Poker as a Proxy Turing Test

I found the recent contest in which an algorithm was able to successfully defeat four professional poker players in a particular version of poker to be very interesting.

What strikes me is not the fact that the algorithm was successful but the way in which it accomplished the task.

check this out it’s all interesting but pay close attention from about the 8 minute mark

and also this

https://www.bloomberg.com/news/articles/2017-01-31/inside-the-20-year-quest-to-build-computers-that-play-poker

From what I gather the professional poker players gave overarching four reasons for their defeat

1) The unusual and difficult conditions of the contest
2) The ad hoc reset restriction that placed a limit on the amount that could be bet on any one hand (equity chop)
3) The fact that the AI “bluffed” randomly
4) The fact that the AI changed strategies at times randomly

You could say that the algorithm succeed in this particular challenge by behaving in a way that was completely unlike a human player.

Traditionally Poker has been as much about understanding the personal proclivities of your opponent (when he will bluff etc) as about skill in the peculiarities of the game.

Apparently this AI won by taking removing that all important personal aspect and turning Poker into just another math problem. It should not surprise that an algorithm is successful in solving a math problem.

When asked if they could employ the strategy of the AI to improve their own games the players said it would be impossible for them to act that way.

from the article

quote:

Les said he’s trying to figure out how to adapt some of Libratus’s irregular betting behavior to his own game. It’s hard. “We just simply do not have the mental capacity to do it,” he said.

and

there’s no clear path to turn Libratus and DeepStack into players that could be confident of beating a group of flawed humans. That’s because the equilibrium strategy that the AIs use fall apart in multiplayer games, when the point isn’t to play perfectly but to identify and exploit the shortcomings in other people’s games.

End quote:

I would suggest that with a little fleshing out least some of the reasons that the poker professionals gave for the AI victory could be used as a powerful proxy Turing test for determining if there is a mind behind a particular process that is of interest to us.

That sort of thing could be useful in evaluating Artificial intelligence as well the cause of things like the universe or the panorama of life. What do you think?

peace

31 thoughts on “Poker as a Proxy Turing Test

  1. Hang on. The computer didn’t learn about it’s opponents and develop strategies specific to each of them? Did it even know that it was playing different people?

  2. Mung: Did it even know that it was playing different people?

    It seems that it’s victory was a direct result of it not knowing any thing at all.
    At least as far as we can ascertain.

    peace

  3. Neil Rickert: I don’t see that as a Turing test, unless you meant a failed Turing test.

    Right, it was a failed test.
    Persons behave in a certain way and the AI won precisely because it acted differently.

    I find that fascinating because we can potentially universalize this insight into the behavior of persons verses algorithms.

    Peace

  4. I say its a misunderstanding about these things.
    Its all a function of memory. With humans they fail to maintain memory.
    so a perfect memory machine, on a curve, will beat the less perfect ones.
    there is no thinking going on. just memory operations.
    Poker is not a thinking mans game.

  5. The lesson from this particular “Turing test” is that when identifying a Mind we should expect to see among other things.

    1) limited random unjustified actions (ie bluffs)
    2) no sudden random changes in over all nature (ie strategy or personality)
    3) No extraordinarily mechanical tedium but rather something like stability punctuated at times by pauses.
    4) Continuance from one action to the next rather than disjunction

    Peace

  6. Robert Byers: there is no thinking going on. just memory operations.

    For the AI it was definitely the case that no thinking was going on. Just number crunching

    What tripped the humans up was that when it comes to poker they expect to be to competing against a cogent agent and they instead were competing against a calculator.

    Poker players win by sizing up their opponents mind. That could not happen here.

    News flash, don’t bet against a calculator in a math contest.

    peace

  7. That sort of thing could be useful in evaluating Artificial intelligence as well the cause of things like the universe or the panorama of life. What do you think?

    Utter bollocks.

  8. Tell you what FMM. Why don’t you go and actually use “this sort of thing” to do something useful.

    I would suggest that with a little fleshing out least some of the reasons that the poker professionals gave for the AI victory

    If it’s only a little get the fuck on and do it.

    The joke is with you people is you are always on the edge of a major discovery, as long as others are willing to do the work for you.

  9. OMagain: Tell you what FMM. Why don’t you go and actually use “this sort of thing” to do something useful.

    As you know I’m already using the lack of the behavior that the poker players saw to identify special cause tampering when I look for patterns in processes. That is in part that is what my “game” does.

    What’s needed is needed imo a consensus that the lack of these behavior is a way to identify mental agents . Hopefully that is what this discussion will be about

    OMagain: Utter bollocks.

    it’s that sort of unwillingness to universalize criteria like the Poker players identified when the subject is things like the panorama of life or the universe that makes it difficult to “do something useful” with their insight.

