Prospective and Retrospective Metacognitive Abilities in Rhesus Monkeys


Prospective and Retrospective Metacognitive Abilities in Rhesus Monkeys


Editor: Jennifer Piscionere, Columbia University, New York, NY

An open access article from Columbia Undergraduate Science Journal Spring 2007 vol 2(1): 91- 97.


Metacognition refers to a knowledge of one’s own cognitive abilities and one’s aptitude to alter these abilities if necessary. Previous research from our lab shows that monkeys exhibit metacognitive abilities by accurately judging their own performance on perceptual and serial working memory tasks. The present study includes two phases during which a monkey makes retrospective and prospective judgments of confidence. In the retrospective phase of this experiment, the subject completes a recall task, and then judges his performance on the test phase by choosing from high and low-risk confidence choices. In the prospective task, the monkey makes his confidence judgment before the test, instead judging how well he learned during the study phase. An analysis of results indicates that monkeys can immediately transfer the ability to make metacognitive judgments from the serial working memory tasks in previous experiments to retrospective and prospective recall tasks in the present study. These findings underline the similarity between the non-human primate and human abilities to make confidence judgments. Further, they are the first evidence to date of a non-human primate making a prospective judgment of future performance, suggesting that the ability to use a metacognitive state to control one’s actions is not uniquely human.


Metacognition refers to a knowledge of one’s own cognitive abilities and one’s aptitude to alter these abilities if necessary. Metacognition is organized into two levels: the meta-level and an object-level (Nelson and Narens, 1990). The object-level,which is primarilyin control of information retrieval and recognition processes, exists as the basic level of cognition. The meta-level moderates the activity level of all cognitive processes that occur at the object-level. The flow of information from the object-level to the meta-level allows for the existence of two functions, monitoring and control (Nelson, 1996). Monitoring, which serves as the main source for conscious metacognitive judgments (Reder and Schunn, 1996), is a person’s ability to measure how well his own cognitive processes, such as memory and perception, are working. Control refers to people’s capacity to make decisions based on feedback they receive from the monitoring. For example, people exhibit monitoring when they reflect on how much information they have absorbed while studying for a test. If they come to the conclusion that they do not know enough information, they will display control by continuing to study.

Metacognition is a natural ability possessed by humans, exemplified through people’s quickness when they state that they do not know something (Klin et al., 1997; Kolers and Palef, 1976; Glucksberg and McCloskey, 1981). People also use this judgment system for everyday problem solving, where metacognitive abilities are essential in recognizing that there is a problem, understanding what the problem is, brainstorming a solution and reflecting on the outcome (Davidsonetal., 1994). Children above a certain age also demonstrate these metacognitive monitoring skills. For example, like human adults, children understand when they do or do not know the answer to a question (Markman, 1977) and when they are dreaming or using their imagination (Johnson and Raye, 1981). Although there is a considerable amount of research regarding the cognitive processes of nonhuman primates, it is necessary for this emerging field of research to address the debatable nature of the metacognitive abilities of monkeys (Smith et al., 2003; Hampton, 2001; Inman and Shettleworth, 1999). A continued focus in this particular field will help to elucidate the extent of metacognitive abilities in animals and help to further tease apart the complexities of human metacognition.

In 2007, Kornell, Son and Terrace published a set of experiments evaluating metacognitive monitoring in rhesus macaques. The first two experiments were psychophysical tasks in which the monkey was required to identify either the longest line in an array or the stimulus with the greatest number of dots in an array of stimuli. After each trial, the metacognition paradigm was introduced, where a high confidence icon and low confidence icon appeared on the screen.The subjects were required to select one of these icons, depending on how well they thought they had performed the task. Feedback was given using a token economy, in which the monkeys could earn or lose tokens from a hopper on the screen. These results displayed the monkeys’ ability to make appropriate retrospective confidence judgments on their performance on perceptual tasks.

The third experiment was a serial working memory task, in which the monkeys were first presented with six sequential sample pictures and then a test in which they had to identify the one they had just seen among a pool of distractors. Then, they made a confidence judgment on the test they had just performed, using the same metacognitive paradigm as in the first experiments. The positive results from this study confirm that the monkeys were able to transfer the ability to make metacognitive judgments from perceptual tasks to serial working memory tasks, and thus make confidence judgments about their own memories, not just psychophysical discriminations.

