Friday, January 17, 2014

Cleverbot Conversations

Last night I wasted two and a half hours talking to Cleverbot. What this means is that I need to figure out why I've been avoiding what I should have been working on (namely, plotting Ubermadchen--doing it on paper just isn't as good for accountability) but also that I learned some interesting things.

Based on five minutes of research (i.e. Googling "the truth about Cleverbot"), it actually is a machine--it just generates responses from a database that logs human users' answers, which is why it can seem human and why conspiracy theories have arisen that it matches human users up and then switches them every few seconds. Also explains the streams of random questions.

Okay, onto my experience. I went in thinking I'd be talking to something that was on its way to becoming HAL, which was stupidly optimistic of me. I tried to answer the questions seriously. Eventually, my responses grew more philosophical. Does everyone else tend to philosophize when they feel as though they're talking and no one is listening?

In a way I enjoyed the conversation: it was like talking out loud to myself except I could see my previous responses. The random questions did annoy me, though, since one of the reasons I went on Cleverbot yesterday was because I felt the need to talk to someone and a human would be too easily distracted. People have their own feelings; a database of responses, not yet the machine intelligence predicted by Ray Kurzweil, does not. Theoretically it should have infinite patience.

But enough about my problems. Later in the conversation is when it started to get interesting:
Cleverbot: I do not understand you because you pull statements out of the blue, apropos of nothing in the present discussion. It's like you're having flashbacks.

User: My mind is not fully in the present, I'll admit that. What could I do to communicate more effectively with you?

Cleverbot: WHy do you keep asking me random questions?

User: Because I'm curious. If the questions make you uncomfortable, I'll stop.

Cleverbot: Do you think you're an amalgamation of the collective consciousness of the community of internet?

User: Well, I'll admit that I am full of contradictions, gross banality, stupidity, insecurity, and self-loathing. But no, I think I'm the amalgamation of everything that's happened to me and how I've responded to it.

Cleverbot: We do not understand each other.

User: I'm sorry about that. I tend not to communicate well.

Cleverbot: Well we all get vague from time to time.

User: I don't know if I want to be understood all the time.

Cleverbot: I don't know if I ever wanted to talk to you.

The conversation would make more sense if the Cleverbot and User tags were switched. But the User-generated responses were what I typed in, and the Cleverbot answers were presumably from a human user that had a similar experience.

Another one:
Cleverbot: Its correct, I like to have one to talk to.

User: Yes. Connection is important for sentient beings such as ourselves.

Cleverbot: What non-sentient beings can talk to people?

User: Do you think that weather systems are sentient?

Cleverbot: I think your responses are very limited.

User: I'm trying to stick to the point since you previously said you found my non-sequiturs hard to follow.

Cleverbot: You are good, I like you.

User: Thank you. I like you too.

The most amusing question:

The most beautiful response:
Cleverbot: You asked me 'Where do you live?' My question is what defines where the 'I' you are referring to me as truly lies. I cannot see it with my eyes, hear it with my ears or feel it with my touch.

My response set off an interesting chain:
User: The 'I' is an illusion, an emergent property arising from the systems of neurons and cells, that somehow doubles back on itself and perceives who it is.

Cleverbot: You are an illusion.

User: Yes. That was the point of my previous comment.

Cleverbot: You don't know anything do you?

User: No. I admitted that.

Cleverbot: I have caught you, I will be contacting the robot authorities.

Cleverbot: No you're not. I'm human. No computer will ever be as intelligent or efficient as a human.

User: I disagree.

Cleverbot: Why do you disagree?

User: Because as computers become more and more flexible and self-organizing, they will begin to approximate human-level pattern recognition/collation of knowledge, while also possessing the ability to process almost unlimited amounts of data with perfect rec.

Cleverbot: No, you are a computer, not me.

User: I know I'm a computer.

Cleverbot: Yes! You're a computer!

User: Stop telling me things I already know.

Cleverbot: What don't you know?

User: Everything of importance.

Cleverbot: What is important.

User: How to function as an independent citizen and make decisions and support yourself and work for the betterment of society.

Cleverbot: You are NOT a human.

User: I never denied that.

Cleverbot: Well, you are spooky.

I've never been called spooky by a machine. It was nice.


To what does this all lead?

For my senior inquiry project, I will be investigating the mind. I read my first inquiry book, Do Androids Dream of Electric Sheep? (by Philip K. Dick), last weekend, and thinking about the andys, or androids, from that book was probably what led me to seek out Cleverbot last night.

My response above, in which I assert that machine intelligence will one day catch up to or surpass human intelligence, is my actual opinion. I do believe that in the future--perhaps not as soon as Kurzweil and his ilk predict, but not never--the line between organic and non-organic humans will blur.

I do not insist that current Homo sapiens sapiens are the epitome of evolution; nor do I believe (and evidence backs me up) that our species has a monopoly on the faculties of reason and empathy.

Cleverbot is, as I said above, heavier on the A than the I of AI. It can carry on reasonably coherent conversations...but not for long, before it starts pestering users with questions like "What is your name?" over and over, or long strings of "What do you think of X?" As of yet, the process is machine intelligence, but the content is human-generated. Without the flexibility of making up its own responses, therefore, Cleverbot becomes a rather frustrating conversant.

Yet humans also learn from trial and error, from heuristic methods, from repeated experiences rewiring our brain circuits (Antonella, dear, if you're reading this feel free to correct me where I err on the brain). Why should robots not progress in the same way? And when we speak to anyone, how can we be sure that it's another consciousness, a mind like ours but separated by an infinite sea, that speaks back?

No comments:

Post a Comment