This item is only available as the following downloads:
ii THINKING ABOUT THINKING: METACOGNITION IN DOLPHINS BY JUDITH LOBO A Thesis Submitted to the Divisions of Social and Natural Sciences New College of Florida in partial fulfillment of the requirements for the degree Bachelor of Arts Under the sponsorship of Heidi Harley Sarasota, Florida May, 2012
ii Dedication To my parents To Yoya To Nane
iii Acknowledgements This thesis would never have been possible without the endless guidance and support of Professor Harley. My first y ear I was tempted into Psychology by her Cognitive Psychology course and I haven't left since. I would like to thank Professor Bauer, a fantastic academic adviser and committee member. Thanks for answering my endless questions and feeding my love of lear ning. I would like to thank Professor Beulig, whose charisma and endless knowledge base helped cultivate a love for brains in me. Thank you to Jennie Caskey, for everything from a friendly ear to providing some of the figures in this thesis. Thank y ou to my father, Edgardo Lobo, who raised me to always ask questions, to not do as you're told, but to think critically. Thank you to my mother, Emma Bique, who drove me to the library once a week when I was a kid and would sit next to me and help me with my homework every night. I owe gratitude to The Hostel, The Tripping Troupe, Those Guys and The Family. Thank you for creating the perfect, chaotic universe for me. I don't know what I would do without you guys. Especially German Ortiz, a great Zen Master for such a stressful time in my life. I owe more than just a thank you to Lydia Dumais. Thank you for saving my life all those times, for the times we would sit in a hammock and be broken hearted
iv together, under the banyan tree and the times w e would sit in a hammock in Costa Rica saving sea turtles you are an amazing companion and a terrible influence. Finally, I need to give due credit to Carlos Arias. We began this journey the first week of school and look how close we are to the finish line. Thank you for the all nighters, the frantic drives to turn in our papers (five minutes before they were due), for playing along with all my crazy tutorial ideas, for always knowing how to make me feel like I was a superhero when I could barely keep i t together, for all our dorky neurobiology conversations. I would have never made it this far if I hadn't bumped into you at that pool table.
v Table of Contents 1. What is metacognition?.......................................................... .............................1 1.1. 2. How is Metacognition studied?...........................................................................5 2.1. 3. Metacognition in C 3.1. 3.2. 3.3. 3.4. 3.5. 3.6. .16 4. 5. 5.1. 5.2. 5.3. 5.4. 5.5. 6. 6.1. 6.2. 7. 8.
vi THINKING ABOUT THINKING: METACOGN ITION IN DOLPHINS Judith Lobo New College of Florida, 2012 ABSTRACT Metacognition is the ability to think about thinking. It is a level of cognition that enables us to be self reflecting, self monitoring, and to pry through the contents of our own minds Metacognition has been found in some non human species such as the rhesus monkey. There has been one study on dolphin metacognition, however, the results can be explained through associative learning rather than metacognition. The current study sought to elicit metacognitive behavior in one spotted and four bottlenose male dolphins. The dolphins were presented with an apparatus that consisted of a transparent base with two opaque plastic buckets placed on top. In the experimental condition, the bucke ts were baited while a transparent or opaque barrier was placed between the subject and the apparatus. If the dolphin was aware of its lack of information in the opaque barrier condition, he should have looked underneath the apparatus to seek more informa tion about the placement of the baiting fish, thereby showing metacognitive behavior. Of the
vii five dolphins, only one moved on to testing. During testing, he did not look under the apparatus in the opaque barrier condition, however, he did display a form of information seeking behavior: he peeked around the opaque (and only the opaque) barrier during processing, there are not enough data to make this conclusion solid. H eidi E. Harley Social Sciences Division
1 M etacognition in Dolphins Thinking About Thinking: Do Dolphins Use Metacognition? On February 21 st of this year, the BB was proposed at the annual meeting of the American Association for the Advancement of Science, the world's most prestigious scien ce conference. The declaration, originally treatment, or be removed from its nat Among other research in favor of intelligence and social complexity in cetacean individuals, one of the arguments for as signing dolphins human rights was their metacognition. Dr. Lori Marino was quoted by the BBC world news article stating that re based on the findings of the only metacognition study executed with dolphins (Smith, Schull, Strote, McGee, Egnor & Erb, 1995), but the study's method was insufficient to demonstrate clearly that dolphins have metacognitive abilities (Hampton, 2009). Me tacognition, thoughts about one's thoughts, is difficult to demonstrate in a non human species because it requires making inferences about an internal state monitoring another internal state in a species that cannot talk. Hence, it is especially hard to ru le out other, more parsimonious explanations for behavior.
2 What is Metacognition? When a professor hands you an exam, you employ two levels of cognition to complete it. The first level, your cognition, is the information that you studied the previous nigh t. The second level is your knowledge about that cognition, i.e., your awareness of how well you remember the material, of how difficult a question is, or the realization that you need to change your approach towards the question: This is your metacognitio Is it possible for the mind to look back at itself? Accord ing to the French philosopher, Auguste Comte, metacognition should not even be possible. He argued that him reason. The organ observed and the organ observing being in this case, identical, how to Tarski, this is a secondary level that is separa ble from the initial level. This secondary level accesses the primary level in a self reflecting way. For instance, meta singing would be singing about singing, meta writing would be writing about writing. Metacognition is cognition about our primary cogni tion, knowing what we know or are thinking about our thinking.
