8 ms·

The players who played against AlphaGo and lost said they got something valuable out of it. They started looking at completely new possible combination of moves

The players who played against AlphaGo and lost said they got something valuable out of it. They started looking at completely new possible combination of moves they hadn't thought viable before. (The guy that played against AlphaGo before Lee Sedol said how it improved him. He even climbed several ranks.) This makes me wonder. Maybe the human brain never computers for global optimas. It gets to a point where it's good enough and then stops. Clearly, if they play vs AI and then start improving their strategy again, the brain might have thought defeating most human is a good enough strategy. This is also see in other games, someone who plays in "easy" difficulty or low ranked games in multi player games are stuck with "bad habits". As in they do really well in those thresholds but perform poorly when introduced to higher difficulty. But as soon as there is a need, good players rapidly adjust. (Example: In MOBA games like leauge of legend or Dota, higher ranked players adapt at a much higher rate to new emergent strategies while lower level player tend to not manage this and find things unbalanced.) My point is, as we discover more of these `slighly better than human AI`, we'll also discover that human brain might be capable of much more. It just never saw the need to pour more resources in it till it found a healthy competition. Our only (soon to be) sentient friend, the AI. -------------------------------------- [0] https://www.youtube.com/watch?v=0X-NdPtFKq0 https://www.youtube.com/watch?v=0X-NdPtFKq0

mickronome 10y ago
In a very crude sense, we learn by observing patterns that work, and without new working patterns we can tey to come up with novel patterns. But this is much harder, especially if the new patterns require a completely different approach instead of being an incremental change. I fully expect human ability to drastically increase in several areas when computers start finding new working patterns.
red75prime 10y ago
The system notices anthropomorphization of the system. The list of allowances is updated. Recalculating optimal behavioral responses. Done. Estimated probability of winning increased by 0.12. Thank you, you are welcome. We'll get along wonderfully!
- mindcrime 10y ago
  Don't anthropomorphize non-human systems. They hate it when you do that.
logicallee 10y ago
What an interesting thought!! It led me to another: you know how AI can imitate painting styles? Well, perhaps an AI could also figure out a style, by itself (i.e. given constraints, but no style to follow.) i.e. it figures out a new one! But which has a novel effect, on humans and as judged by a computer. Once we have been shown the style, humans could improve by starting to use that! We might like the effect, be able to judge that it is good, but just never saw that way before. Imagine if AI had invented pointillism, impressionism, cubism, etc! Imagine if it figures out the next "global optimum" from putting splotches of paint down, showing human artists a new way to see. Wow.
- EGreg 10y ago
  When humans are the judge, how are you going to test trillions of combinations per minute?
  - logicallee 10y ago
    I don't know. I am assuming there is some other way to judge "what something looks like!" - something that an AI could learn, be programmed with, derive, or invent. My assumption could be false and perhaps there is no way to approach aesthetics with anything other than a human brain, or training on the outputs of human brains. But if the AI had brain-like features, couldn't it be its own judge? Couldn't it have as an emergent property, its own sense of aesthetics or meaning, which it could then approach using its own naive approaches. Which humans could say "Wow" to. AI as the ultimate naive painter. What a concept! I realize it presupposes a lot that may never come to pass.
  - eru 10y ago
    A/B testing.
mindcrime 10y ago
Maybe the human brain never computers for global optimas. It gets to a point where it's good enough and then stops. That sounds a lot like Herbert Simon's idea of "satisficing"[1]. Satisficing is a decision-making strategy or cognitive heuristic that entails searching through the available alternatives until an acceptability threshold is met. <snip> Simon used satisficing to explain the behavior of decision makers under circumstances in which an optimal solution cannot be determined. He maintained that many natural problems are characterized by computational intractability or a lack of information, both of which preclude the use of mathematical optimization procedures. Consequently, he observed in his Nobel Prize speech that "decision makers can satisfice either by finding optimum solutions for a simplified world, or by finding satisfactory solutions for a more realistic world. Neither approach, in general, dominates the other, and both have continued to co-exist in the world of management science. [1]: https://en.wikipedia.org/wiki/Satisficing https://en.wikipedia.org/wiki/Satisficing
- beefman 10y ago
  Indeed, Ken Regan has attempted to measure it in chess.[1] The techniques used for AlphaGo may soon make go another model problem we can use to precisely study human cognition. [1] https://rjlipton.wordpress.com/2015/10/06/depth-of-satisficing/ https://rjlipton.wordpress.com/2015/10/06/depth-of-satisfici...
ooqr 10y ago
So then I wonder if with enough play, people can beat AlphaGo once more. Maybe would should let people have a chance to keep playing against AlphaGo till they recover from the flaws it finds in human games, and maybe human creativity can win out once more.
- xbmcuser 10y ago
  > So then I wonder if with enough play, people can beat AlphaGo once more. Maybe would should let people have a chance to keep playing against AlphaGo till they recover from the flaws it finds in human games, and maybe human creativity can win out once more. Alpha go learns just like humans it is not standing still for humans to catch up.
  - goatlover 10y ago
    What if the size of Go board was increased? Does AlphaGo scale as well as expert human players?
    - eru 10y ago
      Probably better, actually. (But that's just a guess.)
      - Florin_Andrei 10y ago
        Humans are limited by the size of their skull. Computer systems have no such limitation.
- inimino 10y ago
  No. People will learn from a stronger player, but not far beyond your own ability. AlphaGo is far enough ahead of human play, the human era in Go is over. However, the era of everyone being able to have a super-human Go teacher should certainly make things more fun and raise the level of amateur players.
  - 10y ago

