How to beat it? Come up with your rock, paper, scissors picks in advance by rolling a 6 sided dice with 1-2 being rock, 3-4 being paper, and 5-6 being scissors.
The AI won’t win 60% of the then. I’d if shocked if it did and whatever program pattern it had to take advantage of people’s tendencies would be useless.
Or play and just pick the opposite of what you initially want half of the time. I suppose that would do it as well.
This isn’t you beating it. It’s a program you developed beating it.
In a game with three choices, how do you pick the “opposite” of what you initially are minded to choose? :rolleyes:
You said you’d wager that you’d win more than 60% of the time. I’d bet you’d win right about 50% of the time using your method.
You mean 33%.
For actual past results involving (hand-coded) AIs playing each other see here. Of particular interest is the “Iocaine Powder” bot which employs a “Sicilian reasoning” strategy. Maybe not as easy a strategy to follow as a human, though, any more than choosing random moves is.
Yep, because that makes it random, and in random RPS matchups it always normalizes out to draws over the long run.
I used an online dice roller to give me 50 results of 1 - 3 and then plugged that into the game.
I won: 16 times.
Computer won: 15 times.
We tied: 19 times.
Yes, yes, yes. I should have been more careful about my phrasing with that, but the 60% win percentage mentioned before was throwing away ties, so I was going with that assumption, but that never was explicitly specified.
Although the strongest chess computers have achieved an ELO rating as high or a bit higher than the strongest human player I would like to see an official FIDE world championship match between the best AI program and the current world champion Magnus Carlsen to settle it once and for all.
Why would that settle it once and for all? Once the AI beats him wouldn’t there just be some other chess champion down the line that should face the AI to “settle it once and for all”?
It was settled once and for all back in 2006. Since then there’s been few matches because almost nobody in the chess or AI community think there’s anything more to settle. They’ve moved on to can a state-of-the-art AI beat world-class chess players while handicapped (down a pawn, 2 pawns or knight)? The answer to that question so far seems to be not so much, although it is capable of drawing about 33% of the time. See, in particular, the Komodo handicap matches which are, to the best of my knowledge, the most recent.
Also, consider that most, if not all, of the top chess AI programs are available to the public generally for free. Don’t you think that Carlsen has already played them in his own time? Don’t you think that if he was beating them privately he might have gone to the AI developer and said “Hey, I’m beating your AI at top difficulty level. We should have a public match.” I can all but guarantee that if any top player were to contact one of these groups and tell them they can beat it, they would jump at the opportunity for a match. The fact that this hasn’t occurred suggests, while not being definitive proof, that Carlsen is not defeating top chess AI software.
And there are obviously still some computer-human matches, or it wouldn’t be possible to assign the computers an Elo rating at all. They’re just not publicized much any more.
As a nitpick, by the way, it’s “Elo”, not “ELO”. It’s named after its inventor, not an acronym.
I could be wrong, but I think I recall that the Elo rating for chess AIs are done separately from humans now in machine only pools.
From shudder Wikipedia:
“Since 2005–06, human-computer chess matches have demonstrated that chess computers are capable of defeating even the strongest human players (Deep Blue versus Garry Kasparov). However, ratings of computers are difficult to quantify. There have been too few games under tournament conditions to give computers or software engines an accurate rating.[33] Also, for chess engines, the rating is somewhat dependent on the machine that the program runs on.”
“Specialized machine-only Elo pools have been created for rating machines, but such numbers, while similar in appearance, should not be directly compared.”
If I think Rock, just throw the paper. Simple.
Mix it up.
Nothing wrong with picking a random time to throw the same solution out 4 times in a row.
Or there is nothing wrong with throwing out what the computer used last.
Just come up with a few random things like this, and NO AI will win 60% of the time.
Originally Posted by Silver lining View Post
How to beat it? Come up with your rock, paper, scissors picks in advance by rolling a 6 sided dice with 1-2 being rock, 3-4 being paper, and 5-6 being scissors.
The AI won’t win 60% of the then. I’d if shocked if it did and whatever program pattern it had to take advantage of people’s tendencies would be useless.
Or play and just pick the opposite of what you initially want half of the time. I suppose that would do it as well.
No. I said the AI won’t win 60% of the time. I posted my initial reply first above.
Your claim was that YOU would beat the AI “better than 60%” of the time using that strategy. Not that you could keep the AI below that number.
If you’re trying to claim that using a random strategy would result in a ~50% win rate, nobody will argue with you because that’s the same as claiming you could flip a coin a bunch of times and get ~50% heads.
You missed my point. There is no “opposite” to one of three choices. :rolleyes:
It is also not valid “Sicilian reasoning”. You would throw Rock to beat Scissors, right? But then your wily opponent would choose Paper instead.
Therefore your “opposite” would be Scissors, not Paper as claimed. 
???
And I would take that wager.
And, yes, nobody at all is arguing that you couldn’t reduce the AI’s winning percentage to chance (50% in the long run, discarding draws) by utilizing a purely random strategy. Of course not. I thought that was pretty clear from post 104 where I wrote “Humans clearly don’t play purely randomly, hence the AI’s better performance.”
And the fact that you think that this is a good strategy is exactly why the computer will win. If you are using any strategy that doesn’t involve rolling dice or some equivalent actual randomization, then the computer will figure out what strategy you’re using, and thus always know exactly what to throw to beat you. Yes, this applies even if you think your strategy is “pick one at random”. Humans are lousy at random, and when we think we’re being random, we’re really just following patterns that we don’t recognize. But the computers usually can recognize them.
Again, this isn’t theoretical. Computers really do consistently beat humans, in real-world tournaments.
I suspect one could come up with a mental pseudo-random generator that would work well enough that a computer couldn’t predict it easily. For instance, here’s one with a cycle length of 999:
Pick a 3-digit number to start (not 447)
Rotate the digits: 137 => 371
Multiply by 3
Add 25
Modulo 1000
Pick the middle digit, and choose RPS based on 0-2, 3-5, 6-8. Repeat on a 9.
Not good enough for real work, but as a scrambler that would confuse an AI, it’s very likely to succeed.