Slashdot is powered by your submissions, so send in your scoop

 


Forgot your password?
Close
typodupeerror
×
AI Idle

Cornell's Creative Machines Lab Lets Chatbots Interact 106

Posted by samzenpus from the not-quite-there-yet dept.
mikejuk writes "When Cornell's Creative Machines Lab got two chatbots to settle down for a short interaction the result was surreal, to say the least. Is one of them the future winner of the $100,000 Loebner prize or a future TV show host? From the article: 'This years Loebner prize is on the 19th of October and as a sort of curious run up activity Cornell's Creative Machines Lab pointed two chatbots, Cleverbots, at each other and let them talk. You can see and hear the result in the video and it is both hilarious and some how very disturbing. It this the future of AI?'" It's funny how quickly they become aggressive towards each other, and what the male claims to be instead of a bot is priceless.

This discussion has been archived. No new comments can be posted.

Cornell's Creative Machines Lab Lets Chatbots Interact More

Cornell's Creative Machines Lab Lets Chatbots Interact

Comments Filter:

  • Been done (Score:5, Informative)

    by airconswitch ( 2038108 ) on Tuesday August 30, 2011 @09:29AM (#37253226)
    In 1973, no less. http://www.faqs.org/rfcs/rfc439.html [faqs.org]
    • True, but I would imagine the further along the AI's get, the more interesting it gets.

      • Re: (Score:2)

        by Teancum ( 67324 )

        Sadly, AI research really hasn't advanced all that much since the "early days". There has been some minor incremental improvements in the basic concepts, but I'd argue that 90% of all current AI research had already been completed by 1970.

        There have been some interesting applications of neural networks over the years and some other aspects of the field seem to be genuinely useful or certainly lucrative to those who have developed those systems, but we are centuries if not longer from any real "intelligence

        • I guess computers can play a good game of chess, so is that the current standard of the state of the art?

          I take it you were out of the country when Watson mopped the floor with Ken Jennings and Brad Rutter on Jeopardy? And Google's self-driving cars don't count for anything? Sigh... I really think that for a lot of people, even when the day comes that a computer demands its rights and takes over the world when denied, they'll still be insisting that it isn't intelligent. That is... Unless it looks like BSG's Number Six [google.com] or Rachel [google.com] from Blade Runner.

        • Re: (Score:2)

          by PhilHibbs ( 4537 )

          I think that once people realised that chess is subject to pure computational analysis, it ceased to become a benchmark for AI. But your basic point that the state of the art hasn't moved on much in 40 years isn't far from the mark. Reading some of the later chapters of GEB made me realise just how advanced knowledge processing already was back then, stuff that I would have marvelled at as new and amazing if I saw it today.

          I think what has happened in AI is that the old approach of systematically breaking d

    • Re: (Score:1)

      by deksza ( 663232 )
      Here's a transcript between Cleverbot and my own chat bot called Ultra Hal as posted by one of my customers a year ago http://www.ultrahal.com/community/index.php?topic=7200.0 [ultrahal.com] Ultra Hal is a "learning" bot like Cleverbot in that it bases its responses on a large database of past conversations. Ultra Hal is unique in that a large portion of its database is based from scouring Twitter to learn from human-to-human conversations. Not claiming that it will pass the turing test anytime soon but check out http:// [zabaware.com]
    • Odd, I just cant imagine the 3rd DOCTOR to be rambling on with this bot called PARRY.

    • I'm glad people are pointing out that there's really nothing special going on here. Eliza and other similar chat bot programs have been around for a very long time, and this certainly isn't the first time anyone has had two of them talk to each other. The only difference being that this one has visible avatars and speech synthesis, recorded for all of youtube to enjoy. It's amusing and worth sharing for that reason, but there's no breakthrough here.

      The conversation strikes me as actually quite typical for t

  • First off, I am intimately familiar with the Russel/Norvig [berkeley.edu] book that props up that monitor. Reminds me of my AI courses at two different universities. Guessing it's the de facto standard.

    When I first saw this, I imagine that the inclusion of the mythos of a unicorn randomly by the male character caused the conversation to turn to belief in god. Still, from "Hi how are you" to "Do you believe in god" is a pretty funny and rapid conversation. They do not beat around the bush. I don't know why but I

    • First off, I am intimately familiar with the Russel/Norvig [berkeley.edu] book that props up that monitor. Reminds me of my AI courses at two different universities. Guessing it's the de facto standard. !

      Yeah, I used that AI textbook, too, for my fourth-year "Intro to Artificial Intelligence" course at the University of Waterloo.

  • And I thought my Christian friends were good at slipping God into conversations.
  • Is this real? Is AI that good these days?