    Apparently everyone is willing to grant that an AI is not a mind because it exhibits these sorts of behavior but “skeptics” are unwilling to entertain the idea that the cause of things like the panorama of life is a mind despite it behaving in a way unlike algorithms.

    IMO What is needed it some sort of consensus that Minds behave in certain ways. If we could get that it would be firs step in “doing something useful” In both AI and ID

    peace

  10. Mung:
    I think what we’re seeing is that the chemicals in certain brains behave predictably.

    it is the dosage that is unpredictable

  11. Mung: I think what we’re seeing is that the chemicals in certain brains behave predictably.

    Sadly it looks like a case of not wanting to even think about what sorts of behaviors that Minds share for fear that to do so would commit them to affirming something they find to be unpleasant.

    They avoid doing this even though it’s obvious that such an exercise would be valuable to something increasingly relevant to modern life. (ie Artificial Intelligence )

    And they call ID a science stopper. 😉

    peace

  12. fifthmonarchyman: For the AI it was definitely the case that no thinking was going on. Just number crunching

    What tripped the humans up was that when it comes to poker they expect to be to competing against a cogent agent and they instead were competing against a calculator.

    Poker players win by sizing up their opponents mind. That could not happen here.

    News flash, don’t bet against a calculator in a math contest.

    peace

    Possibly. You probably know more about poker then me as i know nothing.
    I understand there is great deal of math as such.
    I’m not sure sizing each other up is the margin of victory.
    Its an option the humans just made mistakes. other human might not.
    The thing is that in all card/chess etc games its all about memory unless sizing up opponents does matter.

  13. Robert Byers: I’m not sure sizing each other up is the margin of victory.

    some relevant quotes

    quote:

    Play the players more than you play the cards.
    and
    Watch the other players for “tells” before you look at your own cards.
    and
    Choose the right opponents. If you don’t see a sucker at the table, you’re it.
    end quote:
    Amarillo Slim

    Robert Byers: Possibly. You probably know more about poker then me as i know nothing.
    I understand there is great deal of math as such.

    I’m not an expert but I’d agree, If you remove the personal aspect you are pretty much left with math. Computers are really good at math.

    Robert Byers: The thing is that in all card/chess etc games its all about memory unless sizing up opponents does matter.

    Robert Byers:
    The thing is that in all card/chess etc games its all about memory unless sizing up opponents does matter.

    Sizing up your opponents is what makes poker different from other games IMO.

    However now that you mention it the Chess program I play from time to time doesn’t behave much a person either.

    It does the same sort of things that the poker bot does.

    Perhaps there is a pattern here 😉

    peace

  14. fifthmonarchyman: For the AI it was definitely the case that no thinking was going on. Just number crunching

    I’ll just comment that many computationalists might say that thinking just is number crunching.

    Personally, I’m not a computationalist. And I agree with your assessment that there is no thinking. I’m just pointing out that not everyone will agree, though you probably know that.

  15. Neil Rickert: I’ll just comment that many computationalists might say that thinking just is number crunching.

    I hope that we can use the opportunity to define the boundaries of what “thinking” is.

    It’s obvious that what the algorithm is not behaving like a person.
    So if the computer is thinking then thinking is not what humans do.

    peace

  16. This discussion risks sinking into that same old morass of undefined terms. ‘Mind’, ‘Person’, ‘Thinking’ – unless you provide some definitions for these terms, at the very least operational ones, you cannot hope to have a rational, analytic discussion with conclusions that would stand up to scrutiny.

    Here is something you might want to do:

    We can tell that something has a Mind if 1)…, 2)… 3)….
    We can tell that something is a Person if 1)…, 2)… 3)…
    We can tell that something Thinks if 1)…, 2)…, 3)…

    If you do that, you will end up with a Venn diagram of two non-overlapping circles: one is the set of entities that are either Persons, and/or have a Mind, and/or can Think; and the other one is the set of everything else.

    At that point you might realise that the simplest way of labelling the first group is ‘Human Beings’.

    Basically, all you will end up doing with this Poker example is demonstrate that a computer is not a human being.

  17. faded_Glory: Here is something you might want to do:

    We can tell that something has a Mind if 1)…, 2)… 3)….
    We can tell that something is a Person if 1)…, 2)… 3)…
    We can tell that something Thinks if 1)…, 2)…, 3)…

    from up thread

    Quote:
    The lesson from this particular “Turing test” is that when identifying a Mind we should expect to see among other things.

    1) limited random unjustified actions (ie bluffs)
    2) no sudden random changes in over all nature (ie strategy or personality)
    3) No extraordinarily mechanical tedium but rather something like stability punctuated at times by pauses.
    4) Continuance from one action to the next rather than disjunction

    end quote:

    How is that for a first draft??

    faded_Glory: At that point you might realise that the simplest way of labelling the first group is ‘Human Beings’.

    Basically, all you will end up doing with this Poker example is demonstrate that a computer is not a human being

    I would agree with this as far as it goes.

