Robot

Artificial Intelligence

Cleverbot

Please note that the following exercise involves interacting with one of these chatterbots. Because the chatterbot has learnt from humans, it will quite possibly have been taught to say things that you may find highly offensive. While we have tried to choose chatterbots that mostly say things that aren’t going to offend, it is impossible to guarantee this, so use your discretion with them; you can skip this section and still cover the main concepts of this chapter. Because Eliza and Alice don’t learn from humans, they won’t say offensive things unless you do first!

And again, don’t tell the chatterbots your personal details (such as your full name, date of birth, address, or any other information you wouldn’t be happy sharing with everybody). Just make stuff up where necessary. A chatterbot that learns from people quite possibly will pass on what you say to other people in an attempt to sound intelligent to them!

These warnings will make more sense once you’ve learnt how these chatterbots work.

We saw previously examples of two chatterbots that have rules that help them determine suitable things to say. One of those chatterbots only works in a restricted domain, taking the role of a therapist, and is unable to have very general conversations, and the other is able to have more general conversations. Both these chatterbots had their rules of what to say determined by programmers at the time of programming, and these rules will never be changed unless a programmer decides to change them.

There are other chatterbots that are able to learn their rules from the humans they have conversations with. By looking at how a human responds to various dialogues, the chatterbot attempts to learn how it should respond in various situations. The idea is that if it responds in similar ways to what a human does, then perhaps it will sound like a human. Most of these chatterbots aim to have very general conversations, i.e. they aren’t restrained to one domain such as Eliza the therapist is.

If it is human intelligence you are trying to simulate, then perhaps learning from humans is the way to go?

An example of a chatterbot that learns from humans is Cleverbot.

Click on this link to have a conversation with Cleverbot

Unlike Eliza and Alice, whose rules of what to say were determined by programmers, Cleverbot learns rules based on what people say. For example, when Cleverbot says "hi" to a person, it keeps track of all the different responses that people make to that, such as "hi", "hello!", "hey ya", "sup!". A rule is made that says that if somebody says hi to you, then the things that people have commonly said in response to Cleverbot saying hi are appropriate things to say in response to "hi". In turn, when Cleverbot says something like "sup!" or "hello!", it will look at how humans respond to that in order to learn appropriate response for those. And then it will learn responses for those responses. This allows Cleverbot to built up an increasingly large database.

An implication of learning from humans is that Cleverbot makes the assumption that the humans actually are intelligent, and will teach it to say intelligent things. If for example people told Cleverbot something like "School is boring" in response to Cleverbot saying "hi", Cleverbot might learn that when a person says "hi" to it, it should say "School is boring"!