The target article proposes that “mindreading is prior to metacognition,” meaning that just as we know the minds of others by observing what they do, we know our own minds by observing what we do. According to this view, metacognition – that is, cognition about one's own cognition – requires mindreading abilities. Rhesus monkeys (Macaca mulatta) do not appear to possess mindreading abilities (Anderson et al. Reference Anderson, Montant and Schmitt1996; but see Santos et al. Reference Santos, Nissen and Ferrugia2006). Here we present evidence, however, that rhesus monkeys are metacognitive. We offer a different definition of mindreading than that used by Carruthers, and we contend that the mechanisms of mindreading and metacognition are largely orthogonal.

The target article reports in detail on only a few seminal studies of metacognition in animals (see Smith et al. Reference Smith, Shields and Washburn2003; Smith & Washburn Reference Smith and Washburn2005). We begin by elaborating on subsequent studies that provide evidence of animal metacognition (reviewed by Kornell, in press). For example, Hampton (Reference Hampton2001) tested monkeys in a modified delayed match-to-sample task: On each trial, a sample picture was presented on a touch-sensitive computer monitor, and then, after a delay, the same sample picture was presented among three distractors, and the subject had to touch the sample. On some trials, after viewing the sample, the monkey could choose to skip the test and receive a small reward. If the monkey instead chose to take the test, he could earn a large reward, or, if his response was incorrect, forfeit reward completely. Memory accuracy was better on self-selected test trials than on mandatory test trials. It appears that the monkeys chose to take the test when they knew that they knew the answer, in the same way that a student raises her hand in class when she knows that she knows the answer (see Suda-King Reference Suda-King2008, for similar results in orangutans).

In another study, two male rhesus monkeys were asked, essentially, to bet on their memories (Kornell et al. Reference Kornell, Son and Terrace2007). A given monkey was shown six pictures sequentially for “study,” followed by a display of nine pictures presented simultaneously, one of which had been “studied.” The monkey's task was to select the studied picture. After he responded, two “risk” icons were presented, which allowed the monkey to bet his tokens (which could be exchanged for food). A high-risk bet resulted in the gain of three tokens if the monkey had responded correctly, but a loss of three tokens otherwise. Choosing low-risk resulted in a sure gain of one token. The monkeys made accurate confidence judgments: They bet more after correct responses than after incorrect responses. This finding was especially impressive because the monkeys were originally trained on tasks that involved neither pictures nor remembering (e.g., select the longest line); following that training, they were able to respond metacognitively beginning on the first day of the picture-memory task. The monkeys appear to have learned a general metacognitive response, not one that was task-specific.

In addition to being able to make judgments about their memories, monkeys have demonstrated that they can choose behaviors, based on metacognition, that advance their own knowledge – that is, they have demonstrated metacognitive control (see Nelson & Narens Reference Nelson, Narens and Bower1990). To investigate this ability, we allowed two monkeys to request information when they were uncertain, just as a person might ask for a hint when answering a difficult question (Kornell et al. Reference Kornell, Son and Terrace2007). The monkeys could request a “hint,” that is, a blinking border that surrounded the correct response, on some trials in a list-learning experiment. As the monkeys' response accuracy on no-hint trials improved steadily, their rate of hint requests showed a corresponding decline. By requesting hints when they were unsure, the monkeys went beyond making an uncertain response; they took steps to rectify their ignorance.

Based on the studies described above, we conclude that monkeys have metacognitive abilities – that is, they can monitor the strength of their own internal memory representations. According to the target article, these findings fall short of metacognition, however. Carruthers writes, “It is only if a human reports that she acted as she did, not just because she was uncertain, but because she was aware of being uncertain, that there will be any conflict [with the metacognition is prior account]” (sect. 5.2, para. 3). We do not agree that metacognition requires awareness; we have previously argued that the metacognitive abilities that animals possess are not necessarily conscious (Kornell, in press; Son & Kornell Reference Son, Kornell, Terrace and Metcalfe2005; also see Reder Reference Reder1996). For example, a monkey might make a high-risk bet without being aware that it is monitoring its internal memory trace.

We are not arguing that mindreading cannot subsume metacognitive functions. Indeed, we can learn much about ourselves by observing our own behavior: for example, after playing a round of golf, we decide we are not quite ready for the pro tour. Moreover, numerous experiments have shown that metacognition is largely based on unconscious inferential processes, not direct examination of memories; for example, we infer that we know something well based on the fluency (i.e., ease and speed) with which it comes to mind (Schwartz et al. Reference Schwartz, Benjamin and Bjork1997).

Given the way we, and many other cognitive psychologists, define metacognition, we assert that it is likely that metacognition and mindreading are separate processes. The argument that one should only see metacognition in species that can mindread is, to the best available evidence, false. For example, some have suggested that dogs, which have shown no metacognitive abilities but show high levels of social cognition, may have rudimentary mindreading abilities (Horowitz, in press; Tomasello et al. Reference Tomasello, Hare and Agnetta1999). Conversely, we offer rhesus monkeys as a case study in a metacognitively competent animal that fares poorly at mindreading. In the tasks we describe, metacognitive processing can lead to positive outcomes that are evolutionarily adaptive. Indeed, metacognitive monitoring seems to have its own rewards.