3 Importance of Metacognition Metacognition is so prevalent in our daily lives that it is easy not to detect its influence. It can be used to select among different learning strat egies and to evaluate, revise or even abandon cognitive tasks and goals due to feelings of uncertainty or failure (Flavell, 1979). In light of its influence on learning, metacognition has been related to intelligence (Borkowski et al., 1987; Brown, 1987; Sternberg, 1984, 1986). According to Sternberg, particular task or set of tasks, and then making sure that the task or set of tasks are done p.24). Sternberg asserts that the ability to allocate cognitive resources appropriately, such as deciding when and how to complete a task, is central to intelligence. When he developed the metacognitive model, Flavell (1979) predicted that children who u se cognitive monitoring more often (versus those that do not) would learn more effectively in and beyond an academic setting, i.e., increasing a child's metacognitive knowledge and monitoring skills through systematic training would produce better students Three decades later, Kleitman and Gibson (2011) sought a correlation between academic achievement, self efficacy (judgment of an individual's own competence to complete academic tasks, Shunk, ) and metacognitive behaviors. They administered eight s urveys to 177 sixth grade students: an Academic Self efficacy scale, a Memory and Reasoning Competence Inventory (MRCI), the Progressive Achievement Test in Reading and Mathematics, a line scale Confidence Judgment, a Self handicapping Tendencies survey, a nd the Verbal Comprehension and
4 Concept Formation sub tests of the Woodcock Johnson III Brief Intellectual Ability test. The intelligence, gender, self efficacy judgments towards teachers, perception of classroom environment and school fees of the children served as control variables. The MRCI and Academic Self efficacy scale were used as measures of metacognitive skills. Positive metacognitive beliefs were a positive predictor of achievement and self confidence. Students with stronger metacognitive skills also displayed fewer self handicapping tendencies. Why should we look for metacognition in non human species? Metacognition can be important for learning, however, metacognitive activity can also hamper functioning (Sternerg, 1985). This usually applies to more automatic functioning, like motor skills. For example, tennis players often find that when they think too deliberately about their actions, the quality of their playing declines. For animals, a decline in their ability to avoid predators or capture prey could be fatal. Hence, the question of metacognitive processing in animals is quite interesting. In addition, the deifnition metacognition has an overlap with that of conscious awareness and introspection (Hampton, 2009). Accordi ng to Farthing (1992), consciousness can be defined as subjectivity, awareness, the ability to experience or feel, wakefulness, having a sense of self hood, and the executive control system of the mind. Some metacognitive experiences are best described as items of metacognitive knowledge that are a part of consciousness (Flavell, 1979). Hence, finding metacognition in a species implies an aspect of consciousness is present in that species. The presence of metacognitive abilities may impact human policies instituted to protect non human species. If a species is metacognitive, this implies a secondary, self
5 aware level of cognition. If this is the case, pain and suffering could be explicitly analyzed by the species thereby indicating the need for s pecial protection. Finally, as implemented with human children and college students, animals with metacognition may require special training techniques. If material is presented taking into account metacognitive strategies, we may be able to facilitate th as finding alternative, more effective ways of training new behaviors. Therefore, pragmatic implications. How is Metacognition Studied? Comparative studies on metacognition typically focus on metamemory: metacognitive knowledge and awareness of one's own memory. Studies on metamemory are aimed at determining whether or not a nonhuman subject can distinguish between implicit and ex plicit mental representations as humans do (e.g., Hampton, 2001). Metamemory has two functions: monitoring memory strength and controlling information seeking (Shettleworth, 2010). The skeleton of a metacognition study is a primary memory or p erceptual discrimination task that is completed by the subject along with a metacognitive judgment the subject makes about the primary task, (i.e., a report which indicates how sure the subject is of his or her answer). Metacognitive judgments can occur be fore, during or after the task. Metacognitive judgments can be defined as confidence ratings of how sure subjects are that they responded correctly. If subjects imply a high confidence rating and are correct, they receive a highly valued reward. If they a re incorrect, they do not receive
6 a reward. If they imply low confidence, a lesser valued reward is always dispensed, regardless of accuracy on the primary task. For example, a subject is asked to determine which of two tones has a higher pitch. The tones are played and before the subject can choose, a window pops up, requiring a response that allows the subject to continue, i.e., receives six pellets. If the subject responds confidently and is incorrect, no reward is dispensed. If the subject responds negatively (low confidence), three pellets are dispensed no matter which tone is indicated as having a higher pitch. Other behaviors suggesting metacognitive processing include response latency, expectation of reinforcement, and an index of memory strength (Shettleworth, 2010). Comparative studies of metacognition must include four components (Hampton, 2009). First, there must be a primary task, typically an objectively observable behavior that can be scored as accurate or inaccurate (or as efficient or inefficient). Secondly, there must be an observable variation in behavior of the subject to allow the researcher to find a correlation between the primary and a secondary behavior. This can be caused by an increase in difficulty between trials, e.g., increasing delays before allowing responses. Thirdly, the secondary behavior must be specified and observable, e.g., skipping a question, escaping a trial, requesting a hint or providing a confidence judgment. Finally, this secondary behavior must be correlated with the assessment of the primary behavior (e.g., the subject skips difficult questions more frequently). This correlation can be assessed best when the subject's knowle dge is experimentally manipulated so that the state of their knowledge can be confidently known (e.g., hints can be given during some
7 of the trials so that subjects can provide a more difficult answer, while no hints are provided in other trials). The two most common secondary responses in metacognitive studies are the uncertainty response and information response allows the subject to avoid the task (e.g., Smith, 1995). If the subject chooses to do so when it has not received enough information, then avoidance can be an indicator that the subject is aware of its lack of information. This alternative, uncertainty response yields a smaller reward or some form of delay before the next trial, so that the subject d oes not simply select the uncertainty response on every trial. For example, in the experiment Smith and his colleagues (1995) carried out on dolphins, if the dolphin chose the uncertainty response, the intertrial interval between trials increased. Other s tudies certainty (e.g., Kornell, Son & Terrace 2007). Kornell and his colleagues implemented this in a match to sample task with rhesus monkeys. If the monkeys requested a hint during the trial, they would receive a pellet. If they selected the correct response without a hint, the monkeys received M&Ms. As mentioned earlier, metacognition regulates information seeking behavior. Many comparative studies have exploited this aspect of metacognition by manipulating information availability and observing the subject's information seeking behaviors in order to complete a memory or perceptual discrimination task (e.g., Call and Carpenter, 2001). If the subject seeks information w hen, and only when, it needs more information to complete the task, this suggests metacognition because it implies that the subject is aware of its lack of information immediately available.
