Google's AI crushes Go Grand Master

[Fuzz]

Sounds pretty human to me, then.

[jammies]

I think AI development will eventually discard the idea of copying the human model, and then vastly surpass human-level intelligence in short order. Right now, most AI development is along the lines of how you might design a car if you were basing its locomotion on human legs. Once you put wheels on your car instead, you can go a heck of a lot faster with a much simpler mechanism.

Of course, conceptualizing a different method of intelligence is not exactly a trivial task, else it would already have been done. It could happen tomorrow, it could happen decades from now, or it might not happen for centuries, if ever. It might end up being far simpler to adapt biological brains to the tasks we want AI to perform, or simply interface human brains with computers to create hybrids with the advantages of both.

[Fuzz]

I know a few human brains that could benefit mankind by being wiped and re-purposed. We can start with Trump.

[peter12]

Originally Posted by jammies

I think AI development will eventually discard the idea of copying the human model, and then vastly surpass human-level intelligence in short order. Right now, most AI development is along the lines of how you might design a car if you were basing its locomotion on human legs. Once you put wheels on your car instead, you can go a heck of a lot faster with a much simpler mechanism.

Of course, conceptualizing a different method of intelligence is not exactly a trivial task, else it would already have been done. It could happen tomorrow, it could happen decades from now, or it might not happen for centuries, if ever. It might end up being far simpler to adapt biological brains to the tasks we want AI to perform, or simply interface human brains with computers to create hybrids with the advantages of both.

Well, this is the trick. It ain't easily conceptualized. We can't even conceptualize our own intelligence.

[psyang]

Originally Posted by peter12

As I said above, Fan Hui, Alpha Go's first opponent, has improved his game immensely since his 5-0 series loss to the Google program. So yes, humans excel at transfer learning, and high-end Go players can understand some of the machine's moves, and adapt. Sedol's win is the real story here, in my opinion.

My personal view is that a lot of this so-called AI development is just hubris on the part of the programmers, VC's etc... If it proves to be all but impossible for a machine to develop a general intelligence, like humans, and all that occurs is the automation of a handful of routine-based tasks or token victories against human game players, then I have to ask, "what is the point."

A huge part of these games is that they are duels between equals. A machine doesn't get tired, it doesn't have any tells or give anything away. It also doesn't know why it made a move. Kasparov was beaten by a Deep Blue glitch - a move so out of the ordinary that it threw him for the rest of the series, which at the time, he attributed to intelligence. Now we know that the program had reached a point where it did not have any optimal moves so went to a default random move.

The "human weakness" in these matches definitely favors the AI. Lee was caught several times looking at AlphaGo's human "handler" (who actually placed the stones for AG) to try to spot a tell. Lee himself said it took a while to adjust his game with the knowledge that he was playing an emotionless computer.

Further, the qualities of beauty, emotion, and "flow" in a go game can be seen as human weaknesses as they blind a player to other ways of playing. However, you could argue this unfair advantage that AlphaGo had was beneficial. That there even was another way of playing go was inconceivable before AlphaGo started winning.

Maybe AlphaGo won't be curing cancer in the near future, but it can certainly teach us more about ourselves and the world around us in ways we couldn't by ourselves.

[Locke]

Originally Posted by Fuzz

I know a few human brains that could benefit mankind by being wiped and re-purposed. We can start with Trump.

How do you know we havent already?

What secrets does that combover hide??

[Fuzz]

Originally Posted by Locke

How do you know we havent already?

What secrets does that combover hide??

You're right. He does seam to be very repetitive and simple. Must be V1.0. Time for an OS upgrade.

[peter12]

Originally Posted by psyang

The "human weakness" in these matches definitely favors the AI. Lee was caught several times looking at AlphaGo's human "handler" (who actually placed the stones for AG) to try to spot a tell. Lee himself said it took a while to adjust his game with the knowledge that he was playing an emotionless computer.

Further, the qualities of beauty, emotion, and "flow" in a go game can be seen as human weaknesses as they blind a player to other ways of playing. However, you could argue this unfair advantage that AlphaGo had was beneficial. That there even was another way of playing go was inconceivable before AlphaGo started winning.

