Table of Materials
Hi everybody, we are together this week for an article about the history of the chatbot, from its birth to today. Before you start a few keywords to arouse your interest: Turing Test, Eliza, MIT, Siri and Alexa. Continue reading to understand the link between all of them.
A chatbot is a program that tries to have a conversation with a person for a few minutes whose purpose is to give him the impression of addressing a human. We might then think that the program is supposed to understand what the person is saying, but that impression is wrong, most conversational agents are not designed to understand. They locate keywords or phrases that are called triggers,to find the answer in the database. It works to a certain limit, the conversation is more or less intelligent, and does not require understanding what they are talking about.
To improve the flaws of this techniques, for example associate “how does it work?”with “How does the software work?”, a system based on the keyword recognition method described above
This is why two types of chatbots must be differentiated:
- – simple bots, which is based on the keyword method and
- – intelligent bots, which use linguistic analysis based on natural language comprehension technology.
Before attacking on the first best-known chatbot, Eliza, let me introduce you to it ancestor.
The word ancestor might sound a bit strong to you, but if I tell you that Eliza was created in 1964 and that the chatbot I’m talking about was created back in 1780, you may understand me better.
The man behind the first chatbot searches is none other than a French abbot, Father Mical, he developed two copper talking heads, which were able to pronounce four sentences by clumsily reproducing the human voice, giving the impression that they had a conversation.
Following this invention there was Eliza as evoked above which was created by an MIT professor, Joseph Weizenbaum. Its operation was quite simple:
- – It started with Good morning. Why are you coming to see me? Then depending on the user’s response she will try to find a question in the extension.
- – Another scenario: the user asks her a question, she answers him by questioning the origin of this question: Why this question?
- – And if the caller uttered a sentence that included the word computer (= keyword search system as stated in the definition), she asks: Do you say this because I am a machine?
The following chatbots are based on Eliza such as PARRY created in 1972 which actually interacted with her and then ALICE in 1995 and so on… Chatbots are more and more numerous, among the well-known, Siri and Alexa.
Who's the best chatbot?
But then how could one evaluate the capabilities of a chatbot? This is where the Turing test comes in, it is intended to test the category of chatbots that wants to imitate humans.
Set up by Alan Turing in 1950, this test works in the following way. A jury speaks with 2 users via interposed screens, one of the them is a chatbot and the other one a normal human. At the end of the test, the jury selects the one they think is a robot.
The flaws in the test are as follows, if the objective of the chatbot is to answer a client’s questions but he cannot simulate a certain humanity, then he will not validate the test despite the fact that he may answer very well the questions.
After this test was introduced, the Loebner Prize was established in 1990 by Hugh Loebner to award the most humane computer program. Some chatbots who won this award: ALICE quoted above won the prize in the following years: 2000,2001, 2004. The last winner is Mitsuku.
One should ask whether the Tuning test is still relevant to test chatbots because according to research conducted by MIT, the more the chatbot seeks to appear human the more they distrust it.
A little fun fact before you leave? After the release of alexa, the number of girls named after that name halved from 2015 to 2018.
The final word
Don’t hesitate to follow us on social networks to be kept up to date with our next articles, but in general it’s once a week.