8 Obstacles in comparative metacognitive studies In part because metacognition implies consciousness, it attracts critical scrutiny. Even though there are now substantial data from a variety of tests of metacognition in non humans, they are subject to interpretation as being a result of simpler types of learning or a re action to externally available cues. When subjects determine whether or not they will be able to complete a task, they base their decisions on cues in their surroundings or in their internal state. Hampton (2009) stated that metacognitive cues can be divid ed into two types: public and private cues. Public cues are externally available, not only to the subject completing the task, but to anyone else who is present during the task. These cues can be environmental, behavioral or a competition between urges to respond to a task. Environmental cues are characteristics of the stimulus, like stimulus magnitude or similarity between two options. Behavioral cues can include the subject's own behaviors, such as vacillation or delay before response. Response competi tion occurs when the urge to respond to the primary stimulus competes with the urge to engage in metacognitive behavior, like seeking more information. Private cues are what we perceive to be the internal evaluations of our own knowledge, or introspecti on. Introspection is the private experience of uncertainty or confidence. To make a metacognitive judgment based on a private cue means that only internal cues cues only available to the perceiver are utilized. Metacognitive studies in non humans cannot rely on self reports, therefore, the only way to infer introspection is to rule out all of the public cues.
9 Metacognition in Comparative Cognition Even some private cues preclude metacognitive processing. Morgan's Canon (1906) makes finding metacognitio n in animals an especially difficult goal. According to behavior at the lowest psychological level. This directive makes it particularly difficult for researchers seeking s igns of metacognition because it is a secondary cognitive level. Although the studies described below have explored promising behavioral measures for the empirical investigation of animal metacognition, the findings are still vulnerable to an alternative e xplanation: Animals might have adjusted their behaviors to the perception of task difficulty rather than assessment of their cognitive states. In other words, they may be responding in an apparently metacognitive way because of the uneasiness caused by tas k difficulty rather than by metacognitive uncertainty (Hampton, 2009; Shettleworth, 2010; Smith, 2009). Nonetheless, metacognitive studies have been conducted with a number of species. Rhesus Monkeys Rhesus monkeys have been the most wid ely studied of all the species, and they are the only species to demonstrate metacognition in a task whose results cannot be explained by external cues. One of the most notable non human metacognition studies was Hampton's (2001) specialized matching to sa mple task. Rhesus monkeys were presented with a touch sensitive video monitor that displayed a clip art image in the center of the screen. After selecting it three times, a delay intervened long enough to ensure that the monkeys forgot the image (34 or 38s ). Two choices were displayed, one the left and the other on right side of the screen. The left would initiate a memory test and
10 the right was an escape response. The escape was only available during some of the trials. The monkeys had a higher accuracy wh en they freely chose to take the memory test (58.1%) than when they were forced to take a test (46.8%). These results indicate that when given the opportunity, monkeys chose adaptively to decline memory tests when memory for the sample image was relatively poor. In a second experiment, the same procedure was used, but probe trials were displayed randomly along with the normal trials. This was done to ensure that the monkeys were declining the tests due to lack of memory and not some form of external cue. During probe trials, answer options were provided but no sample initial picture was displayed for the monkeys to study. One of the monkeys declined every probe trial presented, and both of the monkeys declined the first probe trial, indicating that this wa s not a learned response. In their final experiment, Hampton compared how many times the monkeys declined the task after a short delay to versus a long delay. Due to memory decay, it was expected that the longer the delay, the more likely the subjects wo uld be to decline the test trial. Accuracy was high for both monkeys on the short delays, but this wavered with longer delays. The first monkey declined 85% of the long delays and 43% of the short delays. The second monkey declined 32% of the long delays a nd 16% of the short delays. Shettleworth (2010) argues that Hampton's (2001) study is the only study to rule direct cognitive test so that the animal could respond non rand omly, only by consulting Indeed, most of the studies describe responses that can be ascribed to some form of public
11 cue, however, the field is evolving and new models an d stricter methodologies have been developed to rule out non metacognitive possibilities (e.g., Foote & Crystal, 2012). In another study, Hampton, Zivin and Murray (2004) replicated Call and Carpenter's (2001) experiment on the metacognition of human chil dren. Seven male and two female rhesus monkeys were presented with a tray of four opaque tubes. The tubes were connected to the tray on a hinge system, so that when lifted the prize would slide out. Once a choice was made, the other hinges would lock, so o nly one choice could be made on each trial. The tubes were baited while a screen was placed between the subjects and the tubes. In one condition, the subjects were allowed to see the baiting through a transparent screen. In the other condition, an opaque s creen blocked their views of the baiting. Before testing, the monkeys were trained to look down the tubes. On tests with the opaque screen, the subjects were significantly more likely to display information seeking behavior (looking) than when they were al lowed to see the baiting. In addition, they were significantly more likely to choose the baited tube when they looked through the tubes than when they made a selection without looking. Hence, the monkeys used information seeking effectively and efficiently thereby suggesting metacognitive processing. Rhesus monkeys also express their uncertainty and frustration in their behavior during testing. Hampton (2006) did an unplanned retrospective analysis of video taped behavior in order to determine whether a rh esus monkey's apparent frustration predicted his accuracy in a matching to sample task on a trial by trial basis. The subject was seated before a screen where an image would appear. The monkey had to strike the icon twice to proceed. After one of six delay s, (0, 2, 4, 8, 16 or 32s) four images would appear on the
12 screen, and the subject had to select among them. The subject was more likely to strike the computer touchscreen aggressively when committing errors, whereas he generally touched the screen more ge ntly when selecting the correct stimulus. The monkey made an error on 47% of the trials when he hit the screen compared to 25% touch trials. This difference in behavior, which occurred before the monkey received feedback on the accuracy of his choice, sugg ests that he knew whether or not he remembered the correct response. In another experiment, rhesus monkeys were trained to make retrospective confidence judgments and were able to transfer this cognitive monitoring ability to a new task. They were also a accomplish through trial and error (Kornell, Son & Terrace, 2007). The subjects were trained to do two perceptual tasks: a line discrimination task, in which the subjects were required to se lect the longest line, and a numerical task, in which the subjects were required to select the stimulus containing the most or the least number of objects. After a choice, two icons lit up: one indicated high confidence and the other indicated low confiden ce. Completing the task correctly and selecting the high confidence icon won three tokens. Making an error and selecting the high confidence icon lost three tokens. The low confidence icon always dispensed one token. The monkeys chose the high confidence i con significantly more often after a correct response than after an incorrect response, and they chose the low confidence response more frequently after errors than after correct responses. These rhesus monkeys transferred that ability to both a new perce ptual task and a delayed matching to sample tasks. The second task was a simultaneous chaining
