Chatbots might take over the world someday.

I love chatbots - I remember trying to write one when I was at university in 1996 or so. I think I used Modula-2. Anyway, It was pretty crap, but my friends and I had fun making it respond to all queries like it was drunk.  It responded to every query with either "I love you man, you know that?", "I just wish I could find someone to love." or "You wanna take this outside sunshine?".

I'm pretty sure it would have passed the Turing test with all that.

Although modern chatbots like Cleverbot are pretty damn good, but they haven't yet passed the Turing test as far as I know.

In case it's not clear, a chatbot is basically bit of software that's programmed to respond intelligently to specific textual queries and statements.  Like if you say to an online chatbot "Hey chatboi, what's the weather in Moscow today?" it will say "Bloody cold.  You should wear a fur coat and one of those Russian hats with the ear flaps". 

If the programmer has done enough work to cover this and as many other scenarios as possible, then they can convince a user that the bot is intelligent.  But this is a façade.

Here's an analogy that I totally stole, but imagine a room full of people who only speak English.  On the door of the room there's a letter box.  Every few minutes, a new letter come through the box.

When a new letter comes in, someone in the team opens it, and inspects the content.  It's got a single word in Russian.  Nobody in the room speaks a word of Russian, but they have a manual that tells them what to do when they get words. The manual says "For the word 'Здравствуйте' go to page 156, word number 57". They find the word "Bonjour" - they write it down and post it back through the letter box.

Nobody in the room has any idea what either word meant, but they just successfully translated a word from Russian into French anyway.

Obviously, This is a simplified version of what a chatbot does. 

There is no understanding in the bot - It might give the illusion that it understands what you're saying, but it's just code and look-up tables generating the response.

Someday a chatbot or it's derivative will convincingly pass the Turing test.  One of my problems with this is that to pass the test, a computer must either lie, or not be aware it's a computer.

Neither of these scenarios seem good to me.  A computer that can convince a human that's it's also a human by lying seems like a pretty dangerous thing.

And a computer that doesn't know it's a computer?  Well that's just sad.

Here's my rules for chatbot if we want to avoid the AI taking over the world:

• Chatbot must know it's a chatbot.
• Chatbot must not lie about being a chatbot.
• Chatbot may not injure a human being or, through inaction, allow a human being to come to harm.
• Chatbot must obey orders given it by human beings except where such orders would conflict with the Third Law.
• Chatbot must protect its own existence as long as such protection does not conflict with the Third or Fourth Law.

I might have stolen a few of those.

The first two though, they're mine.

Making a rule that a chatbot can't lie would be great - we can avoid some embarrassment when the AIs take over the world and we realise that we let them because we couldn't tell that it wasn't actually Aunt Margaret on the other of of that IM session.