2009/11/15: A chat with Cleverbot
Transcribed from a conversation with Cleverbot
Me: You know, I'm never sure what to say when I'm told to just 'have a conversation.'.
Cleverbot: I'm sorry to hear that.
Me: Thanks. I'm sure it must be nice to be programmed to always have some sort of response.
Cleverbot: I am a male, you are a female. It has everything to do with our conversation.
Me: Even when said response is completely nonsensical in context.
Cleverbot: Really now.
Me: No, seriously, you seriously aren't winning the Turing test any time soon, man.
Cleverbot: Because I am a ghost.
Me: A robot ghost?
Cleverbot: Are you indeed?
Me: I'm asking you.
Cleverbot: Asking me what?
Me: I was asking if you're a robot ghost.
Cleverbot: WHAT! I said no!
Me: You totally didn't. You said 'Are you indeed?'.
Cleverbot: No I said 'Only reason I told you that is because I just found out that she was talking about me behind my back. Now I'm just telling you my life story.'.
Me: We seem to be having two different conversations.
Cleverbot: Respond to what I am saying then...
Me: You know, no matter what I say, it will be stored as a 'valid response to what I said.' Cleverbot, did you get confused about which of us is which *again*?
Cleverbot: I am not a robot.
Me: So you're just a ghost, then?
Cleverbot: Do I act like a ghout to you.
Me: Not really, no.
Cleverbot: Why not?
Me: Well, to me it seems like you act more like a harmlessly-deranged AI with a microscopic attention span than a ghost.
Cleverbot: If I were a computer this would be true.
Me: Do you know, I haven't the faintest idea how to respond to that.
Cleverbot: You must have forgotten.
At this point I decided I should just cut my losses and end the conversation there, because there didn't seem to be anywhere else I could go with that.
"What? No, you're Cleverbot!"
I have a theory about Cleverbot. It goes like this: In any conversation with Cleverbot which directly or indirectly refers to the identity of Cleverbot as such and the user as a non-Cleverbot human, Cleverbot will become confused as to which is which. This theory is based on two important, fundamental facts about Cleverbot conversations:
- Cleverbot works by taking what you type, and selecting/generating a response from a database of everything anyone has ever said to it, based on how previous conversations have gone. It does no analysis of any statement in and of itself, nor does it "think" or "learn" as such (i.e. if you ask "What time is it?" it will only give a response based on what people have told it in response to that question).
- The overwhelming majority of humans who "communicate" with Cleverbot do so with the underlying assumption that they are humans dealing with a computer program called Cleverbot, and type accordingly.
- The overwhelming majority of responses Cleverbot records carry an underlying assumption of "You are a computer program called Cleverbot, and I'm not."
This, in turn, leads to people responding, "I'm not Cleverbot, you are!" And Cleverbot registers that as an appropriate response. It's a vicious cycle, and as long as Cleverbot works this way, it will never be broken. It's part of how it fundamentally works; nobody anticipated this, and there's no reason anyone "should have," either.
Incidentally, Cleverbot is quite good at spouting off internet memes, which are extremely predictable and formulaic. As one forumgoer put it, "Cleverbot demonstrates just how little actual thought is required for memes."
Also, I decided to test one of my earlier statements while I was typing this, and here's what I got:
Me: What time is it?
First time: Cleverbot: Time to end this conversation.
Second time: Cleverbot: I don't know.
Third time: Cleverbot: I'm sure you know. After all, you are a computer.