13 paradigm, in which all list items were displayed simultaneously throughout each trial, on a touch sensitive screen. The subject's task was to respond to the item s in order despite their spatial position on the monitor. In this task, a hint icon was added to the screen. When it was selected, four lines would indicate which item had to be selected next. If the hint was used when making a correct response, a pellet w as dispensed. If no hint was used when making a correct response, more desirable M&Ms were dispensed. Hint seeking behavior was inversely related to accuracy in the novel task. Orangutans, Chimpanzees & Gorillas One of the most emulated metacognitive st udies was conducted by Call and Carpenter (2001). They tested three orangutans and 11 chimpanzees in an object choice setting. The apes were presented with two or three tubes that were presented in parallel, facing perpendicularly toward the subjects. The experimenter hid a food reward in one of the tubes in full view of the subjects in some trials, while the baiting procedure was occluded with a screen in other trials. The apes looked into the tubes more often when the experimenter did not reveal the loca tion of the food than when the hiding process was visible. Subjects were more likely to look into the tubes when some delay was imposed as compared to no delay. Call and Carpenter (2001) also tested bonobos, gorillas and other orangutans and chimpanzees. S ubjects of all four species looked into the tubes at a greater rate when they did not know the location of the food than when they possessed this knowledge. Orangutans have also completed studies with other forms of metacognitive response. Five orangutans ( Pongo pygmaeus ) were tested to see if they could escape a spatial memory test when the location of the reward was unknown (Suda king, 2008). The
14 apes saw two identical cups, under one of which was a preferred reward (e.g., two grapes). The subjects were with which they could receive a less preferred but secure reward (e.g., one grape). The orangutans as a group were significantly more likely to select the yellow container (the uncertainty respon se) more often when they did not see the baiting procedure or when the hiding locations of the preferred reward were switched to a new location. This occurred even when the escape response was presented before the final presentation of the memory test in w hich they would likely err. Rats Foote (2007) demonstrated that rats are capable of internally evaluating how well they can perform a duration discrimination task. The rats were trained to discriminate between different durations of presentation of white noise. Pressing one lever was rewarded for long durations and pressing the other was rewarded for short durations. Before every trial, the rat could place its nose in one of two apertures. If it selected one, the lights would turn off and the test would b egin. If it selected the other, the lights would turn off until the rat retrieved a newly dispensed pellet from a dispenser. During forced choice trials, only the test aperture could be selected. A correct choice yielded 6 pellets, an incorrect choice woul d yield no pellets and choosing to not partake in a test would yield 3 pellets. The rats had a higher performance accuracy when they chose to take the test than during the forced choice tests. Dogs Dogs failed to display information seeking behavior in a food seeking task (McMahon, Macpherson & Roberts, 2010). Six dogs saw four black, cardboard boxes.
15 One of the boxes had a white side. All of the boxes were baited to control for smell cues, but the reward was only available from the box with the white si de. The experimenters turned the box by 45 degrees with each trial, before the dogs were presented with the boxes again, in order to determine if they would search for the white side. Their performance was not above chance levels, and they did not look for the white side when it was out of view. When given the choice between two humans an informant and a non informant dogs will choose the human that can provide them with relevant information (McMahon, Macpherson & Roberts, 2010). In this second experimen t, the dogs were presented with three black boxes. A large cooler obstructed the subject's view of the baiting procedure. The dogs were trained to approach one of the two confederates: one that was always the informant and one that was always the non infor mant. If the informant was chosen, the informant pointed to the baited box. When the non informant was selected, s/he would look away. In order to control for the associative learning that may have occurred with the informant, when the non informant was ch osen, all the boxes were baited. This way, no matter which informant they chose, the dogs would receive a reward. Even though it did not affect whether or not they received a reward, the dogs chose the informant significantly more often. Pigeon s Pigeons are likely to use an escape response in a memory task (Inman & Shettleworth, 1999). Pigeons were tested using a chamber that presented a memory task with a safe key they could use to escape a test. The longer the interval between the initial sti mulus and the test, the more likely they were to use the safe key. Their memory was
16 also more accurate when they could choose the safe key than when they were forced to take the test. However, when the birds had to commit to choosing between the test and t he safe key before the test appeared, they failed to use efficient strategies, suggesting that they do not have metacognition. In Sole, Shettleworth and Bennett's (2003) study, they tested the metacognition of pigeons by employing a perceptual task of va rying difficulty with an uncertainty option. Pigeons classified a display of illuminated pixels on a touchscreen as sparse or dense. Correct responses were reinforced with six food pellets; incorrect responses were unreinforced. On some trials an uncertai nty response option was available. Pecking it was always reinforced with an intermediate number of pellets. Like monkeys and people in related experiments, the birds chose the uncertainty response most often when the stimulus presented was difficult to cla ssify correctly. However, their behavior was not functionally similar to human behavior based on conscious uncertainty or to the behavior of monkeys in behavioral experiments because they may have been using the uncetainty response when the stimulus had an intermediate amount of sparseness, therefore the Dolphin A dolphin was asked to discriminate between low and high frequency tones. The dolphin was trained to respond while statione d in a chamber with three levers: a high frequency tone lever, a low frequency tone lever, and an uncertainty response lever. When a correct response was made, a reward was dispensed. An incorrect response initiated a time out phase. The uncertainty lever terminated the trial and provided an easier trial (higher or lower in frequency) (Smith et al., 1995).