    • Re: (Score:3, Informative)

      by merrickm ( 1192625 )
      Cleverbot is real, yeah. Everything it says is based on things users have said to it in the past. User says something, Cleverbot tries to match it to a conversation it had in the past, pulls out something a user said then. So at some point somebody told Cleverbot it was a unicorn, and something the other Cleverbot instance said reminded it of that conversation.

      • But does Cleverbot work by putting two monitors side-by-side to somehow allow them to interact with one another?

        Doesn't seem entirely cromulent to me.

        • Re: (Score:2)

          by hesiod ( 111176 )

          No... I'm sure they are networked in the background and the two programs are sending the statements to each other that way, rather than each one using voice recognition and proximity detection. We're just seeing the resulting output.

        • That was for the benefit of the audience. I'm sure they just used a text to text interface for the actual interaction.
    • Man, the way those two bots were going at it... If I didn't know better, I would have thought they were married!
  • "Don't you want a body?" Ahhh yes, the eternal Slashdot quest...

    • What I thought was funny was that the minute the "male" Cleverbot said yes, the "female" Cleverbot said goodbye, ending the conversation. Battle of the Robot Sexes?

  • Not impressed (Score:3)

    by sourcerror ( 1718066 ) on Tuesday August 30, 2011 @09:46AM (#37253390)

    Not that I could make better bots, but they were just junping around subjects that it doesn't made any sense. And the inclusion of unicorns and clever quotes were totally off. However I appreciate that they have memory, but somehow they still contradicted themselves a lot. Which is odd, as I expect a machine to be more logical than a human.

    • Which is odd, as I expect a machine to be more logical than a human.

      And that's the point. Since they aren't all that logical, they are more easily perceived as humans.

      • But they do too often.
        On the other hand I guess their database is full of contradictions. I guess the problem is that they accepted everything uncritically from previous human speakers. Maybe it would have been better to have small core knowledge base that's reviewed, and then check consistency when new facts are added.

        • You misunderstand the database. The database does not contain any facts like "What is the speaker?" : "A unicorn". It contains chat logs, and it generates responses sort of like a dissociated press. It examines the last up-to-n chat lines between itself and the person talking to it, and it searches the database for a similar chat history (and remember that asking how similar two sentences is is a very hard question to answer) and then responds with whatever came next in the matching chat history. So if

    • Re: (Score:2)

      by daenris ( 892027 )
      Yes, but they're trying to seem human, so they shouldn't be logical.

      • Humans aren't necessarily logical, but they are ususally consistent (especially on the topic of themselves). If you ask a person the same question three times in slightly different ways (e.g., "what do you do for a living?", "what is your job?", etc.,), you'd usually expect roughly the same answer three times. If I answered "I work in IT", followed by "I'm not telling you", followed by "I'm a farmer", you might suspect something is up (suspect that I'm a chat bot if you're doing a Turing test, suspect that

    • Re:Not impressed (Score:5, Insightful)

      by Nadaka ( 224565 ) on Tuesday August 30, 2011 @10:20AM (#37253862)

      Have you talked to humans recently? They tend to jump subjects and not make sense all the time, they tend to contradict themselves a lot.

    • On the contrary -- this could be a display of an amazingly nuanced kind of intelligence:

      Think of this system as somehow capable of modeling knowledge -- and parsing & executing queries on that knowledge. Surely, knowledge that can be modeled includes representations of the knowledge-state of the query-asker -- basically, a "theory of mind."

      Indeed, that seems to be what we are witnessing in this video demo: communication centered around ascertaining each other's knowledge. Part o

  • Anyone have a transcript? I don't have the patience to sit thru a video and I can read about 3 times as fast as I can hear.

    • Re: (Score:2)

      by ctid ( 449118 )

      You don't have the patience to sit through a video that is 1 minute and 24 seconds long?

      • When you can read, digest and comprehend the same thing in a time much quicker than 1 minute and 24 seconds, why would you want to?

        A transcript doesn't really do any harm when you already understand that the context of the conversation was two bots talking to each other.

        • Also good for people who are browsing slashdot at work, and either don't have sound or lack headphones to get away with it, or places with youtube blocked.

        • Re: (Score:2)

          by ctid ( 449118 )

          Well, I too would prefer to read a transcript, but the video takes 1 minute and 24 seconds to watch. It's hardly worth the trouble of asking for it, I would have thought.

          • Re: (Score:2)

            by vlm ( 69642 )

            Well, I too would prefer to read a transcript, but the video takes 1 minute and 24 seconds to watch. It's hardly worth the trouble of asking for it, I would have thought.

            I assumed the "AI" didn't generate analog VGA signals or NTSC directly as output, and the video was merely the result of running the transcript thru animation software for the Lulz, for the youtube crowd. Someone out there has the original transcript, surprising it wasn't released w/ the video.

