SoylentNews is people
SoylentNews
Chess: Magnus Carlsen to meet top teens in pursuit of record unbeaten run:
Magnus Carlsen, the world champion, will face five opponents aged 21 or younger when he bids for a record eighth victory at Tata Steel Wijk aan Zee in January. The traditional Dutch tournament at the small windswept North Sea town is by general consensus the best all-play-all annual event on the chess calendar and the invited entry for its 2020 version is bold and imaginative.
Fabiano Caruana, the world No 2, and Wesley So, who crushed Carlsen 13.5-2.5 at Fischer Random earlier this month, are in the field, but many dedicated fans will be watching Alireza Firouzja, who at 16 is receiving rave reviews comparing the Iranian teen to the legendary attacking genius Mikhail Tal. Jeffery Xiong, 19, the No 1 US junior, also has a breakthrough opportunity.
For Carlsen, the fifth round at Wijk, which will be played on tour at Eindhoven, could be the day when he sets a world record for the longest run of classical games without defeat. The Norwegian thought he had reached his target when he surpassed Ding Liren's 100 mark but then there was publicity for the Russian-Dutch GM Sergey Tiviakov's 110 against weaker opposition. So Carlsen still needs to stay unbeaten in four games at the Grand Tour final at London Olympia on 2-8 December and his first five rounds at Wijk.
(Score: 2) by JoeMerchant on Wednesday November 20 2019, @11:45AM (2 children)
Computers have proven now: strategy and tricks are useless, any mistake will result in defeat when your opponent has unlimited time/resources/processing power per move.
However, (disclaimer: I'm unfamiliar with modern top level tournament play) I'd really like to see a stratified ranking system with more emphasis on time per move. In this respect, humans could (fairly, in my opinion) be granted a handicap against computers, but for human vs human play: a 30 minute per move championship, a 10 minute per move championship, 3 minutes per move, 1 minute per move, 20 seconds per move, and even 5 seconds per move. I prefer the Fischer time rules, with something like a 10x maximum "bank" of saved time. As for the computers, I think it would be interesting to convert the human brain caloric consumption (during play) to watt-hours and grant the computers instead of equal time, roughly equivalent energy per move to the human players.
🌻🌻 [google.com]
(Score: 2) by All Your Lawn Are Belong To Us on Wednesday November 20 2019, @03:29PM (1 child)
Any mistake could mean a loss, true, and perfect play by both sides should (in theory) always lead to a draw. But remember that the computers are still applying algorithms written by humans to assess what the state of a board is and then use their speed to assess multiple boards. (It's not uncommon for a current computer to run between 10 and 40 million possible board variations per non-book move, several million just for a simple move). Modern Grandmasters try to take advantage of this by finding board configurations that a computer might calculate as suboptimal but have some advantage the computer hasn't been programmed to recognize.
There are both long-term and blitz championships. The FIDE World Blitz Championship is played at something like 3 minutes plus 2 seconds for each move (although I don't follow blitz much). What is referred to as "the" World Championship has been pretty similar for a very long time. The regulation for the 2020 match is: Each side gets two hours to complete the first forty moves, sixty minutes is then added after completion to complete the next twenty moves, then after move 60 another fifteen minutes is added per side with an addition 30 seconds added per move. The idea being that while one can't take all day, nevertheless a player should have considerable time to be able to think.
If the main event ends up in a draw (7-7 over 14 games) it moves to tie breaking games. Four games played at 25 minutes plus 10 seconds increment with first white by draw. If that ends up 2-2 then they play two games at 5 minutes plus 3 seconds increment and alternating colors each game. This 2-game match is repeated five times if necessary (five separate two game matches) and if all five matches are drawn 1-1 they then play a sudden death game (aka an "armageddon" game). The sudden death game colors are drawn by lot, white receives 5 minutes and black receives 4 minutes, and 2 seconds per move is added after move 60. If this game is drawn then black wins (i.e. white is forced to play for the win to get the title but gets an extra minute of thinking time to try to break black). Most major level tournaments, unable to extend matches indefinitely anymore, have some similar type of play of rapid game tiebreaks followed by an armageddon game. Some tournaments allow the players to 'bid' down black's time in order to receive the advantage of getting black (playing defense).