Maybe AlphaGo won't be curing cancer in the near future, but it can certainly teach us more about ourselves and the world around us in ways we couldn't by ourselves.

I think you are right in that they open the human player up to a new cognitive process previously completely unavailable. Professional Go players only get to that point by essentially doing what Alpha Go did - memorizing thousands of matches, classic players, and problems. So you may have a good point there. By removing that, it forces human players to not just adapt but completely reformat their strategy - hence the massive improvement in Fan Hui's game.

You are right to say that these kinds of advances - novel or not - present new challenges to human self-understanding, but I don't know if they are productive, or not.

The thing that really scares me about AI research is how quickly humans want to denigrate themselves in service to these algorithms. Jaron Lanier makes this point in his essay "The Myth of AI." Essentially, by believing that AIs present some kind of new closed system of thinking or innovating or practicing things exclusively in the domain of humans, we a) forget the important role that human data plays in the functioning of the AI, and b) we forget what the system was like before it was closed by the algorithm.

So Go played with a machine programmed to play the game in an intuitive manner is not the same as two humans playing Go, and just because the behaviours are radically different, doesn't make them any better or necessarily interesting.

https://www.edge.org/conversation/ja...the-myth-of-ai

[jammies]

Originally Posted by peter12

Well, this is the trick. It ain't easily conceptualized. We can't even conceptualize our own intelligence.[

Haven't yet, not can't. It's undoubtedly the most complex of all problems we've yet encountered, but I don't see it as being an insoluble problem. We learn more all the time, and its complexity isn't infinite, so eventually we'll get there.

Again, though, that might not be in our lifetimes. Or ever, I suppose, if we destroy ourselves, or no longer have the resources to spend on research due to needing those resources for survival.

[peter12]

Originally Posted by jammies

Haven't yet, not can't. It's undoubtedly the most complex of all problems we've yet encountered, but I don't see it as being an insoluble problem. We learn more all the time, and its complexity isn't infinite, so eventually we'll get there.

Again, though, that might not be in our lifetimes. Or ever, I suppose, if we destroy ourselves, or no longer have the resources to spend on research due to needing those resources for survival.

I think Chomsky, and other skeptics are right - there are some problems, like the origins of the universe, the development of speech in humans, and the problems of consciousness are puzzles that exist outside of our cognitive capabilities as a species.

It is like how a rat in a cage can figure out that taking every second left turn in a maze will lead him to the cheesy exit, but rat is absolutely stumped when the maze is configured so that only every left turn on a prime number will lead to the exit.

Yeah, and I realize that skepticism can be an easy way out, but so is blind faith in progress.

[jammies]

Originally Posted by peter12

I think Chomsky, and other skeptics are right - there are some problems, like the origins of the universe, the development of speech in humans, and the problems of consciousness are puzzles that exist outside of our cognitive capabilities as a species.

That's the allure of developing AI without relying on human models, though - I'm not convinced that consciousness is a mandatory feature of intelligence, so it may not be necessary to understand one to create the other. Also there is the example of evolution to show that intelligence can arise out of non-intelligence, so it is not unreasonable to think that deliberately harnessing evolutionary algorithms can lead to an intelligence unfathomable by the very beings that create it.

[mrkajz44]

I find the "start from scratch" idea the most interesting. And I'm talking truly from scratch - all it knows is there is gameboard and pieces. It might not even know you need to take turns. Then each time it violates the rules, you let the program know it did something wrong (but don't tell it what it did wrong). I wonder if it could slowly learn all the rules of the game this way, and then could truly play games against itself without any sort of interference at all.

[peter12]

Google's DeepMind has already done this with Atari games.

[btimbit]

Originally Posted by Dan02

One thing I have learned from this thread, is damn i am absolutely terrible at GO.

Only thing I learned is that there is a game called GO

[peter12]

Absolutely fascinating article regarding Sedol's victory in Game 4. Basically, he pursued an aggressive, almost random strategy that off-set the probability calculations the machine relies on to predict victory.

https://gogameguru.com/lee-sedol-def...meback-game-4/