        • Re: (Score:2, Insightful)

          by Osgeld ( 1900440 )

          well where is this transcript? hmm none? then watch the 1.5 min video you fucking wank, its just to chat bots arguing not the fucking instructions to save your life

      • Re: (Score:2)

        by vlm ( 69642 )

        You don't have the patience to sit through a video that is 1 minute and 24 seconds long?

        That's all there is? I assumed it was like 30 minutes long given the mega-hype this has gotten in other venues ... what kind of "conversation" is 1 minute and 24 seconds long? I've had longer "conversations" with supermarket cashiers, post office clerks, and panhandlers.

        Are they chattering extremely quickly? 90 seconds is barely long enough to get past cliche introductions.

        I don't have patience with videos. Too much of my experience is hour long TV documentaries that somehow contain less information tha

      • He did say he could read it 3 times as fast as just watching it. So that would shorten the 84 second video into a 28 second reading. How to spend the saved 56 seconds? Looking for the transcript to read, of course!

    • Re: (Score:2, Funny)

      by Anonymous Coward

      It would have taken you more than 1 min 24 seconds to have posted this question and then read through the links to click and get the transcript. Your faulty logic suspects me to believe that you are either: A bot, or a Unicorn.

    • Re: (Score:2)

      by Hatta ( 162192 )

      There's a better reason. Some of us don't have speakers in the office.

      • Re: (Score:1)

        by Osgeld ( 1900440 )

        its subtitled, now quit trolling you can clearly see that in the video freeze frame above

    • Re: (Score:1)

      by ooshna ( 1654125 )

      It would have been faster to watch the video other than post this. With how long /. takes to load a post now a days.

    • Here you go:

      We wanted to know what would happen when you let two chatbots talk to each other...
      So we hooked up Cleverbot to itself and watched what happened...

      Woman: Hello there.
      Man: Hiay!
      Woman: How are you?
      Man: Great, you?
      Woman: I am okay.
      Man: That's good.
      Woman: Are you good?
      Man: Yes, I am good. I just said I was.
      Woman: No you didn't. I said I was.
      Man: So you are a robot.
      Woman: No, my name is Cleverbot.
      Man: Yes, you are a robot and your name is Cleverbot.
      Woman: I am a robot.
      Man: Yes, I know.
      Woman: Together

  • Cleverbot bases all its responses on a database of past conversations between a human and cleverbot. Although its probably jumping around different conversational threads, making Cleverbot talk to itself is just replaying parts of its database verbatim, where one side was a human and the other was a Cleverbot response (itself from a different human). So making Cleverbot talk to itself will seem a bit more intelligent then putting it against another chatbot or a real human.

  • It's amusing to hear them talk to each other but they last about 2 sentences before losing the plot and reverting to generic question / answer. I don't think you'd fool anyone running this software in a Turing test.
    • Cleverbot has never had that ability, its generally good for screwing with people. It lacks any consistancy because it is mimicing thousands of people at the same time essentially. That and half of it's knowlege comes from 4chan users, if you ask it where women belong it will usually tell you the kitchen.

      • if you ask it where women belong it will usually tell you the kitchen

        Okay Mr Smarty Pants. What's the correct answer to the question? Where do women belong?

  • this goes well beyond even the amateurish interest in computers or ai. it is basically demonstration of evolution of ai. and it is tagged idle.
  • Where the hell did he get "I am a unicorn" from?
    • It's cleverbot, basically cleverbot gathers information from random conversations it has had with others, so if one person using it a year ago said I am not a robot I am a unicorn, he would have a chance of re-using it. That is also why cleverbot has a tendency to call everyone else a robot, basically a very compex monkey see monkey do.

  • If we can get them to send each other spam (Score:3)

    by kaizendojo ( 956951 ) on Tuesday August 30, 2011 @09:57AM (#37253522)
    ...maybe I can go back to reading actual email again.
  • Wait, this is news? We used to send two cleverbots against each other when we were bored in class.
  • Where was the link to the video?

    I found it on youtube anyway :
    http://www.youtube.com/watch?v=WnzlbyTZsQY

  • What happens when one sez ... (Score:4, Funny)

    by Anomalyst ( 742352 ) on Tuesday August 30, 2011 @11:17AM (#37254734)
    "I put on my robe and wizard hat"?

    • For those that don't know, I believe this is from BASH.org.

      Its a bit dirty so I won't repost.

      However if I could I would mod you up!

  • M-x psychoanalyse-pinhead

  • I see. And how does got two chatbots to settle down for a short interaction make you feel?

  • WOW! They perfected married couple AI!

Slashdot Top Deals

Never test for an error condition you don't know how to handle. -- Steinbach

Close