17 The dolphin selected the uncertainty response only when the test was difficult. Difficult phases were those that presented a tone with a relatively mid range frequency. This was the first metacognition experiment on animals, and since then the methodologies of metacognition studies have evolved. Several features of this experiment suggest that the dolphin may have taken advantage of publicly observable c ues to guide use of the uncertainty response. The dolphin may not have been responding to a feeling of response could be mediated by associative learning not metacognit ion. The subject may have learned to select the lever when the stimulus had a mid level frequency. The subject may have also used its own publicly observable behavior as a discriminative stimulus for declining tests. The dolphin's hesitant behaviors (e.g., wavering) peaked around his perceptual threshold and were distributed like the uncertainty response, wavering may have served as a behavioral cue. The primary and secondary response were presented simultaneously, which may have provoked response competiti on. Of course, the subject may have used a private form of metacognition. However, because multiple public cues were available, invoking an introspective account may be unwarranted. The Current Study Metacognition is an important facet of human conscious ness; it allows people to probe the contents of their own minds and memories as well as devise and evaluate learning strategies (Nelson, 1996; Flavell, 1979). Metacognition reveals hierarchical structure in mind, because executive functions oversee the pri mary cognition (Metcalfe & Kober, 2005). It is a sophisticated capacity that may be rare in non human species,
18 however, the function of metacognition seems to be to facilitate efficient behavior; therfore non human species may also display this mechanism ( Shettleworth, 2010). Evidence supporting metacognition in non humans has been generated with rhesus monkeys (Hampton, 2001a, 2004b; Hampton, Zivin & Murray 2006; Kornell, Son & Terrace, 2007), rats (Foote, 2007), apes (Call & Carpenter, 2001), capuchin mo nkeys (Basile, Hampton, Suomi & Murray, 2008), and orangutans (Suda King, 2008), but not pigeons (Inman & Shettleworth, 1999; Roberts, Feeney, McMillan, MacPherson, Sole & Shettleworth & Bennet, 2003; Roberts, Feeny, McMillan, MacPherson, Musolino & Petter 2009), dogs (Brauer, Call & Tomasello, 2004; McMahon, MacPherson & Roberts, 2010) or dolphins (Smith et al., 1995). Because self reports cannot be used, these metacognition studies include a primary memory or perceptual discrimination task with easy and difficult trials paired with metacognitive reports (i.e., a report which indicates how sure the participant is of his/her answer). The difficult trials allow the researcher to create a state of uncertainty. In some cases, the researcher provides an escape option in order to allow the subject to decline the difficult task (e.g., Smith et al., 1995), while in others, a confidence judgment may be solicited (e.g., Hampton, 2001). Another common method is to use a primary task in which some trials provide enou gh information for the subject to make the correct response, while other trials occlude some of this information and observe if the subject seeks it (e.g., Hampton, 2009). This information seeking behavior demonstrates metacognition by suggesting that the subject is aware of its lack of knowledge and seeks to remedy it.
19 Morgan's Canon (1906) makes finding metacognition in animals a difficult goal. According to this tradition in comparative psychology, researchers strive to explain animal's behavior at the lowest psychological level. This directive makes it particularly difficult for researchers seeking signs of metacognition because it is a secondary cognitive level. Therefore, in order to infer that an animal uses metacognition to solve a task, we need to rule out three alternative explanations: environmental characteristics of the stimulus; behavioral responses; and response competition, i.e., the urge to respond to the primary stimulus competing with the urge to engage in metacognitive behavior (Hampton, 2009) something that previous studies on metacognition in dolphins have failed to do (e.g., Smith et al., 1995). In fact, there have been few studies that ruled out these possibilities (i.e., only Hampton, 2001). The current study seeks to present five do lphins with a primary memory task that has a variation of difficulty in order to instill a state of uncertainty. The apparatus was designed so that the dolphins could seek additional information outside of publically available cues a metacognitive strateg y. Method Subjects Five male dolphins participated in this study: Ranier (wild caught), Khyber, Calvin, Malabar (all captive born) and Moonshine (beached and rehabilitated, unsuitable for re release). Moonshine is a 12 year old Pantropical spotted do lphin ( Stenella attenuate ). Ranier, Khyber, Calvin and Malabar are bottlenose dolphins ( Tursiops truncatus ) who are 31, 20, 17 and 11 years old, respectively. All subjects had had previous experience with cognitive studies.
