Artificial Intelligence

We saw previously that Eliza is a chatterbot who works with a restricted domain (trying to take the role of a therapist), and tried to sound intelligent by giving very general responses. As you probably found out though, this quickly reveals that Eliza really isn’t intelligent at all, and will just repeat the same things over and over again. In addition, you probably also noticed that Eliza frequently doesn’t make a lot of sense, and comes up with sentences that make no sense to a human.

Try having a conversation with another chatterbot --- Alice. Note that she knows you as 'judge'. You might work out why she calls you this once you have read the sections on the Turing Test later on!

Does Alice sound more intelligent than Eliza? Does she sound as intelligent as a human, or can you trick her into saying things that make no sense?.

Like Eliza, Alice also uses hard coded rules to know what to say, although she has a lot more of these rules and a more sophisticated algorithm for choosing which rules should be used to determine what to say. If you want to know more, just ask Alice how she works, and she’ll tell you!

To use the rules, the chatterbot analyses your last sentence (and possibly earlier sentences also) by matching the sentence against the patterns in its rules. If the sentence matches the rule pattern, it 'activates' the rule. The rule will specify a reply. Sometimes the reply is just a template for a sentence, and the chatterbot analyses the text you typed to extract words that can be inserted into the template. That’s why the replies sometimes use part of what you typed.

The quality of these rules determines how intelligent the chatterbot will sound.

While Alice is undoubtedly more intelligent than Eliza, Alice still has her limitations. Sometimes she even admits to those limitations!

Judge: What will happen in the year 2000
Alice: I cannot predict the future.
Judge: That is the past
Alice: Yes it is one of the past.
Judge: Then why did you say you can't predict the future?
Alice: It was a preprogrammed response completely determined by your statements.

I suppose I was a bit mean for tricking her and saying 'what will'. Her rules assume a human wouldn’t mislead her like that.