Incrementing time per move (Fischer time) is generally the standard for FIDE and international play. In the US delay, where one is given a countdown (typically 5 seconds for games over 30 minutes per side) time starts coming off but nothing is added to the base time, is still very popular (I think more popular than Increment, so long as the tournament is not being rated internationally). There are no top limits to the increment, however. If one considers that the longest recorded game was 269 moves under rules that no longer apply, and 240 is the next highest, and games over 150 moves are exceedingly uncommon, one might reach a theoretical maximum. The 269 mover took a little over 20 hours to play, but a 150-mover calculated by the rules above cannot go over 4 hours maximum per side. In practice the longest games still typically take 6-7 hours.
The biggest challenge is getting the players to actually go for a win. In Grandmaster play the main incentive is "don't lose" rather than "try to win," and as such there are a lot of draws in play. It's hard to get the player to try and risk loss to get the win, unless the player is certain (relatively) that a certain move will lead to a win - typically because the variation has already been analyzed by the player beforehand and then reconfirmed at the board before committing. To try and combat that, some events follow "Sofia rules," named after the location where they were first seriously employed. Normally (like in the championship) a win counts for 1 point, a draw 1/2 point for each player, and a loss 0. Sofia rules count a win as 3 points and a draw as 1, thus giving a greater reward in standing to players who gain wins. But this variation has never been terribly popular.
Human brain caloric consumption would be interesting to try and figure. Anecdotally, Garry Kasparov said he could lose something like 10-15 pounds during a top-level tournament and IIRC he might take a little walk or something but no strenous exercise. One might do that by comparing caloric consumption and weight loss before, during, and after tournaments to establish player baselines and then average them. But I fear what would be found is that the computer is a much more effective machine from the start - yes the power supply must run the whole computer but it doesn't have to run a body on top of it. :)
This sig for rent.
(Score: 2) by JoeMerchant on Thursday November 21 2019, @03:13AM
The algorithm that started beating world champion Go players is Monte Carlo Tree Search [wikipedia.org], and in its most recent forms, this algorithm is not only beating top human Go players, but also Chess and Shogi, with only basic programming of the rules: what is and what is not a legal move, and no other library or strategy programming.
I haven't done a deep dive into AlphaZero or KataGo [github.com] but the 90,000 foot view is: they use random play, I believe sometimes in combination with minimax, to train neural networks. So, much like humans, that training phase is very computationally expensive, but the actual play is relatively efficient and fast.
Thanks, my attention span is dwindling as the years go on, but I don't think I ever had enough patience to spend a full 3 minutes considering a chess move - which is probably why I never achieved more than about 1800 ranking at my peak - at my current attention span capabilities I'm down much closer to 1200. But, when you put the clock around 15/60 seconds Fischer time, I do relatively better since that is well matched to my "interest level" in the game.
Contrast with Go, where 300+ moves are possible, though higher level players can usually forecast the final outcome long before the board fills, and uncertainty diminishes significantly as the board fills, usually leading to faster play.
Philosophically, that's an appropriate final analysis for the "game of Kings," though, I doubt the game's original designer knew this would be the case. Almost like WOPR and Global Thermonuclear War: the only way to win is not to play...
Go implemented something similar not so long ago [wikipedia.org] with Komi: 6.5 or 7.5 points (Japanese or Chinese rules) to the white (second) player to force black to make more aggressive choices, since it was generally agreed that black enjoys a clear advantage over white due to moving first.
I think back in the Deep Blue days, humans would have beat the computer handily even giving the computer the whole body caloric consumption budget. Today, with neural networks implemented on power efficient dedicated architectures, I think the humans are in deep trouble, even just counting the extra energy consumed for play above baseline consumption. Still - ratios could be established and pushing the computers ever lower in total power budgets while they still attempted to maintain a certain level of play would be a neat way to make it interesting to play against computers again. As it is now, a cellphone is a depressingly omniscient opponent.
🌻🌻 [google.com]