Computer scientist Melanie Mitchell argues that artificial intelligence operates as an “alien intelligence” fundamentally different from human cognition. She proposes six principles for better assessing machine capabilities, including awareness of anthropomorphic bias and rigorous experimental controls, drawing parallels to how psychologists study animal and infant cognition. The discussion explores whether AI systems truly reason or merely produce text resembling reasoning, using the cautionary tale of Clever Hans, a horse whose apparent math ability actually reflected subtle human cue-reading, to illustrate the importance of controlled experiments in AI evaluation.
