In artificial intelligence, the confederate effect is the phenomenon of a human being considered a machine from their textual discourse during practical Turing tests staged in the Loebner Prize for artificial intelligence.
It is the reverse of the ELIZA effect, which Sherry Turkle states is "our more general tendency to treat responsive computer programs as more intelligent than they really are" and the cause to "very small amounts of interactivity", causing humans to "project own complexity onto the undeserving object".
In the first Loebner Prize for artificial intelligence, in 1991, which deployed restricted conversational one-to-one Turing imitation games, each interrogator chatted to one artificial conversational entity (ACE) at a time, a female confederate or hidden-human, about William Shakespeare. The phenomenon was seen in the University of Surrey 2003 Loebner Prize for Artificial Intelligence, when both hidden-humans, one male and one female, were each ranked as machine by at least one judge: Judge 7 and Judge 9 ranked the female 'confederate 2' as "1.00=definitely a machine"; the male 'confederate 1' was ranked "1.00=definitely a machine" by Judge 4 and Judge 9. The gender of these two hidden-humans were incorrectly identified (male considered female; woman considered man) in independent transcript analysis ('gender-blurring' phenomenon, see Shah & Henry, 2005).