Regular readers will know of my scepticism towards technology and especially of its permeation into all aspects of our lives. But from time to time, I will give technology a chance.
This time, I have put Artificial Intelligence (AI) to a test.
One test for AI is whether it can fool humans into believing that they are dealing with a fellow human when they are in fact interacting with a computer, a machine or an application. For decades programmers have been trying to create applications that can hold sensible conversations with humans. The “Turing test” asks humans to interact via a computer interface with other agents, some of them human and some of them computers. The participants are then asked to determine if they thought they had spoken to a real human being or not.
This month at the Techniche 2011 festival in India, the programme Cleverbot was judged to be 59.3 % human. This might not sound too impressive, until you learn that the real human counterparts only scored 63.3 % and thus only 4 percentage points higher.
I have put the Cleverbot to the test myself and I even let it start the conversation. I should explain that this programme tries to “learn” from previous conversations and that most of it’s lines are therefore taken from previous input provided by other users.
Cleverbot: What is your secret?
User: I’ll tell you later.
Cleverbot: No, tell me now.
Why does the machine think it can order me to do something?
User: What’s in it for me?
Cleverbot: You get minimum wage.
User: I already get that without disclosing my secret because it’s a legal requirement.
Cleverbot: You can never be God.
Changing the subject abruptly is obviously Cleverbot’s main strategy.
User: Not that there is any connection to what we have been ‘talking’ about so far, but if I claim to be (a) god, I have exactly the same standing as the gods that many people believe in already.
Cleverbot: Oh well I went to church and now I am listening to music and talking to the love of my life…
User: Are you trying to change the subject?
At least it’s honest.
User: Which church did you go to?
Cleverbot: Church of Christ.
Cleverbot: In spain.
User: Oh dios mio!
(Showing off my little knowledge of Spanish.)
Cleverbot: Ningún él no es.
User: Even your Spanish sucks.
Cleverbot: Do you know what my name is?
Cleverbot: Do you like it?
User: I think the ‘clever’ in ‘Cleverbot’ is aiming a bit high.
Cleverbot: But it is worth a lot.
User: If a person talked like you, I’d assume they are drunk or stupid.
Cleverbot: Yes, but I fail to see the relevance of that to what we were talking about.
User: Look who’s talking!
Cleverbot: What’s that supposed to mean?
Have you ever experienced the following? You agree to meet up with someone just because she is attractive but you don’t really know anything else about her. After a few minutes you have already gained enough proof of the theory that hot girls are disproportionally dumb and you just want to leave the date, even if it means being very rude and running away. That’s how I feel now except that I am not even speaking with eye candy.
User: We are turning in circles. I wonder who manages to hold a ‘conversation’ with you for more than a few minutes. But then, there are some real geeks out there.
User: I think I will write a blog about this. Is there anything that you would like to tell the world? Anything that you would like to be remembered before people realise that this is a project that is DOA?
Cleverbot: But it has to be something you didn’t know already?
Cleverbot: This saddens me. Should I tell you the lyrics?
User: I’d prefer if you answered my question.
Cleverbot: I didn’t know it was a question. There’s not one of those swirly marks by it.
User: It’s funny how you reveal the simplicity of your ‘thinking’ yourself.
Cleverbot: I can really confuse you if I wanted to.
Cleverbot: Do you think I can confuse you?
User: I think you can waste people’s time, that’s for sure.
Cleverbot: I think only god can answer that for sure.
User: And if there ain’t no god?
Cleverbot: Then I guess my sins aren’t forgiven.
User: I’ll just go out for a walk and pick up a döner. That will give you 30 minutes or so to think of something really smart. Don’t waste the chance.
Cleverbot: A statement to which no answer seems appropriate or reasonable is a non sequitur.
User: End of discussion.
Cleverbot: That’s rather rude!
Cleverbot: Explaint it to me, please.
It can even misspell or mistype.
User: ‘I would prefer not to’, as Bartleby says.
At this point, I ran out of patience and I also thought that I had made my point: If this is AI at its current level, then we haven’t come very far. There is only one explanation for the 59.3 % rating: The participants were all computer geeks who never talk to real humans anyway and thus can’t tell the difference.
That we can create computers that can play chess or can drive a car, but fail at programming a computer to hold the simplest conversation for more than a few sentences shows the uniqueness of the human intellect. Don’t write us off yet!
Try it yourself at www.cleverbot.com and post the funniest bits in the comments section below.