20 Facilities All the dolphins we re tested in their home pools at Florida public facilities. The bottlenose dolphins reside in three adjoining tanks separated by gates. The experiment was carried out in a concrete tank, out of the view of the patrons and denoted as B pool in Figure 1. B p ool is rectangular; 6.1 m by 7.62 m and 1.83 m deep. ______________________________________________________________________
21 The pantropical spotted dolphin was housed in a marine hospital and research laboratory in southwest Florida. His oval tank holds 757,000 liters of salt water, measures 38.1m by 15.25m, and is depicted in Figure 2. Testing was conducted dockside by a gate that split the tank into two sections: the shallower ea st end, 3.7m deep, and the deeper west end, 1.3m deep. _______________________________________________________________________ F igure 2. Oval tank at marine hospital and research laboratory _________________________________________________________________ _____________________ Apparatus The apparatus consisted of a transparent, plexiglass circular base with an 18 cm radius. There were two opaque, plastic buckets placed on its surface. PVC arms with cushioned ends were attached to the buckets, so that the d olphins could use their rostrums to push the buckets over an opening in the middle of the base, thereby releasing fish through the apparatus and into the water. The buckets were placed downwards and had openings at the tops into which fish were dropped. Th e base was outfitted with polyethylene foam to keep the apparatus floating on the surface of the water. The apparatus can be seen in Figure 3.
22 Two visual barriers, one transparent (rectangular plexiglass,1.21m by 1.52m) and one opaque (rectangular, lightw eight, white plastic, 1.52m y 1.82m) were used for experimental tests. See Figure 4 for the testing setup. _______________________________________________________________________ Figure 3. Apparatus with two buckets __________________________________________________________________________________ Training Procedure The dolphins were initially exposed to the apparatus with only one bucket and were cued to push the arm inwards once the bucket was baited. When they were able to push the arm consistently, a second bucket and was added. The training sessions consisted of 12 trials; each bucket was baited six times, and the order was semi randomized and counterbalanced w ith no bucket being baited more than twice in a row. Responses were recorded via data sheets and digital video technology. One subject, Khyber, reached the training criterion of at least 80% performance accuracy averaged across 5 sessions. He moved on to being trained with the transparent barrier which was placed between the dolphin and the apparatus during baiting. Then he moved on to the testing procedure.
23 Testing Procedure At the beginning of each test trial, the subject was positioned in front of a ba rrier which was placed between the dolphin and the apparatus during baiting. See Figure 4 for the testing setup. The transparent barrier was used for half of the trials (N= 6), and the opaque barrier was used for the other half (N=6). The order was counter balanced such that neither barrier was used more than three times in a row. To control for cuing effects, one trainer held the barrier while turned away, while another trainer baited a bucket. The baiter then turned away, and the barrier was lifted. A thir d person held the apparatus in place throughout the trial. The third person turned away during the baiting process. Each session consisted of 12 trials, with each bucket being baited six times. A trial was completed after the subject pushed one of the arms far enough for the fish to fall through the apparatus's opening. After one selection was made, the subject would be directed back to his position behind the barrier in preparation for the next trial. Again, responses were recorded via data sheet and video _______________________________________________________________________ Figure 2. Testing Setup _________________________________________ _______________________________ Transparent barrier Opaque Barrier
24 Video Coding The responses of the subject were coded by a single researcher. For each trial, every time the subject moved his head around the side of a bar rier sufficiently to see the he would station himself properly before witnessing the bai ting it was code d as responses where compared with the data sheets in order to determine whether Khyber made the correct response after the peeking trials.
25 Results Training Before dolphins could be tested for met acognitive behavior, they experienced a variable number of 12 trial training sessions to learn to choose the baited bucket from two alternatives. Of the five dolphins, only one dolphin reached criterion and could be tested for metacognition. During the tra ining sessions, Rainer only completed four sessions, and he performed at low levels when uncued (38%). Moonshine completed eight sessions and displayed a strong preference bias, selecting the first bucket on nearly all of the trials (M=50%, SD=3.06). Malab ar achieved a mean accuracy of 66.2% (SD=19.3) across 11 sessions, and Calvin achieved a mean accuracy of 59.6% (SD=32.9) across nine sessions. Khyber achieved a mean accuracy of 81.25% (SD=15.3) across eight sessions. See Figure 5 for a graph of these dat a. Figure 5. Mean performance accuracy of each subject for each 12 trial training session
26 Testing In the first half of the 6 testing sessions, Khyber was still adjusting to having two barrier types. His performance on the first three sess ions on the transparent barrier condition averaged 50%. However, in the second half of the testing sessions, his performance accuracy improved to average 78% in the transparent barrier condition, although it stayed at 48% in the opaque barrier condition. S ee Figure 6 for performance accuracy across sessions. __________________________________________________________________________________ Figure 6. Khyer's mean performance accuracy in each condition in each test session _________________________________ _____________________ ____________________________ Peeking The subject did not swim underneath the apparatus in order to gain information about the placement of the fish, however, on five of the opaque barrier trials the subject did display what could b e interpreted as uncertainty behavior. After Khyber was stationed in front of the opaque barrier, he poked his head around the barrier during the baiting. This occurred on 13.89% of the opaque barrier trials and did not occur during any of the
27 transparent barrier conditions. The peeking behavior occurred during the first, second, third and sixth sessions. The first occurrence of peeking was during the first trial in which s in the one but not the other (for graphic depiction of peeking see Table 1). Figure 7. Subject looked around the screen in two of the opaque trials Session Trial Suc cessful Peek Unsuccessful Peek Accuracy 1 1 Correct 8 Incorrect 2 2 Correct 3 1 Incorrect 6 6 Correct Table 1. Occurrence of Peeking by Session, Type of Peeking and Accuracy within trial
