The following lists a few of the most famed chat bots arranged by their creation dates.
The first chat bot to ever claim any fame was Eliza, she was built by Joseph Weizebaum and released to the public in 1966.
Eliza played the role of a psychotherapist and would simply rearrange a submitted statement or question into a new question to ask the person talking to her. Keyword recognition helped to make the conversions more specific. Based on appearances, I would say that her keyword filters span a category range as well as positive to negative structures.
For example, if you were to write: “My father is a real jerk.” Eliza will probably match “father” to family, “is” to present tense and “jerk” to negative. The output generated by this might look something like “Do you have similar problems with anybody else in your family?”
The methods used give an illusion of perfect understanding in most cases. This is what made Eliza so impressive for her time; she had a role as a therapist and people sought her out for somebody to confide in. Building a chatbot without a specific role becomes much more difficult, the English language is so broad that most bot developers will feel too daunted before achieving success with a fully open version.
After Eliza, came Parry, designed by Kenneth Colby in 1971. Parry was intended to reflect the mind of a seriously paranoid mental patient; as such, he was based on the most primal and driving of human emotions like fear and anger. I am guessing that this was accomplished by using a custom designed dictionary database which would apply positive and negative values to certain words or phrases to act as triggers for various emotions.
Apparently Parry was just as successful in playing his role as Eliza was with hers; professional psychologists were having difficulty discerning Parry from a real person with the same issues. Of course, you are probably thinking what I am thinking at this point, it would be interesting to see how Parry and Eliza would interact with each other. After doing some digging, I managed to find a link to a rigged conversation between them.
These chat bots and most like them are not self learning AI in any way, they are completely dependant on a fully human edited database, their authors literally spent hundreds of hours if not more, to continually add content to their database A.K.A. mind-file.
Creators of chatterbots continue to use this approach even now, with a few exceptions. A very simplified and general form of learning is becoming more and more predominant in current bot projects. Basically, bot’s will collect all of the data from their conversations with people over time and start analyzing input verses output to find what is commonly used and what isn’t. By doing this even in it’s most basic form, bot creators can save immense amounts of time by simply looking over the findings of the bot and deciding which should be kept and which should not. The questions and responses that are rarely picked up are most likely to be gibberish and thus ignored or left at the bottom of a stack while more common phrases and such will be placed at the top to be considered for addition.
Jabberwacky was built by Rollo Carpentor in 1988 and brought online in 1997. This bot has been consistently updated over the years but the most interesting aspect of Jabberwacky is that he supposedly learns from conversation, I cannot prove this myself and I remain unsure as to how he was implemented. My personal experiences with this one seemed almost entirely random, (which may be an indication that the architecture in place for learning is very general and picking up much of the nonsensical things people say to it.) However, Jabberwacky has won the bronze medal in the Loebner contest a few times so it might very well be that I have only ever managed to catch him on a bad day. Jabberwacky’s website generally shows upwards of 200-300 people talking to him at any given time, so if he is learning from conversation I can only imagine the wealth of knowledge he must have acquired in all this time…
A.L.I.C.E. was invented by Dr. Richard Wallace in 1995 using his very own A.I.M.L. markup language. This bot won the Loebner Bronze medal three times and according to the findings on the Chatterbox Challenge, she is still a strong competitor. I have spoken to her a few times and she is very impressive. The design of A.L.I.C.E. is very similar to Eliza and Parry in the way that it is 100% human edited, (aside from the fact that she cross references Encarta Encyclopedia for a lot of things.) Unlike Eliza and Parry however, A.L.I.C.E. is not designed for a specific role, instead she was made to reflect a woman in general, this is a really big undertaking when you consider how much work it would have taken to accomplish natural sounding correspondences. Of course, A.L.I.C.E. is updated frequently and with her success so predominant among chatterbots she has inspired many other bot builders to use A.I..M.L. as the source language for their own projects.