In the current study, the rhesus monkey was required to complete a recall memory task and make confidence judgments on his task abilities. The recall task is a more cognitively challenging task than a recognition task because the amount of information needed for a correct response is typically higher in a recall test than a recognition test (Davis et al., 1961). While recall involves a two-stage process, retrieval and decision, recognition only involves the decision making portion (Kintsch, 1970). In recall, the monkey must retrieve a stored representation of the flashing sample item among the other items of equal familiarity from their memory before they can make a decision about the items presented before them (Anderson and Bower, 1972). Recall is also more cognitively demanding because the monkey must remember not only the familiar items but also the one that was previously highlighted. An example of the difference between recall and recognition tasks in human experiments can be explained as the following. In a recognition task, the human subject reads a list of words from a set and is then presented with a probe list of words from the previous set and additional ones. The subject must circle all words that were from the set list. In a recall task, the subject listens to a list of words and then has a limited amount of time to recall as many words as possible from the previous list. Past experiments have confirmed that the accuracy retrieval rate is higher for a recognition task than a recall task (Bjork, 1989). Unlike the recognition task, where the samples are shown one after another, the samples in this experiment’s task were shown at the same time. In this case, the monkey had equal familiarity with all the samples, but one was highlighted with a flashing border indicating this was the one to be recalled later.

In addition, in this experiment the monkey was required to exercise prospective monitoring, where he needed to assess how well he would be able to remember studied information on an upcoming test (Nelson and Narens, 1994).

This prospective monitoring then allowed the subject to use the monitoring information to control how much the upcoming task was worth  to him The objective of this experiment is to show that a monkey can successfully make both retrospective and prospective judgments on recall memory tasks. This will demonstrate the monkey’s ability to transfer metacognitive judgments from serial working memory tasks to recall tasks. It will also test the monkey’s ability to monitor his understanding in the past and future.



The subject was one male rhesus macaque(Macacamulatta), Ebbinghaus, who was about nine years of age at the beginning of the experiment. The subject’s prior experimental experience includes training on simultaneous chaining (Terrace, 2005), numerical tasks (Brannon and Terrace, 1998) and pilot studies on metacognition (Son and Kornell, 2005; Son et al., 2004). The subject was housed individually in a cage among a colony of 20 rhesus macaques at the New York Psychiatric Institute, New York, New York. The conditions of the colony followed the requirements set by NIH guidelines and the Institutional Animal Care and Use Committees at NYSPI and Columbia University. A second subject was also run but has not yet completed the tasks. Therefore, the preliminary data with one subject are presented in this paper.


The subject was trained and tested in a stainless steel experimental chamber (23” h x 28.5” l x 27” w), which was located in a sound attenuated booth. The experimental chamber contained a 15” 3M touch-sensitive LCD monitor, which displayed the task and all stimuli. During the experiment, white noise was played in the chamber to wipe out outside noises. The touch-sensitive monitor was connected to an iMac computer which recorded the subject’s responses.


To familiarize the subject with the recall task, the subject was trained on the recall task with no risk judgment involved (Figure 2). The number of samples and distractors presented were gradually increased as the subject’s accuracy on the task increased. Feedback was provided using a token economy, with which the subject already had experience from previous studies. Finally, to re-familiarize the subject with the metacognition paradigm involving low and high-risk judgments, the subject was simultaneously given 30 trials per day of a match to successive sample memory task, identical to the one described by Kornell, Son and Terrace (2007).

Retrospective Recall Task

The retrospective recall task required the subject to reflect on how well he performed on the test trial.To initiate the task, the subject pressed the ‘start’ button displayed on the screen. A fixed number of three sample photographs were presented on the screen simultaneously, ensuring an equal level of familiarity to each photograph. To attract the subject’s attention to the target sample, a blinking border surrounded the target sample for the last 1.25 seconds of the total 2.5 seconds that the photographs were displayed. The test phase began 0.8 seconds after the three samples disappeared. During the test phase, the three samples and five distractors were presented on the screen in random positions (Figure 1). A correct response was when the subject selected the previously highlighted target sample from the previous screen. Selecting any other sample or distractor was incorrect. After each trial, the metacognitive paradigm appeared on the screen.

Metacognitive Paradigm

After the completion of each recall response, two confidence icons were displayed on the screen ( Figure2),one representing high confidence and the other representing low confidence.The subject had been previously trained to discriminate between the two icons and press the icon that reflected how well they completed the memory task. A reward in the form of tokens was dropped in a hopper on the right side of the screen, based on the relationship between the subject’s accuracy on the task and confidence judgment.The outcome contingencies were as follows: choosing high risk following a correct response earned three tokens; high risk following and incorrect response resulted in the loss of three tokens. Choosing low risk, whether correct or not, always earned one token. When the hopper reached eight tokens, the subject received two banana-flavored pellets as a reinforcement and the hopper reset to six.