28 D iscussion Khyber did not look under neath the apparatus before making a selection during any of the opaque barrier trials; however, performance accuracy was clearly better with the transparent barriers than with the opaque barrier on the later test sessions. In addition, Khyber displayed a f orm of uncertainty behavior during five of the opaque barrier trials; he attempted to see the baiting by looking around the opaque barrier. When he was able to see the baiting, he selected the baited bucket. This behavior was also displayed with the first encounter with the opaque barrier and was more common among the first few trials. This was probably due to the trainer correcting his behavior while stationed in front of the barrier. This kind of unexpected, uncertainty behavior is a demonstration that me tacognitive ability is probably present in dolphins. The uncertainty behavior was spontaneous, untrained and unrewarded; it was not a product of associative learning. Of course, more data would be needed to confirm this assertion. The secondary response wa s designed so that it would be less efficient than a primary response in order to prevent the subject from simply swimming underneath the apparatus on every trial. Trainers reported that Khyber would look up from beneath the apparatus when he did not recei ve a reward. However, perhaps the impulse to respond by checking underneath before making his choice was overridden by the subject's long training of making a choice first, a common problem in comparative cognition studies on metacognition (for a review, s ee Hampton, 2009). Considering that Khyber had a lot of experience choosing a bucket before looking during the training trials, this is a possibility. Out of five subjects, only one of the dolphins performed with a high enough accuracy during the training sessions to move on to experimental testing. This lack of
29 success may have been due in part to the apparatus. The design of the apparatus was iterative, and it may have required yet more iterations. Anecdotal evidence from the trainers suggests that the d olphins were often not attending to the baiting. The pushing response that was requested of the dolphins may be unorthodox. In fact, Moonshine was actively discouraged from pushing objects or people prior to the experiment. The dolphins also often selected the closest choice possible. During several of the sessions, some of the dolphins displayed a strong preference bias towards one of the buckets. The next study could be improved in several ways. A simpler primary task would allow more subjects to be succ essful in training. I would also recommend using a method more akin to Hampton's (2001) study of metacognition on rhesus monkeys by modifying monkeys were presented with a match to sample task. After the sample image was presented on a touch screen a long delay (32 and 38s) elapsed. Two icons would then appear on the touch screen, one that would initiate a trial and one that would present a small reward without a trial. I n a second experiment, Hampton added probe trials in which no sample image was provided. Finally, in a third experiment, the sample image was followed by short and long delays. Hampton's results eliminated the possible influence of public, external cues, t hereby making metacognition the most probable explanation. If the dolphins participated in such a design and displayed a higher accuracy in the trials they choose to initiate, as well as declining the probe and long delay trials more often, the data would provide evidence for metacognition in dolphins. Metacognition has been found in several non human species (e.g., Basile, Hampton, Suomi & Murray, 2008; Foote, 2007, Kornell, Son & Terrace 2007). Although
30 dolphins have been studied previously (Smith et al ., 1995), the elicited metacognitive behavior may have been due to associative learning or external cues (Hampton, 2009). Dolphins are a likely candidate for metacognition due to their wide range of cognitive abilities (for a review see Herman, 2010), and large, highly convoluted brains (Elias & Schwartz, 1969; Wood & Evans, 1980). Recent research has also revealed the presence of von Economo neurons, a type of large, bipolar neuron, in the neocortex of bottlenose dolphins (Butti, Sherwood, Hakeem, Allman & Hof, 2009). These neurons were thought to occur only in humans and great apes and to be a part of the circuitry that supports human social networks (Allman, Watson, Tetrault & Hakeem, 2005). They are also implicated in reasoning, intuition and in overcomi ng uncertainty (Allman, et al., 2002). It should be noted that there is a trend in cetacean literature to depict dolphin as being sagacious creatures. There is an overlap between the domain of neurobiologists and behavioral researchers when examining dol phin intelligence. The assertion that dolphins are intelligent may lead behavioral researchers interpreting dolphin behavior as being intelligent. This in turn may lead to the interpretation that dolphin brains are comparatively complex and large. This fee dback further reinforces that dolphin behavior is intelligent yet the specialization of research may be causing a loss in translation between the disciplines ( Bradbury, 1986) Metacognition in non human species implies that they have an internal world i n which they can evaluate and remember pain and discomfort. Finding metacognition in dolphins would strengthen the need for stricter policies protecting cetaceans, as was
31 po licies, such as the Declaration of Rights for Cetaceans, on research that has been refuted may do more harm than good and stunt future efforts. Dolphins can self imitate, make analogies between their body parts and that of a human, which suggests a sens e of ownership of self and self awareness. They can interpret their body parts and their movements as being their own (Herman, 2012). Self awareness in conjunction with metacognition is a strong argument for consciousness in dolphins because it implies a s monitoring thoughts. Future research on the metacognition of dolphins could help us determine that a non human species does indeed have consciousness.
32 References Basile, B. M., Hampton, R. R., Suomi, S. J., & Murray, E. A. (2009) An assessment of memory awareness in tufted capuchin monkeys ( Cebus apella ). Animal Cognition, 12 (1), 169 180. doi:10.1007/s10071 008 0180 1 Bradbury, J. (1986). Social complexity and cooperative behavior in delphinids. In R. Schusterman, J. Thomas & F. Wood (Eds.), Dolphin Cognition and Behavior a Comparative Approach (pp. 361 372). Hillsdale, New Jersey: Lawrence Erlbaum Associates, Publishers. Brauer, J., Call, J., & Tomasello, M. (2004). Visual perspective taking in dogs (canis familiaris) in the pr esence of barriers. Applied Animal Behavior Science 84 299 317. doi: 10.1016/j.applenium.2004.03.004 Borkowski, J., Carr, M., & Pressely, M. (1987). "Spontaneous" strategy use: Perspectives from metacognitive theory. Intelligence, 11, 61 75. Brown, A. L. (1987). Metacognition, executive control, self regulation, and other more mysterious mechanisms. In F. E. Weinert & R. H. Kluwe (Eds.), Metacognition, motivation, and understanding (pp. 65 116). Hillsdale, New Jersey: Lawrence Erlbaum Associates. Call, J. & Carpenter, M. (2001). Do chimpanzees and children know what they have seen? Animal Cognition, 4 207 220. Commander, N. E., & Valeri Gold, M. (2001). The learning portfolio: A valuable tool for increasing metacognitive awareness. The Learning Assistan ce Review 6 (2), 5 18.
