just how big it's super-computers are. Their cost? hmmm..... 20B+ $ (?). Out of my pocketbook anyway.
There is very little to learn from the match regarding new chess theory. SF performance is rated at 2200/2300 making bad, multiple mistakes every game. It's data bases were shut off. It was running on a standard conventional lap top with last years version.
Once all the hoopla settles down, people will see that the staged event only showed off a step forward in AI programming and has little relevance to chess theory. It may well change how developers program chess engines. There is a lot of learning to be had. Current programs remain clueless about all kinds of strategic positions.
That is interesting, can you tell me where I can read more about SFs performance being rated at 2200/2300? I have not seen that assessment.
This is not quite true. Stockfish 8 still played around at its level. This is an estimate of the level of games on the ICCF scale, since 64CPU at 1 minute per move is approximately 1 hour per move at 1CPU, which roughly corresponds to the level of 2150-2250 on the ICCF re-play schedule.
Those. The level of published games roughly corresponds to the level of games between players with a rating of 2300-2400 against 2200 on the ICCF scale, i.e. the published games are not the highest in the chess history.
To do this, there should have been much higher computer power or time control.
It does not matter you would be cheating if you used either or both. Cheaters are sick
Google is not averse to money. If someone offered google a lot of money for exclusive use of AlphaZero, they would bite. No doubt that the GM would have a huge edge if they got to use AlphaZero and no one else could.
Using AlphaZero 1 hour per move can allow you to be at the 2600 ICCF level, but it's really very expensive even among potential competitors, because ICCF top GMs already use clusters of more or less different capacities.
Sooner or later chess can be appear for analysis on many thousands of processors on cloud servers "Only for science", who knows !?