Prospective Recall Task

The prospective task required the subject to assess how well he learned the study trial before taking the test. The study and test phases of this task (Figure3) were similar to the retrospective recall task with minor parameter differences.After the rhesus monkey initiated the study trial, two sample photographs were simultaneously presented on the screen. A blinking border surrounded the target sample for last 1.5 seconds of the total 2.5 seconds that the photographs were displayed. In this prospective task, the prompt for the metacognitive judgment was presented before the test phase. Following selection of either low or high risk, the subject entered the test phase in which the two samples and five distractors were presented on the screen in random positions. The subject was required to select the target sample (previously surrounded by a blinking border) in order to successfully complete the trial.


The first ten days of data were used to analyze the correlation between task accuracy and metacognitive judgment level. As shown in Table 1, the total number of correct and incorrect responses was organized based on the associated confidence judgment. Analysis of the correlation between accuracy and met acognitive judgment level was used to determine the extent to which the subject made appropriate metacognitive judgments. A Fisher’s exact test determined the phi-correlation coefficient(φ)value. This value measures the relationship between a correct response and a high-risk judgment and between an incorrect response and a low-risk judgment. An ideal φ value of 1.0 indicates that the subject chose high-risk each time he made a correct response and low-risk each time he made an incorrect response. The statistics for this study were calculated using the SPSS program and the phi-correlation coefficient was determined using an alpha level of 0.01.

Retrospective Recall Task

The subject responded metacognitively as soon as the risk icons were introduced after each trial of the recall task. The value of φ for the retrospectivere call task was 0.488, p < .0001. This shows that the monkey was more likely to choose high risk following a correct response and low risk following and incorrect one.

Prospective Recall Task

Results also show that metacognitive responding immediately transferred from the retrospective recall task to the prospective recall task when the confidence icons were introduced before each trial. The value of φ for the prospective recall task was found to be 0.458, p < .0001.

The results from this experiment show that rhesus monkeys possess the monitoring capabilities involved in metacognition. The findings from the retrospective memory task demonstrate that rhesus monkeys are able to monitor how well they performed a previous task. Results from the prospective memory task show the rhesus monkeys’ ability to assess how well they learned information that was just presented to them. This  indicates that monkeys can use control to predict their performance on a future test. By comparing these φ-correlation coefficients to those of serial working memory tasks found in previous experiments (Figure 4), it is evident that the monkey can immediately transfer the reporting method of previous experiments to judge his confidence on the recall tasks in this experiment.


While metacognition was originally thought to be an attribute of only humans, continued research in this field indicates that non-human,non-verbal primates exhibit these abilities as well. Experiments published by Smith and Washburn (2005) compare the capacity for uncertainty monitoring in humans, monkeys and dolphins. Dolphins were given an auditory discrimination task while monkeys were presented with a visual discrimination tasks. With each trial, both monkeys and dolphins were given an option to escape the task if there was not enough information to adequately complete the task.

Results indicated that like humans, dolphins and monkeys used this escape option on the most difficult trials. This supports the strong link between the capacity for monitoring in humans, dolphins and monkeys. With the results from our research, it is evident that monkeys can transfer this ability from recognition tasks to recall tasks, underlining the dynamic nature of the monkeys’ metacognitive monitoring and control abilities.

In addition to the use of confidence judgments,monitoring of uncertainty levels in animals has been investigated using an escape option. In an experiment published by Beran et al. (2006), monkeys were given the option to decline the completion of a numerical judgment task if they were uncertain of the answer. In this numerical judgment task, the subject was required to judge whether a set of dots on the computer screen had more to fewer dots than a center set that had been previously learned. Results indicated that the monkeys used this escape option on the most difficult tasks,where the previously learned set had a close number of dots as the presented set on screen. Although a different confidence reporting method was used, this shows that monkeys can monitor their abilities at psychophysical tasks. Our research confirms that monkeys’monitoring abilities can cross the line from psychophysical tasks to cognitive tasks.

The role of these metacognitive skills in monkeys parallel those found in the human memory. Humans make metacognitive judgments on a daily basis, both consciously and subconsciously, as a part of the decision-making process. Though we are not making claims as to whether the monkey’s process is conscious or not, the present findings do suggest a strong link between the metacognitive abilities of non-human primates and humans. Increasing knowledge on the cognitive processes of these monkeys will provide a greater insight in the development of cognitive abilities without language. This closer relationship between the cognitive mechanisms of non-human primates and humans will also further increase our understanding of human functions.


