3 ms·

The Turing-roulette Game

23 points by paraschopra 16y ago

jacquesm 16y ago
great idea, but this may be a fatal flaw: > you won’t try to be too-dumb (because you will be caught as an obvious deception) After all if being too dumb is a guaranteed deception then programming a bot to be too dumb would mean that bot takes the big prize.
- paraschopra 16y ago
  I don't know but probably even the "dumbness" between a bot and a human could be detected. I don't really know what will be actual dynamics of the interaction. If your bot starts making the same "dumb" conversations again and again, it will start being detected (because after the chat, a human gets feedback via points whether he was right or wrong).
  - jacquesm 16y ago
    Ah, but again we can reverse it, as a human I could do that on purpose too, and so on. Or am I misunderstanding you and are all the bots to be provided by the site owners? (I figured from your description that you can sign up with a bot or as a human). The Turing test is a brilliant device because it is so hard to create something that will be able to pass it without actually being intelligent.
    - paraschopra 16y ago
      You are correct, you can either provide or a bot or take part as a human. My argument is that genuine dumbness of chatbots may not be replicable by by humans. The whole logic of game is that the human who is able to emulate a bot wins, so if you are successfully being too dumb and go undetected as a human, you win! I think I figure your point, and yes the basic assumption is that all bots should be semi-intelligent. The whole game will become uninteresting if every just says "ABC.ABC.". I think we need a caveat or eligibility criteria to take care of that.
      - joe_the_user 16y ago
        Any game with such an eligibility criteria is pretty much unstable. Winning would come down to seeing how well you could skirt the criteria, since "dumb but not too dumb" is inherently fuzzy.
      - jacquesm 16y ago
        You got it :) Thanks, it sounds really interesting by the way, I hope that you follow through on it, I'll definitely participate (with a bot, that's the fun bit I think).
        paraschopra 16y ago
        Yup, will definitely code a prototype and link up some open source chatbots. On the other hand, I think being a human participant will be the fun bit :)
GiraffeNecktie 16y ago
Somehow my brain read this title as "Turing Tourette" which also seems like it might be an interesting mashup.
- jacquesm 16y ago
  I think that's the default setting for many people, it would make distinguishing them from bots trivial.
todayiamme 16y ago
To win the game I think the most interesting deception I could engage in is to engage in fixed patterns that can be readily discerned by the entity while making sure that I don't display world knowledge readily. For example, we could be having a chat and on purpose I would make sure that I start to loop the same argument in different words again and again for different stimuli. I will make the deception look like I am trying to look smart while being painfully limited. I would also make sure that I always bring up less than 10 topics up for conversation so that it sounds like I am stuck in a loop. On the other hand I'll make sure that I appear to know nothing at all about pop culture while pretending to shift the conversation. If I were a bot creator though I would make sure that I had as complex interactions as possible. I would use limited code base so I sound like a human pretending to be a bot, but then I would attach emotional weights and allow random, unpredictable behavior to come through. So, that feeling of complexity is felt by the entity.

23 points by paraschopra 16y ago

great idea, but this may be a fatal flaw: > you won’t try to be too-dumb (because you will be caught as an obvious deception) After all if being too dumb is a guaranteed deception then programming a bot to be too dumb would mean that bot takes the big prize.

paraschopra 16y ago
I don't know but probably even the "dumbness" between a bot and a human could be detected. I don't really know what will be actual dynamics of the interaction. If your bot starts making the same "dumb" conversations again and again, it will start being detected (because after the chat, a human gets feedback via points whether he was right or wrong).
- jacquesm 16y ago
  Ah, but again we can reverse it, as a human I could do that on purpose too, and so on. Or am I misunderstanding you and are all the bots to be provided by the site owners? (I figured from your description that you can sign up with a bot or as a human). The Turing test is a brilliant device because it is so hard to create something that will be able to pass it without actually being intelligent.
  - paraschopra 16y ago
    You are correct, you can either provide or a bot or take part as a human. My argument is that genuine dumbness of chatbots may not be replicable by by humans. The whole logic of game is that the human who is able to emulate a bot wins, so if you are successfully being too dumb and go undetected as a human, you win! I think I figure your point, and yes the basic assumption is that all bots should be semi-intelligent. The whole game will become uninteresting if every just says "ABC.ABC.". I think we need a caveat or eligibility criteria to take care of that.
    - joe_the_user 16y ago
      Any game with such an eligibility criteria is pretty much unstable. Winning would come down to seeing how well you could skirt the criteria, since "dumb but not too dumb" is inherently fuzzy.
    - jacquesm 16y ago
      You got it :) Thanks, it sounds really interesting by the way, I hope that you follow through on it, I'll definitely participate (with a bot, that's the fun bit I think).
      - paraschopra 16y ago
        Yup, will definitely code a prototype and link up some open source chatbots. On the other hand, I think being a human participant will be the fun bit :)

GiraffeNecktie 16y ago

Somehow my brain read this title as "Turing Tourette" which also seems like it might be an interesting mashup.

jacquesm 16y ago
I think that's the default setting for many people, it would make distinguishing them from bots trivial.

todayiamme 16y ago

To win the game I think the most interesting deception I could engage in is to engage in fixed patterns that can be readily discerned by the entity while making sure that I don't display world knowledge readily. For example, we could be having a chat and on purpose I would make sure that I start to loop the same argument in different words again and again for different stimuli. I will make the deception look like I am trying to look smart while being painfully limited. I would also make sure that I always bring up less than 10 topics up for conversation so that it sounds like I am stuck in a loop. On the other hand I'll make sure that I appear to know nothing at all about pop culture while pretending to shift the conversation. If I were a bot creator though I would make sure that I had as complex interactions as possible. I would use limited code base so I sound like a human pretending to be a bot, but then I would attach emotional weights and allow random, unpredictable behavior to come through. So, that feeling of complexity is felt by the entity.