My good friend GM Maurice Ashley was on CNN. Click here to watch the interview.
Special thanks to IM Ken Regan for sending me the information.
Chess Daily News from Susan Polgar
GM Ashley on CNN
About Author
I just watched the video. It is fantastic!!
Ashly is talking about his belief in pursuing a passion all the way.
Very inspiring and charismatic man. I do recommend whole heartedly to everybody: Watch this video! It’s a must!!
Maurice Ashley did very well to become a grandmaster, but did he not want to try to become an elite player or was that good enough? He is not a great player- I would reserve that distinction for players in the elite- an average grandmaster is very good but not a great of the game. But that is just my definition- like an all-time-great is in the elite of all time.
If “elite” means “in the top 20”, then only 20 people can be “elite”. The distribution of talent is a famous topic for study, and in other areas such as violin proficiency and math, the same pattern of a sparse number of supreme outliers is often found.
Another topic is whether the objective standard of these outliers increases over time. In athletics the answer has been yes, especially in swimming and running (has the latter slowed?). Is it yes in chess? Well, again I was up past midnight doing programming that may help find the answer…
The other life question is, “elite at what?” Elite at life, each in eir own way, is the goal here. Being “Onesimus”—useful—is also important.
“If “elite” means “in the top 20”, then only 20 people can be “elite”. The distribution of talent is a famous topic for study, and in other areas such as violin proficiency and math, the same pattern of a sparse number of supreme outliers is often found.”
What? To be king of the ants?
I mean really? How many people on Earth get up from bed every morning and think about being #1 in chess. I think 4 or 5 people tops.
King of the ants indeed.
*FFFSSSSSSSS* Sound of fat heads deflating.
Actually I know a man who has won almost every top prize a computer scientist could be eligible for (CS does not have a Nobel), and yet to judge from our conversations in April, he still gets up in the morning and thinks about chess. (He is USCF Class A.)
Where did he make his GM norms?
“Anonymous said…
Where did he make his GM norms?”
Between coffee and toast at breakfast after his disturbing dream of being Chess World Champion while in the nude and defeating the current champion that is dressed in a rabbit suit.
Hey I beat this guy online before 🙂
“Actually I know a man who has won almost every top prize a computer scientist could be eligible for (CS does not have a Nobel), and yet to judge from our conversations in April, he still gets up in the morning and thinks about chess. (He is USCF Class A.)”
Has he won the Gordon Bell Award?
It is the closest thing to a Nobel for CS folks.
Here’s some info about Gordon:
http://research.microsoft.com/~GBell/
The Gordon Bell award is specialized to the subfield of high-performance computing, and is awarded by the ACM as listed here. The award listed first on that page, named for Alan Turing, is one of the ones I was talking about.
Ken,
HPC and CS are needed to solve chess. The most recent chess records were pulled off by HPC.
Turing award is geared more toward AI, Theorhetical CS, Decision Support Systems, and Statistical Analysis.
“Between coffee and toast at breakfast after his disturbing dream of being Chess World Champion while in the nude and defeating the current champion that is dressed in a rabbit suit.”
I knew it! Yet another fake GM!
“NOI” – True, but I’m entitled to go by what the ACM itself says about the Turing Award on the page I linked.
If by “solve chess”, you mean “play Elo 300+ points better than any human”, then I agree HPC + CS has done that. Ditto with 6-piece, now many 7-piece, and I may live to see all 8-piece tablebases.
But if you mean “determine whether any given position is a win for the player to move”, I would wager this impossible even if every grain of sand on Earth could process at Planck-time-flops (scaling based on the example on the first page of CRC Handbook chapters I co-authored here).
For this century, take the position shown here and go to Move 50, White to move. It has just the Kings, a White Bishop facing a Black Knight, and four pawns each on the same side of the board. After 30+ trillion nodes on my quad-core (at 6m nps on 3 cores that’s 4m seconds = 1/5 year of continuous operation) and 1MB of analysis, I still don’t know if White can win! (I suspect a harrowing draw—the entire relevant 6-piece EGTBs graciously sent to me last month have cleared up some loose ends but not more.) Is it feasible for current HPC to “tablebase forward” from such a position, using engine evals at certain promotion positions that can’t be tablebased?