    I think that when we say something is intelligent or has a mind what we really mean is that it in some measure it is like us.

    That is how it should be

    We are homo sapiens (Latin: “wise man”) after all. When it comes to Turing tests we are focusing on the “wise” part of the equation not the man part.

    I think that when we look for the Designer in ID or the mind in Strong AI what we are really looking for is something that thinks like us. Something we recognize as familiar but not identical to who we are.

    I don’t see anything wrong with that. Do you?

    If a proposed AI or Designer is radically “alien” then there would be no point of intellectual contact and no way to socially interact.

    IMO That is what happened in in the Poker contest and that is why the humans lost.

    peace

  18. fifthmonarchyman: I think that when we look for the Designer in ID or the mind in Strong AI what we are really looking for is something that thinks like us.

    If God thinks like you do, that would explain so much

  19. dazz: If God thinks like you do, that would explain so much

    If you don’t mind lets try and keep this sort of taunting out of this thread, I would hope that we could be civil and address the topic rather than personalities

    Peace

  20. Quote:
    The lesson from this particular “Turing test” is that when identifying a Mind we should expect to see among other things.

    1) limited random unjustified actions (ie bluffs)
    2) no sudden random changes in over all nature (ie strategy or personality)
    3) No extraordinarily mechanical tedium but rather something like stability punctuated at times by pauses.
    4) Continuance from one action to the next rather than disjunction

    How is that for a first draft??

    That is very post-hoc, and only tailored to the specifics of this particular poker example, and extremely vague and ambiguous at that. Try criteria that are not open to dispute between different observers. Not everybody will agree on what is ‘limited’ or ‘unjustified’ for example.

  21. One of the characteristics of humans playing games is that, over time, they adopt strategies that win.

    Which means there is no fixed winning strategy.

    Stock market strategies reflect the fact that when many players adopt a winning strategy, it ceases to be effective.

    The goal of any AI would not be to imitate a particular human being, but to have the capacity to learn.

  22. faded_Glory: That is very post-hoc, and only tailored to the specifics of this particular poker example, and extremely vague and ambiguous at that.

    Actually it’s an attempt to phrase criteria that I think we already use but in the context of this particular game. I did that on purpose.

    When I first watched the video I was struck by how these guys seem to independently discover behavior that I was already looking for when I tried to eliminate “non-mental” causes in data represented in line graphs.

    When I do that in addition to general randomness what I try and filter out

    1) large random spikes in the data
    2) sudden changes in the overall pattern of the data that appear to be random
    3) long periods of monotony
    4) unexplained disjunction in the pattern.

    I just took those 4 criteria and rephrased them to correspond to what the poker players were saying

    I have no idea if there is warrant for choosing these criteria or if it is an exhaustive list of “non intelligent” behavior.

    What I’m interested in is if these sort of things can be universalized as sort of an all purpose Turing test that can be used for the product of any process.

    It would be cool If it could

    I would love to hear any suggestions in that vein

    peace

  23. petrushka: One of the characteristics of humans playing games is that, over time, they adopt strategies that win.

    Which means there is no fixed winning strategy.

    I agree, but humans don’t generally randomly abandon winning strategies. The poker bot had at least one excellent strategy that the humans had never seen before and for some unknown reason it simply stopped using it at some point.

    What persons do usually is stick to what works and try and make minor improvements along the way if needed.

    I will grant that I’ve heard that in poker it’s sometimes a good idea to alter your strategy as your opponent begins to get a handle on it. In a way changing up your strategy is a single unified strategy in itself

    But from what I understand the humans were totally unable to understand why the bot made the radical change it did.

    The change seemed random and ad hoc just like the “bluffs” that the computer was making

    peace

  24. fifthhollowman,

    The strategies developed by the computational agents were not only unforeseen, but also unforeseeable, by the developers of the agents.

    When computational agents pass the Lovelace test, which fifthmarksman regards as supremely important, you beat a cowardly retreat to the Imitation Game.

  25. Tom English: When computational agents pass the Lovelace test, which fifthmarksman regards as supremely important, you beat a cowardly retreat to the Imitation Game.

    What?? I don’t think I’ve ever claimed that the Lovelace test was supremely important.

    I’m not even sure how you could demonstrate that a particular AI was actually being creative. It seems to me that folks could always claim that any thing new that it produced was not original but derived from inputs.

    Heck we even have difficulty with this sort of thing when it comes to human designers as witnessed by copyright lawsuits etc.

    I would be very interested if you have suggestions for making the Lovelace test more objective and universal.

    For instance do you think that the cause of life’s diversity passes the Lovelace test? If not why not?

    As far as the AI’s strategies being unforeseeable. A Random pattern is unforeseeable by definition and randomness is not something we expect from persons.

    peace

    PS

    I do respect your expertise but don’t you think that communication would be easier if you did not begin by calling me a coward.

Leave a Reply