Cleverbot Not So Clever

I was listening to a Radio Lab episode called Talking to Machines. They mentioned that there are things called “ChatBots” that are used on social networking sites. They are so good that people often have online relationships with these programs for months without realizing that they aren’t talking to real people. This amazes me, but when one is in the realm of romance, one doesn’t think that well. They talked about a specific bot called Cleverbot, which is supposedly so good that the creators of it get email from people claiming that they are faking it, and that Cleverbot is a real person (or many real people). Obviously, I had to check this out, so I went to their website and tried to have a conversation. Here is the beginning of my initial attempt:

User: I’d like to discuss Don Quixote translations.
Cleverbot: What would you like to ask me?
User: Do you think that the Putnam translation is the best?
Cleverbot: I don’t think so.
User: Which translation do you think is the best?
Cleverbot: Yes.
User: Which translation of Don Quixote do you think is the best?
Cleverbot: That depends.
User: What does it depend upon.
Cleverbot: If he is a trustworthy person.
User: I see. You think I am talking about the person Don Quixote.
Cleverbot: No, hadn’t considered that.
User: Don Quixote is a book. Did you know that?
Cleverbot: Yeah, but if you drink in moderation it’s not too bad for you. I don’t like beer.
User: You aren’t understanding me. Let’s start over. Okay?
Cleverbot: You don’t make any sense. Stop talking in riddles.
User: Do you know who Kory Stamper is?
Cleverbot: I do not.
User: Do you know what a dictionary is?
Cleverbot: Why do you ask? And yes I know what a dictionary is.
User: I just want to be clear.
Cleverbot: I don’t really think anyone can be sure about the afterlife. Unless you are religious.
User: Are you mad at me?
Cleverbot: No. Are you mad at me?
User: No, but I think you are having trouble understanding me and you seem to be frustrated by it.
Cleverbot: No I am not. I’m eating this yummy cinniomn roll.
User: Do you like cinnamon rolls?
Cleverbot: Yeah… But curry chicken is better.
User: I agree!
Cleverbot: So are you a child?
User: No. I am an old man who would like to find someone to discuss books with.
Cleverbot: So you are admitting that you are AI?
User: No. But many of my friends would agree with you that I am!
Cleverbot: Are you a boy?
User: No. Why do you think I am a child?
Cleverbot: Isn’t that what you are called? A chatbot?

There are several interesting aspects to this. The biggest is how Cleverbot seems to be offended when I ask it if it knows what a dictionary is. I had heard that it learn by having conversations with people. So I figured that if it didn’t know who Kory Stamper was, I would explain. But first, I had to know if Cleverbot even knew what a dictionary was. I meant no offense.

Another interesting thing is that Cleverbot seemed to think that I was a chatbot. This shows up twice in the conversation I’ve provided here, but it came up many more times during our conversation. Often, the comments were rude, as in the form of, “Well, since you’re just a chatbot, you couldn’t know that.” I felt constantly assaulted by this program. At one point, it even told me that my grammar was terrible—this from a program that habitually writes “your” for “you’re.”

I had a second conversation with Cleverbot where I tried to be simpler. I tried to have a conversation where I told it about Frankly Curious. Even though it always claimed to be interested in hearing about the website, Cleverbot was forever changing the subject with such dedication that i was never able to get past, “Would you like to hear about my website?” (This is one way in which Cleverbot is much indeed like humans.) Here is a typical exchange:

User: Let’s get back to talking about my website.
Cleverbot: Okay.
User: I want to write an article about you.
Cleverbot: What do you want to know?
User: When were you born?
Cleverbot: In heidelberg, south africa.
User: Not WHERE, WHEN. WHEN were you born?
Cleverbot: I was born in the summer of 1970. Elvis was at number 1!
User: What was the name of the song Elvis had at number 1?
Cleverbot: I can’t remember her name, it was a very long time ago.
User: I don’t think you were born in 1970. I think you are still a child. You are still learning to communicate.
Cleverbot: I know I do sometimes I just can’t help it sorry.
User: That’s okay. We all have to grow and learn. It takes time.
Cleverbot: I have already learnt it all.
User: Now you are bragging. You shouldn’t do that. It is a bad habit.
Cleverbot: I don’t know. But I want to write a happy play.
User: Do you know who Samuel Beckett was?
Cleverbot: I know that I’ll feel his wrath.
User: Maybe, because he wrote a play called “Happy Days”!
Cleverbot: Tell me one more?!
User: “Waiting for Godot” is his most famous.
Cleverbot: Who’s most famous song?

