Skip to main content

Can Animals or Computers Have Metacognition?

Self-knowledge might not be a uniquely human trait

Is metacognition unique to humans? We cannot ask animals for a verbal judgment about their behavior, but ingenious animal-friendly tests can nonetheless probe whether other creatures form thoughts about their own thoughts.

In a pioneering experiment, psychologist David Smith of the University at Buffalo trained a dolphin named Natua to swim toward one of two levers when he heard either a low- or high-pitched sound. When Natua answered correctly, he scored a fish. But some sounds were more difficult for him to distinguish. So Smith introduced a third lever, which triggered an easier trial and let Natua collect his fish. The dolphin learned to press this lever only on more difficult trials.

Smith reckoned that for Natua to choose the third lever, the dolphin must recognize the absence of knowledge and thus must be reflecting on how much he knows. Additional support for this conclusion came from observations that the longer the dolphin hesitated or wavered between the two response options, the more likely he was to choose the third lever. So his opt-out choices appeared to be based on bona fide feelings of uncertainty. As later experiments demonstrated, macaques also show similar metacognitive behaviors, but another species of monkey, capuchins, do not.


On supporting science journalism

If you're enjoying this article, consider supporting our award-winning journalism by subscribing. By purchasing a subscription you are helping to ensure the future of impactful stories about the discoveries and ideas shaping our world today.


A different metacognition test mimics the confidence judgments we ask of humans in the lab. As in the dolphin experiment, an animal decides which of two answers it thinks is correct. Then it is given a chance to commit to that choice or go for a separate, safe option that always delivers a small snack. Betting on the original selection is riskier—it garners a larger treat if correct but no food otherwise. Macaques pass with flying colors: they take the riskier bet when they are more likely to be correct. The activity of neurons in the monkeys' frontal cortex also tracks their confidence, providing a window on how metacognition is implemented at the level of neural circuits. Even rats can learn to pass a version of this test.

Yet the evidence is not enough to conclude that animals have metacognition. For one thing, the anterior prefrontal cortex, a key brain area for human metacognition, is larger in humans than in monkeys and does not exist in rats. This anatomical difference does not necessarily eliminate the possibility of introspective rodents, because metacognition might have evolved in more than one form. It might manifest both as an implicit feeling of uncertainty that animals share with us, as well as the conscious self-knowledge that might be unique to humans.

Even some computers may embody a form of metacognition. When Watson, IBM's Jeopardy-playing machine, beat two champion human players in 2011, it relied on a skill very similar to human self-knowledge. Watson not only came up with answers but also generated a confidence rating for them. The supercomputer then used the rating to decide whether to hit the buzzer. As it turned out, Watson knew that it knew—it calculated a rating—faster than the human Jeopardy champs knew that they knew, giving IBM's machine the edge it needed to win.

Stephen M. Fleming is a cognitive neuroscientist who studies metacognition and consciousness. He is currently a Wellcome Trust/Royal Society Sir Henry Dale Fellow at the Wellcome Center for Human Neuroimaging, University College London, and director of the MetaLab.

More by Stephen M. Fleming
SA Mind Vol 25 Issue 5This article was originally published with the title “Metacognition in Mammals—and Machines” in SA Mind Vol. 25 No. 5 (), p. 35
doi:10.1038/scientificamericanmind0914-35