what is the speed difference between humans and computers

[Uri Blass]

Suppose humans are trained to use exactly the same algorithm that chess program use when they play chess at fixed depth on a single core(it is impossible to use the same algorithm that chess program used in games because it means looking very often in the clock that humans cannot do).
Suppose that humans are also allowed to use pen and paper and books so they will have no problem of memory.

I guess that chess programs are going to be more than 1,000,000 times faster and that the difference is going to be bigger at bigger depth because humans are going to need time to find the notes that they have but the question is what is the speed difference and if humans can get the difference to be smaller.

Note that I guess that the difference is dependent on the program.


I think that it may be interesting to have some tournament of humans at 120/40 time control with rules that humans have to follow the code of chess programs without a computer and need to give the source of the program before the tournament(and a player is losing if he plays a move that cannot be reproduced by his program when the referee is going to use the program to check it).

I do not think that the existing chess programs are optimized for this task
and people may write a different code for the tournament.

Uri

[Peter Hegger]

Hello
Would a human using pen, paper and books have enough time to calculate exactly as a chess program does in a 40/120 game?
Aren't computers already calculating a million times faster (maybe not on single core) than humans in NPS even when humans are allowed to play their own way?
regards,
peter

[tjfroh]

Uri,

Speed of calculation is irrelevant. We have visual circuits with hundreds of billions of neurons firing a few times per second. Coupled with an associational cerebral cortex of even more neurons, this gives us the ability to make the "right move" or to invent any number of novel ideas or devices.

The computer is just another hammer to drive nails. It is a very fast hammer but all it does is drive bits (nails).

Tj

[bob]

Suppose humans are trained to use exactly the same algorithm that chess program use when they play chess at fixed depth on a single core(it is impossible to use the same algorithm that chess program used in games because it means looking very often in the clock that humans cannot do).
Suppose that humans are also allowed to use pen and paper and books so they will have no problem of memory.

I guess that chess programs are going to be more than 1,000,000 times faster and that the difference is going to be bigger at bigger depth because humans are going to need time to find the notes that they have but the question is what is the speed difference and if humans can get the difference to be smaller.

Note that I guess that the difference is dependent on the program.


I think that it may be interesting to have some tournament of humans at 120/40 time control with rules that humans have to follow the code of chess programs without a computer and need to give the source of the program before the tournament(and a player is losing if he plays a move that cannot be reproduced by his program when the referee is going to use the program to check it).

I do not think that the existing chess programs are optimized for this task
and people may write a different code for the tournament.

Uri

If you have to "follow the code" you can forget about it. I do not believe a human on the planet can do what Crafty does, line by line, in 3-4 minutes of time. So the idea of "following the code" is pointless. If you want the human to follow the same "approach" you might get somewhere. Humans might search 1 node per second, unless you require them to use a computer-like evaluation. In which case you are back to one node per move once again...

[Uri Blass]

Uri,

Speed of calculation is irrelevant. We have visual circuits with hundreds of billions of neurons firing a few times per second. Coupled with an associational cerebral cortex of even more neurons, this gives us the ability to make the "right move" or to invent any number of novel ideas or devices.

The computer is just another hammer to drive nails. It is a very fast hammer but all it does is drive bits (nails).

Tj

I disagree
Speed of calculation is not irrelevant and the question is interesting.

I suspect that part of the advantage of strong chess players is speed of calculation.

chess programs are optimized for speed of calculation of chess programs
and not for speed of calculation of humans so the speed difference between stockfish and human who follow stockfish is an upper bound for the speed difference.

It is possible to develop chess programs that are optimized for speed of calculation of humans but unfortunately there are no tournament that encourage it.

Uri

[Michael Sherwin]

Suppose humans are trained to use exactly the same algorithm that chess program use when they play chess at fixed depth on a single core(it is impossible to use the same algorithm that chess program used in games because it means looking very often in the clock that humans cannot do).
Suppose that humans are also allowed to use pen and paper and books so they will have no problem of memory.

I guess that chess programs are going to be more than 1,000,000 times faster and that the difference is going to be bigger at bigger depth because humans are going to need time to find the notes that they have but the question is what is the speed difference and if humans can get the difference to be smaller.

Note that I guess that the difference is dependent on the program.


I think that it may be interesting to have some tournament of humans at 120/40 time control with rules that humans have to follow the code of chess programs without a computer and need to give the source of the program before the tournament(and a player is losing if he plays a move that cannot be reproduced by his program when the referee is going to use the program to check it).

I do not think that the existing chess programs are optimized for this task
and people may write a different code for the tournament.

Uri

I think that what is meant here is if a human could follow a formal procedure to come up with a move rather than an ad hoc method of just scanning the board looking for 'stuff', then what would that be. Could the procedure be written down? Can this 'program' be used to verify the humans move choice? Can a computer then be programed to follow this same procedure?

[Rafael Vasquez]

Uri,

Correct me if I´m wrong but I thougth that chess programs are just
the numerical methods for trying to do a chess calculation.
What human GM has that a normal human doesn´t is consistency.
By training a GM achieves consistency in finding good moves.
There are different methods for doing this. Kasparov has a method.
Kramnik has another method.
The difference in human is that the speed of aplication can vary if the human is sick, getting older, getting distracted, etc...
I remember Kasparov played a game below his level against Fritz4 in Germany and he complained about being distracted by the crowd.
If the same human method is fully translated to a program, then the program is stronger by being more consistent in aplying it to EVERY move and at the same consistent speed.

Rafael

[lmader]

I think it is a huge mistake to presume that humans think in purely algorithmic terms, such as the way computer programs play chess.

The way in which the human mind is trained to play chess may appear algorithmic from the types of regimens that a player uses, such as studying tactics and end-games, etc. And reading out the tactics of a position can be very linear. But when it comes to the "evaluation function", if you will, that a human employs when looking at a position, this is something that develops in a human over years. The kind of intuition that a player has for a position includes years of experience, memory, and emotional responses that all come together to weight things in extremely ineffable ways.

So although it might be interesting to give a human an algorithm to follow, this would not simulate anything about how humans think about chess.

[Dann Corbit]

http://www.frc.ri.cmu.edu/~hpm/talks/re ... er.aug.gif