"A" must can beating "C" sometimes?İs it true according to you?
No
I can add that the difference in elo does not tell you if the weaker has positive probability to beat the stronger program.
A program may be weak in different ways.
random mover have chances to win against evey program that can lose to the perfect player when some deterministic weak program may have no chances because it is going to lose the same game again and again.
The Elo formula is not valid so far in the tails. (In fact for Human games it is valid nowhere, see the Sonas rating system.)
In practice any program, no matter how weak can win against a computer opponent, as computers can crash or die and will then forfeit on time. The probability for this is far greater than 1e-308. More something like once every 10 years, so 1 in 100,000 in a 1-hour game.
Any program thats is rated mere 1700 elo can beat the legal "random mover"(with no understanding of chess) any number of times, let alone the hypothetical 10 digit elo rated engine.
His question would make more sense if he uses ~1200 rated Xadreco/ChadChess as a lower limit and 3300 elo Rybka as a upper limit... and raise the question that way.
If thats the case, 1500 elo is the minimum limit below which the program can never get a single point.
For example, the engine like Bikjump may be able to get a draw or even win on a good day against Rybka after so many games, the engine rated below 1800 will never have a chance against the 3300. Therefore 1500 Elo difference limit is my best guess.
swami wrote:Any program thats is rated mere 1700 elo can beat the legal "random mover"(with no understanding of chess) any number of times, let alone the hypothetical 10 digit elo rated engine.
If the random mover is lucky it can beat every non perfect player.
The fact that you can expect not to see it beating 1700 player is only because the probability is very small and not because it is impossible.
swami wrote:Any program thats is rated mere 1700 elo can beat the legal "random mover"(with no understanding of chess) any number of times, let alone the hypothetical 10 digit elo rated engine.
If the random mover is lucky it can beat every non perfect player.
The fact that you can expect not to see it beating 1700 player is only because the probability is very small and not because it is impossible.
Uri
That probability is so incredibly small that the word "never" is absolutely justified.
swami wrote:Any program thats is rated mere 1700 elo can beat the legal "random mover"(with no understanding of chess) any number of times, let alone the hypothetical 10 digit elo rated engine.
If the random mover is lucky it can beat every non perfect player.
The fact that you can expect not to see it beating 1700 player is only because the probability is very small and not because it is impossible.
Uri
That probability is so incredibly small that the word "never" is absolutely justified.
I'd think the probability is nil.
What's the chance of TSCP (which really has enough working knowledge in chess and is 1700 elo btw) losing a single game to a 'legal random mover'?
Absolutely Zero unless there's some GUI or an engine bug.
swami wrote:Any program thats is rated mere 1700 elo can beat the legal "random mover"(with no understanding of chess) any number of times, let alone the hypothetical 10 digit elo rated engine.
If the random mover is lucky it can beat every non perfect player.
The fact that you can expect not to see it beating 1700 player is only because the probability is very small and not because it is impossible.
Uri
That probability is so incredibly small that the word "never" is absolutely justified.
I'd think the probability is nil.
What's the chance of TSCP (which really has enough working knowledge in chess and is 1700 elo btw) losing a single game to a 'legal random mover'?
Absolutely Zero unless there's some GUI or an engine bug.
I do not know what is the shortest game that you can mate against tscp
but I guess it is 19 moves or something like that so
my guess is that the probability may be 1/10^20 or something like that(the probability for repeating the shortest game is smaller but there is more than one possible game).
It means that if you play million games in one year you may need to wait 100 milion of milion of years to see one game when it happens.
I assume for the discussion that tscp is deterministic and does not suffer from factors like hardware failures.
swami wrote:Any program thats is rated mere 1700 elo can beat the legal "random mover"(with no understanding of chess) any number of times, let alone the hypothetical 10 digit elo rated engine.
If the random mover is lucky it can beat every non perfect player.
The fact that you can expect not to see it beating 1700 player is only because the probability is very small and not because it is impossible.
Uri
That probability is so incredibly small that the word "never" is absolutely justified.
I'd think the probability is nil.
What's the chance of TSCP (which really has enough working knowledge in chess and is 1700 elo btw) losing a single game to a 'legal random mover'?
Absolutely Zero unless there's some GUI or an engine bug.
I do not know what is the shortest game that you can mate against tscp
but I guess it is 19 moves or something like that so
my guess is that the probability may be 1/10^20 or something like that(the probability for repeating the shortest game is smaller but there is more than one possible game).
It means that if you play million games in one year you may need to wait 100 milion of milion of years to see one game when it happens.
I assume for the discussion that tscp is deterministic and does not suffer from factors like hardware failures.
Uri
Very good argument, Uri.
So I suppose the probability of Tscp winning(or draw) a single game against Rybka is lesser than the probability of "legal random mover" winning(or draw) a game against Rybka or TSCP?
Interesting indeed.
Engine with absolutely "No knowledge" can be sometimes better than engine with some "knowledge" (poor knowledge itself restricts the engine from seeing the unique moves)
Can someone play Random mover Vs. Rybka 3 1ply games until the random mover wins to see how much luck it needs?
I recall that for a random mover to beat another random mover at least 50 games were needed, it'd be interesting to know how would this curve shape against increasingly stronger opposition.
