Thursday, July 31, 2008

Psychological Proofs and Reverse CAPTCHAs

Guest Post from Amir Michail.

The goal behind a psychological proof is simply to convince most people of something. And that's all that's required. A psychological proof may be complete nonsense but as long as it convinces most people—namely, those of average intelligence—then it can be considered a success.

Such proofs can be useful on the web, particularly when a proof that would be convincing to intelligent experts is impossible. Moreover, such proofs need not be deceptive (e.g., as with phishing). You could have a nonsensical/weak psychological proof to convince most people of something that is actually true anyway, so no harm done.

As an application of psychological proofs on the web, consider "reverse CAPTCHAs" within the context of chatbots. It is important to give people chatting with chatbots some confidence that *all* the bot replies are really bot replies—and not something typed in by a human watching the chat.

The problem is to somehow convince most people that you are really a chatbot and not a human. The method used should give them more confidence than simply telling them so.

While chatting with the bot from chatbotgame.com, you can get this confidence by clicking "Convince me you're a bot!". The idea is to show you the rule/method that was used to generate each bot response. Note that you can see rule usage in other chats by clicking accepted/rejected. This gives you more confidence that a rule was submitted prior to seeing its bot response in your chat.

I would be interested in knowing about other uses of psychological proofs on the web to convince most people of something that is true anyway.

7 comments:

  1. Thank you for introducing me to that time-waster. The Digg-style chatbot idea actually works surprisingly well. The conversation was almost believable, if vapid.

    ReplyDelete
  2. I think I'm missing the point here: why do you want to convince someone they are talking to a bot rather than a human?!

    ReplyDelete
  3. I think I'm missing the point here: why do you want to convince someone they are talking to a bot rather than a human?!

    People may talk to a chatbot out of curiosity to see how good its AI is. They may also say things to a chatbot that they would not say to a human (even if the chat is not private).

    ReplyDelete
  4. The rules seemed pretty simplistic. A chatbot with truly good AI would not be able to explain its utterances in such a succinct fashion. Or alternatively, the explanations would be just as "chatty" as the main conversation.

    ReplyDelete
  5. The rules seemed pretty simplistic. A chatbot with truly good AI would not be able to explain its utterances in such a succinct fashion. Or alternatively, the explanations would be just as "chatty" as the main conversation.

    We don't have any chatbots with truly good AI yet. Also, you might be able to fake convincing conversations using massive knowledge and simple algorithms.

    In any case, consider another example: targeted advertising in gmail. How would you convince users that the advertising is targeted in an automated way?

    ReplyDelete
  6. The rules seemed pretty simplistic. A chatbot with truly good AI would not be able to explain its utterances in such a succinct fashion. Or alternatively, the explanations would be just as "chatty" as the main conversation.

    Keep in mind that the psychological proof can be nonsense as long as it convinces most people of average intelligence. So you can try to fake a succinct one.

    ReplyDelete
  7. The author of the bot is not claiming to give a prove. He just wants to rise the confidence level.

    "as long as it convinces most people—namely, those of average intelligence"

    I think that intelligence is only a small factor. It is normal not to question everything all the time. Educated people learn more false ideas than uneducated people, so at the end of the day the educated people believe in more false ideas.

    "You could have a nonsensical/weak psychological proof to convince most people of something that is actually true anyway, so no harm done."

    But is it really true? If you can't proof it, then quite often it is false. This is my experience in computer science. In politics the politician and journalists don't care if something is true, as long as it is helpful to them. And there is harm done.

    ReplyDelete