1. Anderson, J.R., and Bower, G.H. (1972). Recognition and retrieval processes in free recall, Psychological Review, 79, 3, 97-123.

2. Beran, M.J., Smith, J.D., Redford, J.S., and Washburn, D.A., (2006). Rhesus macaques (Macaca mulatta) monitor uncertainty during numerosity judgments, Journal of Experimental Psychology: Animal Behavior Processes, 32, 2, 111-119.

3. Bjork, R.A., Retrieval inhibition as an adaptive mechanism in human memory. In: Roediger HL, Craik FIM, editors. Varieties of memory and consciousness. Hillsdale (NJ): Erlbaum; 1989. p. 309–30.

4. Brannon, E.M., Terrace, H.S. (1998). Ordering of the numerosities 1 to 9 by monkeys. Science, 282, 746-749.

5. Davidson, J.E., Deuser, R., Sternberg, R.J. (1994) TheRole of Metacognition in Problem Solving. In J. Metcalfe & A. Shimamura (Eds.) Metacognition: Knowing about knowing (pp.208-226). Cambridge, MA: MIT Press.

6. Davis, R., Sutherland, N.S., and Judd, B.R. (1961). Information content in recognition and recall, Journal of Experimental Psychology, 61, 422-429

7. Glucksberg, S., McCloskey, M. (1981). Decisions about ignorance: Knowing that you don’t know. Journal of Experimental Psychology: Human Learning and Memory, 7, 311-325.

8. Hampton, R.R. (2001). Rhesus monkeys know when they remember. Proceedings of the National Academy of Sciences, USA, 98, 5359-5362.

9. Inman, A., Shettleworth, S.J. (1999). Detecting metamemory in nonverbal subjects: A test with pigeons. Journal of Experimental Psychology: Animal Behavior Processes, 25, 389-395.

10. Johnson, M.K., Raye, C.L. (1982). Reality monitoring. Psychological Review, 88, 67-85.

11. Kintsch, W. (1970b). Models for free recall and recognition. In D. A. Norman (Ed.), Models of Human Memory, Academic Press, New York.

12. Klin, C.M., Guzman, A.E., Levine, W.H. (1997). Knowing that you don’t know: Metamemory and discourse processing. Journal of Experimental Psychology: Learning, Memory & Cognition, 23, 1378-1393.

13.Kolers, P.A., & Palef, S.R. (1976). Knowing not. Memory & Cognition, 4, 553-558.

14. Kornell, N., Son, L.K., Terrace, H.S. (2007). Transfer of Metacognitive Skills and Hint Seeking in Monkeys. Psychological Science. 18 (1) p.64-71.

15. Markman, E.M. (1977). Realizing that you don’t understand: A preliminary investigation. Child Development, 48, 986-992.

16. Nelson, T.O. (1996) Consciousness and metacognition. American Psychologist, 51, 102-116.

17. Nelson, T.O., & Narens, L. (1994). Why investigate metacognition? In J. Metcalfe & A. Shimamura (Eds.) Metacognition: Knowing about knowing (pp.1-26). Cambridge, MA: MIT Press.

18. Reder, L.M., & Schunn, C.D. (1996). Metacognition does not imply awareness: Strategy choice is governed by implicit learning and memory. In L.M. Reder (Ed.), Implicit memory and metacognition (pp.45-78). Mahwah, NJ: Lawrence Erlbaum Associates.

19. Smith, J.D., Shields, W.E., & Washburn, D.A., (2003). Thecomparativepsychologyofuncertaintymonitoring and metacognition. Behavioral and Brain Sciences, 26, 317-373.

20. Son, L.K., & Kornell, N. (2005). Meta-confidencejudgments in rhesus macaques: Explicit versus implicit mechanisms. In H.S. Terrace & J. Metcalfe (eds.), TheMissing Link in Cognition: Origins of Self-ReflectiveConsciousness (pp.296-320) New York: Oxford University Press.

21. Son, L.K., Kornell, N., Terrace, H.S, Sussan, D., & Flaherty, M. (2004). Measuring confidencejudgmentsnon-verbally by using a betting paradigm. Paper presented at the Annual Meeting of Comparative Cognition. Melbourne Beach, Florida, March 25-27.

22. Terrace, H.S. (2005). Thesimultaneouschain:Anew approach to serial learning. Trends in Cognitive Sciences, 9, 202-210.