Bots that Chatter

You: Today has not been very good so far.

Eliza: Why today not been very good so far?

You: My dog ate my homework and I fell down the stairs.

Eliza: Are you here because your dog ate your homework and you fell down the stairs?

You: I’m not sure why I’m here.

Eliza: You’re not sure why you’re here? Would you like to talk about that?

Eliza was first created in the 1960’s by Joseph Weizenbaum of MIT and is one of the oldest examples of a chatterbot. Designed to simulate a patient/psychologist relationship, Eliza reverses, rewords, and repeats what the user inputs in order to find the heart of what’s on the user’s mind.

A chatterbot (or chatterbox) is a program that is designed to communicate with humans. They are commonly associated with the Turing Test because a communication program is one of the easiest ways for a robot to convince a human that it is another human.

The most common and the simplest chatterboxes contain a database of phrases that a user might type in. They have one or two automated responses they can spit out for each phrase. When designed well, these can be rather intriguing.

Another less common chatterbot remembers what people tell it and stores each entry as a response to a statement. This is supposed to make it more human-like by using actual human responses; however, they are usually not very good at determining what should follow what. In addition, the responses can only be as good as those of the people doing the input.

Alice the chat bot, winner of the 2004 Loebner Prize. (Photo courtesy of Dr. Wallace, reproduced with permission)

It is easy to think of a chatterbox as nothing more than a fun way to kill a couple of hours. Not all chatterbots are innocent, however. The growing popularity of instant messenger and chat rooms has caused malicious chatterboxes to come about. They post links to websites containing viruses or subjects most Internet surfers should stay away from. Most people have learned what they look like and when to avoid them, but some can be very tricky.

Some of the more intelligent chatterboxes contain routines that randomly vary the speed characters appear on the screen. This makes the bot more realistic and believable.

The biggest problem with chatterbots today is that they aren’t very good. Unless a programmer has planned for millions of different responses, a person will notice very quickly that the bot’s responses are not those of a human. Because language is so flexible, keywords are difficult to plan for and place correctly. In addition, most chatterbots cannot remember more than one or two statements, causing a conversation of any sort to be difficult. Learn more about Natural Language Processing.

See also: Robots in Sports, AI Toys, Intelligent Games