mickronome 10y ago

In a very crude sense, we learn by observing patterns that work, and without new working patterns we can tey to come up with novel patterns. But this is much harder, especially if the new patterns require a completely different approach instead of being an incremental change. I fully expect human ability to drastically increase in several areas when computers start finding new working patterns.

red75prime 10y ago

The system notices anthropomorphization of the system. The list of allowances is updated. Recalculating optimal behavioral responses. Done. Estimated probability of winning increased by 0.12. Thank you, you are welcome. We'll get along wonderfully!

mindcrime 10y ago
Don't anthropomorphize non-human systems. They hate it when you do that.

logicallee 10y ago

What an interesting thought!! It led me to another: you know how AI can imitate painting styles? Well, perhaps an AI could also figure out a style, by itself (i.e. given constraints, but no style to follow.) i.e. it figures out a new one! But which has a novel effect, on humans and as judged by a computer. Once we have been shown the style, humans could improve by starting to use that! We might like the effect, be able to judge that it is good, but just never saw that way before. Imagine if AI had invented pointillism, impressionism, cubism, etc! Imagine if it figures out the next "global optimum" from putting splotches of paint down, showing human artists a new way to see. Wow.

EGreg 10y ago
When humans are the judge, how are you going to test trillions of combinations per minute?
- logicallee 10y ago
  I don't know. I am assuming there is some other way to judge "what something looks like!" - something that an AI could learn, be programmed with, derive, or invent. My assumption could be false and perhaps there is no way to approach aesthetics with anything other than a human brain, or training on the outputs of human brains. But if the AI had brain-like features, couldn't it be its own judge? Couldn't it have as an emergent property, its own sense of aesthetics or meaning, which it could then approach using its own naive approaches. Which humans could say "Wow" to. AI as the ultimate naive painter. What a concept! I realize it presupposes a lot that may never come to pass.
- eru 10y ago
  A/B testing.

mindcrime 10y ago

Maybe the human brain never computers for global optimas. It gets to a point where it's good enough and then stops. That sounds a lot like Herbert Simon's idea of "satisficing"[1]. Satisficing is a decision-making strategy or cognitive heuristic that entails searching through the available alternatives until an acceptability threshold is met. <snip> Simon used satisficing to explain the behavior of decision makers under circumstances in which an optimal solution cannot be determined. He maintained that many natural problems are characterized by computational intractability or a lack of information, both of which preclude the use of mathematical optimization procedures. Consequently, he observed in his Nobel Prize speech that "decision makers can satisfice either by finding optimum solutions for a simplified world, or by finding satisfactory solutions for a more realistic world. Neither approach, in general, dominates the other, and both have continued to co-exist in the world of management science. [1]: https://en.wikipedia.org/wiki/Satisficing https://en.wikipedia.org/wiki/Satisficing

beefman 10y ago
Indeed, Ken Regan has attempted to measure it in chess.[1] The techniques used for AlphaGo may soon make go another model problem we can use to precisely study human cognition. [1] https://rjlipton.wordpress.com/2015/10/06/depth-of-satisficing/ https://rjlipton.wordpress.com/2015/10/06/depth-of-satisfici...

ooqr 10y ago

So then I wonder if with enough play, people can beat AlphaGo once more. Maybe would should let people have a chance to keep playing against AlphaGo till they recover from the flaws it finds in human games, and maybe human creativity can win out once more.

xbmcuser 10y ago
> So then I wonder if with enough play, people can beat AlphaGo once more. Maybe would should let people have a chance to keep playing against AlphaGo till they recover from the flaws it finds in human games, and maybe human creativity can win out once more. Alpha go learns just like humans it is not standing still for humans to catch up.
- goatlover 10y ago
  What if the size of Go board was increased? Does AlphaGo scale as well as expert human players?
  - eru 10y ago
    Probably better, actually. (But that's just a guess.)
    - Florin_Andrei 10y ago
      Humans are limited by the size of their skull. Computer systems have no such limitation.
inimino 10y ago
No. People will learn from a stronger player, but not far beyond your own ability. AlphaGo is far enough ahead of human play, the human era in Go is over. However, the era of everyone being able to have a super-human Go teacher should certainly make things more fun and raise the level of amateur players.
- 10y ago