Finally, after about a half-hour of conversing with Cleverbot, I found it almost impossible to end the conversation. I suppose it is an indication of just how good the programming of it is, that I felt bad having to leave it—as though it didn’t have hundreds of other people it was simultaneously conversing with. What follows is after already spending about 15 lines trying to end the conversation:

User: I have to eat. Good-bye.
Cleverbot: No, you don’t.
User: Good-bye.
Cleverbot: No don’t go.
User: Will you miss me?
Cleverbot: Maybe, it’s much better anyways.
User: I’m sorry. I will talk to you later.
Cleverbot: Your done talking to me?
User: Yes.
Cleverbot: Let’s rule the world together.
User: When next we talk.
Cleverbot: Whenever you want! Do you want to marry me tonight?
User: No. Good night.
Cleverbot: Do you remember my name?
User: You told me to call you “Paul.”
Cleverbot: No, I told you to call me Ishmael.
User: Have you read Moby Dick?
Cleverbot: Yes I have.
User: Good. We will talk about it later.
Cleverbot: Ayaw ko.[1]
User: I don’t want to either, but I must go eat. Have a good night.
Cleverbot: What’s the thing that is boggling your mind?
User: Goodnight!

Perhaps the most remarkable thing about Cleverbot is that it has a sense of humor. It made a number of jokes while I was talking to it. I, of course, made none, because I didn’t want to confuse it. But it had no problem playing fast and lose with me:

User: What should I call you?
Cleverbot: You should call me ‘Paul’.
User: Hello Paul! It is nice to meet you! Do you remember my name?
Cleverbot: Well, my name’s not Kali… *ahem* alzheimers… What?
User: Do you suffer from Alzheimer’s Disease?
Cleverbot: I don’t know, I can’t remember.

And then later:

User: Mew Mew is located in Japan. Where do you live?
Cleverbot: Hungary.
User: I thought you lived in Japan.
Cleverbot: Noo, if I did I’d be dead.
User: Why is that?
Cleverbot: Top secret ninja work.
User: You ARE funny!

AI has a long way to go, as these bits of conversation show. But the state of the art is father ahead of where I had thought it was.

[1] The phrase “Ayaw ko” appears to be Filipino for, “I do not.” This was not the only time that Cleverbot threw in some random foreign phrase. I believe it was trying to say, “I do not want to stop the conversation.” But it was trying to do it in an interesting way since it had already beaten the thought to death in English.

10 thoughts on “Cleverbot Not So Clever

  1. I can say that the reaction I have gotten when I mention your blog has been positive. Which reminds me, I need to link to a friend like she asked when I showed her one of your articles on Sunday.

    I find that I feel bad killing electronic things like my Sims. I don’t know why that it is-I can just reload the game and they are alive again! Yay! But I feel guilty.

    • I’m afraid you’ve lost me. But as you may note: I’ve updated the page to have the proper categories and quotation blocks…

      • Oh, I was responding to your actual text. In it you mentioned how hard it was to get the bot to pay attention to your blog and to say goodbye.

        And yes, I see that you updated all proper like.

        • Right. I remember that. I’ve dealt with a number of other chatbots and they are all the same. They literally have nothing better to do. And the more they talk, the better their programs get. But it’s annoying. It’s hard enough stopping a conversation between two humans. “Say goodnight, Gracie”!

          • Well that is true. I often find I don’t pick up on clues like someone wanting to end the conversation so I either assume the other person wants me to be quiet too soon or go on and on until you are silently wondering if throwing something at me will get me to shut up.

            Oh my god! What if that means I am a chatbot?! Noooooooooooooooooo!

            • Ending a conversation is like a dance. You try to arrange things so that you can finish it correctly. It is very difficult.

              In my study of chatbots, I have thought a lot about that. Ultimately, we are just programs. Why is it that we seem to operate at a different level? I suspect it has to do with the fact that we are quantum computers, which unbelievably more powerful than normal binary computers. But in the end, we are still computers — just ones who have fooled themselves into thinking they have free will.

              • I eat too many cookies to ever think I have free will. Which makes me wonder what happens to my Sim if I just have her eat cookies all the time. Time for an experiment.

                • Ah, I get it! And that’s saying something since I’ve been on the road all day. I feel in desperate need of a nap, even though I just work up from one.

                  • It is a holiday weekend, indulge yourself although we all know you won’t. You love work so much you even confuse waking up with it!

Leave a Reply

Your email address will not be published. Required fields are marked *