2015-2020 if there will be top supercomputer in full power & software like MoGo or Zen (if last can scaled like MoGo)
P. Reichart - 13 years ago
The question even more interesting is: When will we be able to teach a computer think like a human - AND - How do we ever know a computer "thinks" like a human? Up to now the solution more or less was found in increasing computing speed - the computer just tries, and tries, and tries... the faster the better... However, what is a human being doing? What about the fact of successful intuition? How can we "teach" a machine to be "intuitive"?
To make the matter short: I am not afraid of a winning machine. Cars can go faster than humans can run. AND? Is that kind of progress we should be afraid of - or just enjoy it? BUT what if a machine ever starts to "think" its own way? I suppose winning the bet in question by the computer would not necessarily be a time of happiness - would it? What - just to look a little further into future - would we feel/think about a machine ever "saying": I bet no human being will ever be able to beat me in any game... Good luck, humanity!
Jean Michel - 13 years ago
I am European 5dan and have been very favorably impressed by the progress of MCTS.
Putting more computer power now works. From the speed of progress I expect programs to beat me within ten years and be world champion within 20 years. So I expect 2025-2030 as the right timeframe. Already the computers reached professional level on 9x9, and I expect them to
reach professional level in 13x13 in less than ten years.
Charles Leedham-Green - 13 years ago
The basic reason that MCTS is so spectacularly successful is that it manages to look far enough
ahead, in fact to the end of the game, in a reasonable way, but avoiding an exponential
explosion. Thus it can read out a long ladder without too much trouble. Where it will fail
is in situations where (unlike a ladder) the number of critical variations increases exponentially.
If a fight becomes complicated enough the exponential explosion will hit the computer, but
not the professional. The professional will be able to discard irrelevant forcing moves, and
allow for the effect of relevant forcing moves, without having to take into consideration
issues of timing and so forth when these are irrelevant. The machine will punish errors
provided that the situation is simple enough; but my hunch is that a professional will be able
to steer the game into a position that is too hard for any machine; the number of critical lines
is exploding exponentially, and the machine rolls over.
Mark Schreiber - 14 years ago
I think a more interesting poll is when will a computer achieve amateur 7 dan. This is more interesting because this will happen sooner than beating the world champion and amateur 7 dan is very strong. Maybe you can add another poll and we will see how the numbers compare.
There is absolutely anything that suggests that a gobot could beat human in any foreseeable future. Not before significant improvement with AI technology. And up to date we have absolutely anything that even hints to the direction of strong AI. Thus I have bet €1000 to support my opinion. If you do not believe me, see my website for instructions how to challenge me.
David Doshay - 14 years ago
The decade with the most votes by far is 2020-2030. I can believe that a computer might beat the human world champ in one game by then, but not in a 10 game match. I added another 20 years under the assumption that it would be the same kind of a bet. Right now humans quickly figure out how to stop loosing to the present crop of MCTS programs. I predict that this is a situation that will not change for a very long time: after loosing even a few games in a row, the person will suddenly understand how to stop loosing and start winning, and then the computer will not win any more games.
Darren Cook - 15 years ago
Hi pookpooi! At the moment 2025-2030 is actually ahead, but 2015-2020 and 2020-2025 are only one vote behind.
A quarter of people are saying within the next 10 years; it would be amazing if that turns out to be the case. But if it does I'm fairly confident just 10 to 20 times faster hardware won't be the reason - it would have to be another breakthrough algorithm.
pookpooi - 15 years ago
I'm looking at Top500 performance chart and all of present evidences point to 2025-2030 (even though Go is software problem, not hardware)
2015-2020 if there will be top supercomputer in full power & software like MoGo or Zen (if last can scaled like MoGo)
The question even more interesting is: When will we be able to teach a computer think like a human - AND - How do we ever know a computer "thinks" like a human? Up to now the solution more or less was found in increasing computing speed - the computer just tries, and tries, and tries... the faster the better... However, what is a human being doing? What about the fact of successful intuition? How can we "teach" a machine to be "intuitive"?
To make the matter short: I am not afraid of a winning machine. Cars can go faster than humans can run. AND? Is that kind of progress we should be afraid of - or just enjoy it? BUT what if a machine ever starts to "think" its own way? I suppose winning the bet in question by the computer would not necessarily be a time of happiness - would it? What - just to look a little further into future - would we feel/think about a machine ever "saying": I bet no human being will ever be able to beat me in any game... Good luck, humanity!
I am European 5dan and have been very favorably impressed by the progress of MCTS.
Putting more computer power now works. From the speed of progress I expect programs to beat me within ten years and be world champion within 20 years. So I expect 2025-2030 as the right timeframe. Already the computers reached professional level on 9x9, and I expect them to
reach professional level in 13x13 in less than ten years.
The basic reason that MCTS is so spectacularly successful is that it manages to look far enough
ahead, in fact to the end of the game, in a reasonable way, but avoiding an exponential
explosion. Thus it can read out a long ladder without too much trouble. Where it will fail
is in situations where (unlike a ladder) the number of critical variations increases exponentially.
If a fight becomes complicated enough the exponential explosion will hit the computer, but
not the professional. The professional will be able to discard irrelevant forcing moves, and
allow for the effect of relevant forcing moves, without having to take into consideration
issues of timing and so forth when these are irrelevant. The machine will punish errors
provided that the situation is simple enough; but my hunch is that a professional will be able
to steer the game into a position that is too hard for any machine; the number of critical lines
is exploding exponentially, and the machine rolls over.
I think a more interesting poll is when will a computer achieve amateur 7 dan. This is more interesting because this will happen sooner than beating the world champion and amateur 7 dan is very strong. Maybe you can add another poll and we will see how the numbers compare.
There is absolutely anything that suggests that a gobot could beat human in any foreseeable future. Not before significant improvement with AI technology. And up to date we have absolutely anything that even hints to the direction of strong AI. Thus I have bet €1000 to support my opinion. If you do not believe me, see my website for instructions how to challenge me.
The decade with the most votes by far is 2020-2030. I can believe that a computer might beat the human world champ in one game by then, but not in a 10 game match. I added another 20 years under the assumption that it would be the same kind of a bet. Right now humans quickly figure out how to stop loosing to the present crop of MCTS programs. I predict that this is a situation that will not change for a very long time: after loosing even a few games in a row, the person will suddenly understand how to stop loosing and start winning, and then the computer will not win any more games.
Hi pookpooi! At the moment 2025-2030 is actually ahead, but 2015-2020 and 2020-2025 are only one vote behind.
A quarter of people are saying within the next 10 years; it would be amazing if that turns out to be the case. But if it does I'm fairly confident just 10 to 20 times faster hardware won't be the reason - it would have to be another breakthrough algorithm.
I'm looking at Top500 performance chart and all of present evidences point to 2025-2030 (even though Go is software problem, not hardware)