Tweek Posted March 9, 2016 Share Posted March 9, 2016 Doesn't it all depend on how many packs of cards are in the pile? Probability gets skewed when you can't card count. Most casinos have several packs in play for blackjack. Link to comment Share on other sites More sharing options...
thebees Posted March 9, 2016 Share Posted March 9, 2016 The boy did well and he knew he'd made a mistake early in the game, maybe he will win next time Link to comment Share on other sites More sharing options...
HeliX Posted March 10, 2016 Share Posted March 10, 2016 Doesn't it all depend on how many packs of cards are in the pile? Probability gets skewed when you can't card count. Most casinos have several packs in play for blackjack. This just makes it more difficult to card count, not impossible. If you know the number of decks in play it's still perfectly viable to count cards. That said, to be good enough at counting cards that you are also able to achieve the whole "making it not entirely obvious" part both in regards to your betting behaviour and the speed with which you play is the bit that requires real skill. Link to comment Share on other sites More sharing options...
Chinahand Posted March 10, 2016 Author Share Posted March 10, 2016 Two down - two victories to silicon! After the match, Lee Se-dol said: "Yesterday I was surprised but today it's more than that, I am quite speechless. "Today I feel like AlphaGo played a nearly perfect game," he said. What is slightly creepy is that we have no idea how the computer is able to play so well - it just learnt to do it itself, and it isn't telling! The next major breakthrough needed in AI - getting AI to explain to us puny humans how they do it. Link to comment Share on other sites More sharing options...
Mr. Sausages Posted March 10, 2016 Share Posted March 10, 2016 There's obviously a man inside the machine. https://en.wikipedia.org/wiki/The_Turk Link to comment Share on other sites More sharing options...
notwell Posted March 10, 2016 Share Posted March 10, 2016 Doesn't it all depend on how many packs of cards are in the pile? Probability gets skewed when you can't card count. Most casinos have several packs in play for blackjack. This just makes it more difficult to card count, not impossible. If you know the number of decks in play it's still perfectly viable to count cards. That said, to be good enough at counting cards that you are also able to achieve the whole "making it not entirely obvious" part both in regards to your betting behaviour and the speed with which you play is the bit that requires real skill. I agree . Counting cards is relatively straightforward. Not being detected is the challenge any serious card counter faces. Link to comment Share on other sites More sharing options...
pongo Posted March 10, 2016 Share Posted March 10, 2016 Once a game is solved it's time to invent new and better games rather than trying to detect people using a proven strategy or a computer. Link to comment Share on other sites More sharing options...
pongo Posted March 10, 2016 Share Posted March 10, 2016 What is slightly creepy is that we have no idea how the computer is able to play so well This is very unlikely to be true. It sounds much more like marketing spin. Link to comment Share on other sites More sharing options...
Chinahand Posted March 10, 2016 Author Share Posted March 10, 2016 All they know is the weights of the neural network. Why they are these weights and what might be a better setting is totally unknown. The network has set these weights through it learning from so many tens of thousands of standard games and then playing against itself for millions of more games. If you built another network and started it off in a different configuration it is highly unlikely it would end up with the same weights after a similar process and the programmers would have no idea which might end up consistently beating the other. This type of deep machine learning is truly creative; the programmers are not deterministically telling it what to do, rather it is genuinely learning what to do itself and it is basically impossible to reverse engineer this into a heuristic to teach a person how to play. Link to comment Share on other sites More sharing options...
notwell Posted March 10, 2016 Share Posted March 10, 2016 Once a game is solved it's time to invent new and better games rather than trying to detect people using a proven strategy or a computer. In the case of poker computers have some use but really only to calculate your outs and chances of holding a top hand at a given point. With online poker most of the games are so short in time frame (turbo and hyper) that it is a lottery anyway. Link to comment Share on other sites More sharing options...
pongo Posted March 10, 2016 Share Posted March 10, 2016 the programmers are not deterministically telling it what to do, rather it is genuinely learning what to do itself and it is basically impossible to reverse engineer this into a heuristic to teach a person how to play. You can teach a person how to build and program a machine to use to solve the problem. If you built another network and started it off in a different configuration it is highly unlikely it would end up with the same weights after a similar process and the programmers would have no idea which might end up consistently beating the other. After an infinite number of games against each other they would presumably play to the same standard. If it could be shown that they might not then that would be genuinely surprising. Link to comment Share on other sites More sharing options...
Chinahand Posted March 13, 2016 Author Share Posted March 13, 2016 Stinky has already announced it in my Chinahand's Random Stuff Thread, but I thought I'd add something here. AlphaGo won 3 games in a row - taking an unassailable lead in the 5 game tournament, but in the fourth match Lee Sedol was victorious; the first time a human has defeated the neural network. That shows how things are still reasonably finely balanced - a human can still defeat the machine. We gave up trying to do that with cars probably 100 years ago, chess computers 10 years ago - I wonder if in a years time with probably double the processors and double the memory Go will join the list where the artificial has such a lead over the human competitions between us become pointless. The fascinating issue is that the result of this might be expert systems - will economists start trusting AI to make decisions on interest rates, or doctors on operations? The Computer Says No is a joke of our time - but it will be interesting seeing how they start influencing our affairs more and more over the next few decades. Link to comment Share on other sites More sharing options...
the stinking enigma Posted March 13, 2016 Share Posted March 13, 2016 sorry chinahand i thought that was the right one. intelligence in a game with set rules is one thing. if AI reaches or surpasses human levels of cunning or deviousness then thats quite a scary prospect for me. i hope these machines at least have an inbuilt safe word in case it all goes a bit too far Link to comment Share on other sites More sharing options...
Chinahand Posted March 13, 2016 Author Share Posted March 13, 2016 After an infinite number of games against each other they would presumably play to the same standard. If it could be shown that they might not then that would be genuinely surprising. It is senseless to talk about an infinite number of games, but even in that extreme example I'm not certain they are guaranteed to converge. Firstly there are likely to be various equilibrium points where a step-wise alteration of the network weights will find its current position is a local maxima - it is at the top of a peak, but not the highest peak. Also there is an element of the random walk in how it learns. If you have flipped a coin a thousand times and find that you have flipped heads 104 times more than you've flipped tails there is nothing pulling the score back towards equal heads and tales - the law of large numbers will make the probability of a head or tale converge to 0.5, but the absolute distance from equilibrium will not converge to zero - it will just wander, and could wander quite a distance from zero. Two coins flipped an infinite number of times will quite definitely NOT have the same number of heads. Link to comment Share on other sites More sharing options...
Chinahand Posted March 13, 2016 Author Share Posted March 13, 2016 sorry chinahand i thought that was the right one. No problem at all! Link to comment Share on other sites More sharing options...
Recommended Posts
Archived
This topic is now archived and is closed to further replies.