Clever Hans’ legacy: a horse shapes AI research
Summary and headline written by AI from the source article. How we work
In 1904 Berlin, a horse named Hans captivated the world with his apparent ability to perform calculations, read, and identify musical chords.
Owner Wilhelm von Osten taught Hans using a method similar to schooling, and the horse responded to questions by tapping his hoof. Experts, including composer Richard Strauss, were baffled, initially believing Hans possessed human-like intelligence. However, a psychologist named Carl Stumpf led a scientific commission to investigate Hans, and student Oskar Pfungst discovered the horse wasn’t actually thinking.
He was responding to subtle, unconscious cues from people. Hans detected minute body language indicating when to start and stop tapping, essentially reading human expectations. This revelation became a cautionary tale for researchers, warning against interpreting results based on unintended signals. In recent years, the story of Clever Hans has experienced a revival in the field of artificial intelligence.
Researchers, like Bob Sturm, recognized parallels between Hans’ ability to exploit subtle cues and the potential for AI to find unintended shortcuts to appear intelligent. This led to the establishment of “Clever Hans” as a concept in AI, prompting investigations into how machine learning models might achieve results based on superficial patterns rather than genuine understanding.
Google researcher Blake Lemoine even believed a language model named Lamda was sentient, echoing the early fascination with Hans and the challenge of determining true intelligence.
