3 ms·
Ask HN: Have Chess Programs Become Even Better These Days?
There was a lot of media coverage when Big Blue from IBM had beaten Kasparov, but what is the state of Chess programs these days with Deep Learning and the latest tools of the trade? Have they improved significantly so that no human can every hope to compete, even the best chess champions ? Is there any hard data available recently ?
- jcr 11y agoYour best bet is to learn about the Elo Rating System [1], and then about the strength of human players throughout history [2]. Keep in mind that "Elo Rating" is imperfect, particularly in historical context, but it's still a reasonably solid measurement. Next up, learn about computer chess, past and present. The main computer chess wikipedia article [3] is a good place to start. Finally, start looking at current computer chess competitions [4] and the programs being used. Whether the claimed "Elo Ratings" of computer chess programs are realistic, is another matter. Suffice to say, many of the current programs can beat most human players. When Magnus Carlsen became the current (human) chess champion in 2014 (FIDE Elo:2882), I recall reading how uses computer chess programs for practice. [1] https://en.wikipedia.org/wiki/Elo_rating https://en.wikipedia.org/wiki/Elo_rating [2] https://en.wikipedia.org/wiki/Comparison_of_top_chess_players_throughout_history https://en.wikipedia.org/wiki/Comparison_of_top_chess_player... [3] https://en.wikipedia.org/wiki/Computer_chess https://en.wikipedia.org/wiki/Computer_chess [4] http://www.computerchess.org.uk/ccrl/ http://www.computerchess.org.uk/ccrl/
- Lordarminius 11y agoAvid Chess Player here. Chess programs are way beyond what they were whe Deep Blue defeated Kasparov in 1997. No living Grandmaster could hope to defeat a top-notch chess program today - hence the trend towards computer vs computer matches.
- firebones 11y agoMakes me wonder if the next frontier of chess is to figure out a reliable way to have computers play at each particular level authentically rather than making random mistakes. For instance, establish a particular ELO against humans, then play computer versus computer enough with various deficiencies introduced (worse opening book, worse calculation or scoring) until you establish a desired level. Could Google's deep learning approach they used for Go be turned against chess to beat existing programs? Or is there something about chess that, despite the smaller search space, makes it less amenable to the Google approach?
- aaron695 11y agoYou'll find a phone using open source software can beat any human who's ever existed. [edit] Interestingly I can't find someone actually willing to say it from evidence. Interesting in-denial from the chess world.