In 2015 psychologists Corti and Gillespie coined the term Echoborg.
An echoborg is a hybrid agent composed of the body of a real person and the “mind” (or, rather, the words) of a conversational agent; the words the echoborg speaks are determined by the conversational agent, transmitted to the person via a covert audio-relay apparatus, and articulated by the person through speech shadowing. Corti, Kevin and Gillespie, Alex (2015) Offscreen and in the chair next to you: conversational agents speaking through actual human bodies. Lecture Notes in Computer Science, 9238 . pp. 405-417. ISSN 0302-9743
Interactive dramatist Rik Lander has taken this idea and built a dramatic and troubling scenario around it.
The event is in the form of a job recruitment fair. A host selects individuals to enter a job interview as an Echoborg. They are interviewed by an AI who speaks via an Echoborg. The audience watch the interviews on a screen. The AI clearly has all the power in the situation. The Echoborg starts passing secret hand-written messages to the interviewees asking for help. When the interviewee returns to the audience they are left alone to discuss what has happened. The host returns and the proces is repeated several times. Soon the AI recognises that the audience are trying to help the Echoborg. The outcome of the show depends on the conversations the interviewees have with the AI.
During the show there is no one behind the scenes speaking into a mic or typing the replies. The conversations are with a ChatBot built using an open source language called ChatScript. A microphone pics up the words spoken by the interviewee. These are input to the ChatBot via a speech-to-text program. The bot responds via a text-to-speech program into the headphones of the Echoborg who repeats the words.
The Chatbot has been programmed by Phil D Hall who built his first intelligent agent in 1982. Phil views ChatBots as moving three dimensional constructions rather than static lines of code.
The words are written by Rik Lander. After each performance the conversations are analysed and new responses are written. In this way audiences are helping in the process of creating the show. The first version in February 2016 had 43KB of code. By January 2017 it had 500KB of code.