The previous conclusion suggests that conscious, interpersonal metacognition may be a human trait. However, what about elementary and possibly non-conscious metacognitive skills, such as trust judgment or error detection? Would non-verbal methods for analyzing metacognition be applicable to animal cognition? And would they show that some animal species have the capacity for introspection?
Kornell, Son and Terrace (2007) propose at least two ways of testing metacognition without language. On the one hand, introspection can be assessed by examining whether the animal is able to place a degree of trust in its own responses. On the other hand, metacognitive control can be examined by demonstrating that the animal "knows that it doesn't know" because it will actively seek additional information.
The first approach, based on the judgment of subjective confidence, has been the subject of a series of experimental studies by J. David Smith at New York University. The principle is simple: animals are given a simple two-choice psychophysical decision (e.g. whether a sound is high or low-pitched). The animals are then trained to use a third response in which they receive a small fixed reinforcement, whatever happens. The result: many animal species learn to use this "escape route" to good effect. A dolphin, for example (Smith et al., 1995) only uses it to specifically refuse difficult trials. The animal also shows hesitant responses (it swims slowly and even shakes its head!), precisely in response to the stimuli for which its performance is weakest.
