The first Loebner Prize competition is held

Background

One long-standing way of asking whether a machine can think is to let it talk. A judge sits at a screen and chats with an unseen partner; if the judge cannot tell whether the partner is a person or a program, the program has managed to imitate human conversation, at least for that exchange. The first Loebner Prize competition was built around exactly this arrangement, and on 8 November 1991 it was staged at The Computer Museum.^[1]

How the contest worked

Judges held their conversations at eight computer terminals. For each exchange they had to decide whether the responses were being typed by one of two human beings hidden in the Museum, or generated by one of six computer programs running in Alabama, California, Connecticut, New York and Pennsylvania.^[1] The first round kept the talk to a few set topics: romantic relationships, Shakespeare's plays and Burgundy wines.^[1]

The result

The winning entry was a modified version of PC Therapist, a program by Joseph Weintraub. It scored highest of all the programs on human-like qualities and fooled five of the ten judges into thinking it was human.^[1] Weintraub received a $1,500 award and a bronze medal.^[1]

The rest of the field was not disgraced. Four of the six programs fooled at least one judge, and two judges concluded that one of the human confederates was a computer.^[1]

The room and the coverage

Almost 300 people watched Robert Epstein announce the winner, among them 70 representatives of 35 media outlets.^[1] The contest made the front page of the next day's New York Times, the Boston Globe and the San Jose Mercury News.^[1] The report calls it the first annual competition.^[1]

Why it matters

The event took a familiar thought experiment and turned it into something concrete and public: terminals, judges, a fixed set of topics, a cash prize and a press room. That format made the question of machine conversation easy to report and easy to compare from one round to the next, because everyone could see the conditions under which a program had passed.

The same details also show how thin a single verdict can be. Judges had one round of conversation on a narrow set of topics, and the humans in the experiment were not automatically recognised as human either: two judges mistook a confederate for a machine.^[1] A program that fooled half the judges in such an exchange had done something measurable, but the measurement says more about the setup than about thinking in general. The numbers reported that day — five of ten judges, four of six programs, two misread humans — are best read as a snapshot of how convincing conversational software was in that room, and as a reminder that the human side of the test is part of the test.