The first Loebner Prize

On 8 November 1991 the Computer Museum in Boston held the first Loebner Prize, a version of Turing's imitation game with judges, cash and a restricted topic. The winner was Joseph Weintraub's PC Therapist, a descendant of ELIZA that discussed "whimsical conversation" and, according to its author, succeeded partly because it made typing mistakes. Several judges rated it human. Marvin Minsky offered a hundred dollars to anyone who could make the contest stop.

The prize ran annually for nearly thirty years and never produced a program that most researchers considered to have passed the test. What it produced instead was an argument about the test itself. Judges could be fooled by evasion, humour and errors; programs won by manipulating the judge rather than by understanding anything; and the contest measured, at best, the design of a conversation rather than the presence of a mind.

Those objections had been made in 1950 and are made again about every chatbot. When large language models arrived in the 2020s and could sustain conversations that no Loebner judge would have doubted, the general reaction was not that the test had been passed but that it had never been the right test.