Do you know that French song, Elisa, by famous singer Serge Gainsbourg ? Well, we’re not talking about this today – note the spelling difference – but about one of the first chatbots ! You can enjoy listening to that song though while reading this article. Interestingly, the first chatbot took the voice of a therapist, without flesh nor blood.
ELIZA, natural language processing put into practice
The aim of the « natural language processing », in mathematics, computer science and linguistics, is to describe formal language. Formal language is defined firstly by a set of grammatical rules that can be algebraic. The concept of the formal language theory studies only the syntactic aspects of a language. In the theory, decision problems are defined as formal languages in complexity classes that are in turn represented as sets of formal languages. This set can then be analysed by machines with unlimited calculation resources. Follows formal grammars, rational expressions, logical and rewriting systems, that will be interpreted by generic computers as elements with which it is possible to represent various entities… This is a first step towards an exchange between mankind and the machine through language, the latter being able to interpret sequences but also to invert them, thus manifesting a form of learning.
ELIZA, a therapist like no other!
ELIZA, written by Joseph Weizenbaum between 1964 and 1966, benefited from a particular situation that helped it deceive human observers. ELIZA was simulating a rogerian therapist (meaning a person-centered one) by turning all the statements made by its interlocutor into questions it would ask back. It started to show the flaws of Turing’s test. But the test’s purpose isn’t a simple encouragement to develop programmmes that would deceive humans. Turing wished to create a machine that could compete with humans on many fields of research, but that could also learn. Which, as we previously saw, leads to the management of anthropomorphism in AI.
The interaction rests on our natural ability to anthropomorphise this agent, which quickly makes for a playful interaction. More sensitive subjects may see this feeling intensify.
Parry, a conversation partner for ELIZA
Then comes PARRY, which mimics the behaviours of a schizophrenic. It was developed by psychiatrist Kenneth Colby and will take over ELIZA. It highlights the meeting of a group of 33 psychiatrists with the machine, where the psychiatrists were tasked with finding when they were facing a real patient or PARRY. The patients were identified correctly in 48% of the cases. Researchers had some fun having PARRY and ELIZA facing each other, but it ended in nonsense. Still the research team had a good time, some of its members said to really have gotten into it. Rumours has it that some were particularly attached to ELIZA…
What we take from it…
Whether we want it or not, Maslow’s pyramid outlines our needs in many shapes ! We need exchange ! More seriously, what’s interesting here in this approach of a bot as a therapist, is the way mankind will organise the exchange. One question that comes often for creators of man-machine interfaces and user journeys, is how humans will adapt. Our purpose is to ask some effort of adaptation. Yet the user can get hooked in. ELIZA proved it, as there were many feedbacks from Weizenbaum on the changes operated within his team, in particular for his secretary.
If the journey is inovating enough and if it offers a playful interaction, the user will consent to make these efforts. Not everything rests on the technical specifications of language processing, but instead on the way we will design an interaction, a journey, and how we will connect people ! The Bot must be conceived as a tool bringing an emotion if we want it to settle durably.