Something Hikaru Said

[rcmaddox]

Just minutes before the games yesterday, in the Chess.com chat, Hikaru responded to members speculating about how long it will be before a chess engine can give piece odds to a GM.

He was concise. He said it will never happen. Hikaru believes that piece odds will now and forever favor the GM.

Any dissenters?

[royb]

Never is a very long time as the saying goes.

Surely that time will come and I believe it will be well within Hikaru's lifetime (I would guess even much sooner -- within 10 years). No real knowledge involved in that guess, other than a knowledge of how software and hardware has improved over the last 10 years as a rough estimate.

[Dan Cooper]

Sounds like a challenge to me.

[bob]

Sounds like a challenge to me.

How that challenge has changed over the years:

1970: no computer will be able to beat an IM (David Levy) within 10 years (was actually stretched to maybe 15 years).

1980 or so: No computer will be able to beat a GM. Until Deep Thought.

1980 or so: No computer will ever be able to beat the current world champion (a GM obviously) and backed up by the Fredkin Foundation prize. Again, along came deep thought in 1997.

2016: No computer will ever be able to beat a GM if he is given knight odds.

2025: No computer will ever be able to beat a GM if he is given queen odds.



etc...

[Laskos]

Sounds like a challenge to me.

How that challenge has changed over the years:

1970: no computer will be able to beat an IM (David Levy) within 10 years (was actually stretched to maybe 15 years).

1980 or so: No computer will be able to beat a GM. Until Deep Thought.

1980 or so: No computer will ever be able to beat the current world champion (a GM obviously) and backed up by the Fredkin Foundation prize. Again, along came deep thought in 1997.

2016: No computer will ever be able to beat a GM if he is given knight odds.

2025: No computer will ever be able to beat a GM if he is given queen odds.



etc...

Actually Naka seems right to me for GMs above ELO 2700, at least against simple 32 men tablebases and aside occasional obvious large blunders by GM. Maybe a "GM-fooling" super-engine can be set in the future, to play nasty tricks on humans, and in this case Naka is still right for the case of a future GM who learns a bit how to play against such an engine.

[bob]

Sounds like a challenge to me.

How that challenge has changed over the years:

1970: no computer will be able to beat an IM (David Levy) within 10 years (was actually stretched to maybe 15 years).

1980 or so: No computer will be able to beat a GM. Until Deep Thought.

1980 or so: No computer will ever be able to beat the current world champion (a GM obviously) and backed up by the Fredkin Foundation prize. Again, along came deep thought in 1997.

2016: No computer will ever be able to beat a GM if he is given knight odds.

2025: No computer will ever be able to beat a GM if he is given queen odds.



etc...

Actually Naka seems right to me for GMs above ELO 2700, at least against simple 32 men tablebases and aside occasional obvious large blunders by GM. Maybe a "GM-fooling" super-engine can be set in the future, to play nasty tricks on humans, and in this case Naka is still right for the case of a future GM who learns a bit how to play against such an engine.

My take from watching this stuff for 50 years now is that humans make mistakes, no matter how strong they are. A knight up and then they drop a pawn, and then another pawn, and before you know it... The only obvious strategy to me is to trade whenever possible so that the knight becomes more and more important. But most computer programs understand the "when behind in material, trade pawns but not pieces and vice-versa" which might tend those games to draws more than anything else.

[Laskos]

Sounds like a challenge to me.

How that challenge has changed over the years:

1970: no computer will be able to beat an IM (David Levy) within 10 years (was actually stretched to maybe 15 years).

1980 or so: No computer will be able to beat a GM. Until Deep Thought.

1980 or so: No computer will ever be able to beat the current world champion (a GM obviously) and backed up by the Fredkin Foundation prize. Again, along came deep thought in 1997.

2016: No computer will ever be able to beat a GM if he is given knight odds.

2025: No computer will ever be able to beat a GM if he is given queen odds.



etc...

Actually Naka seems right to me for GMs above ELO 2700, at least against simple 32 men tablebases and aside occasional obvious large blunders by GM. Maybe a "GM-fooling" super-engine can be set in the future, to play nasty tricks on humans, and in this case Naka is still right for the case of a future GM who learns a bit how to play against such an engine.

My take from watching this stuff for 50 years now is that humans make mistakes, no matter how strong they are. A knight up and then they drop a pawn, and then another pawn, and before you know it... The only obvious strategy to me is to trade whenever possible so that the knight becomes more and more important. But most computer programs understand the "when behind in material, trade pawns but not pieces and vice-versa" which might tend those games to draws more than anything else.

Knight odds were played recently and 2150-2200 ELO FM beat Komodo on 24 cores (~3400 ELO CCRL or CEGT) consistently. The value of such odds against an FM is inferred from this and from some simulations as at least 1300 (computer) ELO points, probably more. Also, simple rule is that the ELO value of the handicap increases with strength. Say, Knight odds are less than 600 ELO points if played between an 1800 ranking and 1200 ranking. I believe that Knight odds value given to a top GM (2750+) are at lest 1800 computer ELO points, probably more, and to beat such a GM the engines would need to be above CCRL ELO 4600, which in several extrapolations, is above 32 men tablebasses.

I belive the biggest odds a perfect engine (simple 32 men tablebases) can give to a top GM and still win on average would be 2 important pawns, f7 pawn + c7 pawn. Maybe even this is too much to give to a prepared Carlsen.

However I do agree that such things as blunders, bad time management, the inadequate style of playing can make GM lose, but a human can also pass out playing against 32 men tablebases, I am not addressing these issues.

[Dirt]

Actually Naka seems right to me for GMs above ELO 2700, at least against simple 32 men tablebases and aside occasional obvious large blunders by GM.

It probably depends on the time control. At a one minute game even Hikaru would probably lose today. At game in two hours plus increment it will probably never happen.

[Laskos]

Actually Naka seems right to me for GMs above ELO 2700, at least against simple 32 men tablebases and aside occasional obvious large blunders by GM.

It probably depends on the time control. At a one minute game even Hikaru would probably lose today. At game in two hours plus increment it will probably never happen.

Yes, I am talking here always only about 2 hour normal time control.

[Sean Evans]

Just minutes before the games yesterday, in the Chess.com chat, Hikaru responded to members speculating about how long it will be before a chess engine can give piece odds to a GM.

He was concise. He said it will never happen. Hikaru believes that piece odds will now and forever favor the GM.

Any dissenters?

Hikaru recently lost to Komodo in an odds match.

http://www.chess.com/news/komodo-beats- ... attle-1331