Demonstration: The Kuleshov Effect

The Kuleshov effect can work to the benefit of conversational systems designers, by making irrelevant responses seem relevant. Thanks to Robby Garner for drawing attention to this phenomenon in the context of conversational systems.

In 1918, the Russian filmmaker Lev Kuleshov carried out a revealing experiment. He inserted the same expressionless shot of a well known actor into three different situations. When audiences saw the movie sequences, they interpreted the actor's expression as pensive, sorrowful and happy respectively, depending on the context in which the reaction shot appeared. Moreover, critics praised the actor for the expressiveness of his acting. However, the actor was not performing in any of these shots, and at the time the film was shot he was totally unaware of the contexts in which his face would be used. The implication is that audiences will read into an impassive face emotions that they themselves are feeling.

A similar effect can happen with conversational systems. If the system gives a response which seems unconnected to your utterance, people will strive to interpret it in a positive light. We've all had this experience in everyday conversation, when we try to make sense of something someone has said to us which didn't seem to fit with what we were expecting.

A demonstration is set up below to respond with random responses to anything you say. Try holding a normal conversation with Ezra the cat (assuming one can have a normal conversation with a cartoon cat!). You may be surprised how often his responses seem relevant. If so, you're experiencing the chatterbot Kuleshov effect. The images of Ezra are also selected at random, so if you find them expressive, this is the original Kuleshov effect in action as well!

More about the Kuleshov effect from Wikipedia