Oops, I meant 5m sec ~= 1/6 year, though if you add day-at-a-time sessions with “Deep” Shredder, Rybka, HIARCS, Junior, and Zappa to my long-term Fritzing, it adds to 1/5 year.
“”NOI” – True, but I’m entitled to go by what the ACM itself says about the Turing Award on the page I linked.”
I am aware of the ACM and IEEE. I am a published member of both as well. Quantum computing will change your little evaluation to determine whether any given position is a win for the player to move. The calculation bits in a Quantum computer do not reside in classical time/space, thus are not bound by Amdahls law, so many algorithmic shortcuts can be made using the breakneck calculating speed of the Quantum computer. It has been theorized that some return bits will actually materialize before the actual evaluation is performed. Einstein and Heisenberg are in their graves having fits. 🙂 You will have to update your book chapters pretty soon. 😉
Cheers
No one important
NOI – I’m aware of quantum computing, and I’m skeptical of its scalability. (Concretely, I think Grover’s algorithm requires linear resources, at least for counting, rather than the square-root-of-n runtime advertised by the standard quantum complexity model—I’m doing research to try to formalize this position.) Even so, see Scott Aaronson’s March 2008 Scientific American article on quantum computing, which gives evidence that NP-hard problems are beyond the reach of QC and classifies nxn chess as evidently even harder. We were part of a brief discussion at last month’s Computational Complexity conference of how that might scale to concrete 8×8 chess, especially when the 50-move rule is set aside, and I stand by my projection.
Kwregan – I’m not pretending to fully understand the discussion, but the compute power you reference seems to be minuscule compared to what is possible today.
With sufficient funding, you can assemble a system much more powerful than a consumer grade quad-core. And certainly you would run it a lot longer than 1/6 of a year.
How about several 64 core systems, with a couple of petabytes of data storage, running for a decade. I know that this wouldn’t scratch the surface of solving chess from a theoretical point of view, but fantasize about the size of the tablebase that you could build!
Yes indeed—and that’s the serious (chess) purpose I have in continuing the discussion. Is anyone trying to set up the kind of “forward tablebasing” algorithm and database I’m musing about? Specifically at promotions that don’t result in immediate captures, I would use the engine eval to depth 25, and work backwards from those. The four things that make it tough are () running millions! of depth-25 evals, () minimaxing 32-bit values (int+float), () identifying and preserving mainlines, and () mixing forwards and backwards-working modes, while the standard tablebase algorithms conveniently only need to work backwards.
A lesser but easier idea is to employ a standard “evolutionary algorithm”/”simulated-annealing”-type heuristic. In this endgame I’ve found several cases where DF10 gives eval over +4.00 to depth 25 in positions I know are drawn, but +5.00 has been safe (though in the Adams-J.Polgar K+P endgame from Corus’08 I found +5.22 that goes bad). The algorithm is simple (say to prove a White win):
[1] Run to depth 25 max.
[2] If you get eval >= 5.00 to White, mark the position as “won” and backtrack.
[3a] If not, then with probability p, play the PV forward 1 ply (or maybe a few ply) and goto [1].
[3b] Or with probability (1-p), jump to a White-to-move node earlier in the current line, play a different move (“mutate”), and goto [1].
[4] Stop if-and-when the root is marked “won”.
The probability p of going forward can be made to decrease according to a “satisfaction factor” with the current line. E.g. if many moves go by with little or no rise in eval, and worse the defender has other nice options rather than forced moves in this line, you’ll “mutate” fairly rapidly. With humans now giving up on over-the-board play, this is the next frontier: can computers duplicate (my) human analysis unassisted?
Mind you, what I need more than terabytes and multi-processor arrays is (1) persistent hash, and (2) user-storable evaluations. Rybka 3 hopes to give (1), and Crafty’s version of (2) is the only one I know to be flexible enough, and not with the best GUIs for my purposes. It now takes me the better part of a day just to intelligently play a big enough piece of my completed analysis into Fritz to teach its hash table that some lines are won, and if Microsoft decides its latest patch absolutely must be installed overnite, all that work is gone!