33 Chiang, L. H. (1998). Enhancing metacognitive skills through learning contracts. Paper presented at the annual meeting of the Mid Western Educational Research Association, Chicago. (ERIC Document Reproduction Services No. ED425 154). (2012, February 21). Dolphins deserve same rights as humans, say scientists. BBC World News Retrieved from http://www.bbc.co.uk/news/world 17116882. El awareness. Journ al of Developmental Education, 21 (2), 10 17 Farthing, G. W. (1992). The psychology of consciousness New Jersey: Prentice Hall. Flavell, J. H. (1979). Metacognition and cognitive monitoring: A new area of cognitive developmental inquiry. American Psycholog ist, 34 (10), 906 911. doi:10.1037/0003 066X.34.10.906 Foote, A. (2007). Metacognition in the Rat. Current Biology 17(6), 551 555. Hampton, R. R. (2001). Rhesus monkeys know when they remember. PNAS 98 (9), 5359 5362. doi: 10.1073/pnas.071600998 Hampton, R.R. (2009). Multiple demonstrations of metacognition in nonhumans: Converging evidence or multiple mechanisms? Comparative Cognition & Behavior Reviews, 4, 17 28. Retrieved from http://psyc.queensu.ca/ccbr/index.html Hampton, R. R., & Hampstead, B. M. (20 06). Spontaneous behavior of a rhesus monkey ( Macaca mulatta ) during memory tests suggests memory awareness. Behavioural Processes, 72 (2), 184 189. doi:10.1016/j.beproc.2006.01.007 Hampton, R. R., Zivin, A., & Murray, E. A. (2004). Rhesus monkeys (macaca mulatta] discriminate between knowing and not knowing and collect information as needed before acting. Animal Cognition, 7 (4), 239 246. doi:10.1007/s10071 004 0215 1
34 Harley, H. E., Fellner, W., & Stamper, M. A. (2010). Cognitive research with dolphins ( Tu rsiops truncatus ) at disney's the seas: A program for enrichment, science education, and conservation. International Journal of Comparative Psychology, 23 (3), 331 343. Herman, L. M. (2010). What laboratory research has told us about dolphin cognition. Int ernational Journal of Comparative Psychology, 23 (3), 310 330. Herman, L. M. (2012). Body and self in dolphins. Consciousness and Cognition: An International Journal, 21 (1), 526 545. doi:10.1016/j.concog.2011.10.005 Inman, A., & Shettleworth, S. J. (1999) Detecting metamemory in nonverbal subjects: A test with pigeons. Journal of Experimental Psychology: Animal Behavior Processes, 25 (3), 389 395. doi:10.1037/0097 7403.25.3.389 Kleitman, S. and Gibson, J. (2011). Metacognitive beliefs, self confidence and primary learning environment of sixth grade students. Learning and Individual Differences 21: 728 735 Kornell, N., Son, L. K., & Terrace, H. S. (2007). Transfer of metacognitive skills and hint seeking in monkeys. Psychological Science, 18 (1), 64 71. doi :10.1111/j.1467 9280.2007.01850.x Mann, J., & Smuts, B. B. (1998). Natal attraction: Allomaternal care and mother infant separations in wild bottlenose dolphins. Animal Behaviour, 55 (5), 1097 1113. doi:10.1006/anbe.1997.0637 Morgan, C. L. (1906). An intro duction to comparative psychology. London: Walter Scott Publishing.
35 Marino, L. (1998). A comparison of encephalization between odontocete cetaceans and anthropoid primates. Brain, Behavior and Evolution, 51 (4), 230 238. doi:10.1159/000006540 McKeachie, W. J. (1988). The need for study strategy training. In C. E. Weinstein, E. T. Goetz, & P. A. Alexander (Eds.), Learning and study strategies: Issues in assessment, instruction, and evaluation (pp. 3 9). New York: Academic Press. McMahon, S., Macpherson, K., & Roberts, W. A. (2010). Dogs choose a human informant: Metacognition in canines. Behavioural Processes, 85 (3), 293 298. doi:10.1016/j.beproc.2010.07.014 Mercado,E., I.,II, & DeLong, C. M. (2010). Dolphin cognition: Representations and processes in memor y and perception. International Journal of Comparative Psychology, 23 (3), 344 378. Metcalfe, J., & kober, H. (2005). Self reflective conciousness and the projectable self. Origins of self reflective conciousness 57 83. Nelson, T. O. (1996). Consciousness and metacognition. American Psychologist, 51 (2), 102 116. doi:10.1037/0003 066X.51.2.102 Nist, S. (1993). What the literature says about academic literacy. Georgia Journal of style="" Reading, (Fall Winter), 11 18. Roberts, W. A., Feeney, M. C., McMillan N., MacPherson, K., Musolino, E., & Petter, M. (2009). Do pigeons (columba livia) study for a test? Journal of Experimental Psychology: Animal Behavior Processes, 35 (2), 129 142. doi:10.1037/a0013722 Shettleworth, S. J. (2010). Cognition, evolution, and behavior (2nd ed.) New York, NY US: Oxford University Press, 243 249.
36 Schunk, D. H. (1991). Self efficacy and academic motivation. Educational Psychologist, 26, 207 232.1992 21550 00110.1207 Smith, J. (2009). The study of animal metacognition. Trends In Cognitive Sciences 13 (9), 389 396. doi:10.1016/j.tics.2009.06.009 Smith, J. D., Schull, J., Strote, J., McGee, K., Egnor, R., & Erb, L. (1995). The uncertain response in the bottlenosed dolphin (tursiops truncatus). Journal of Experimental Psychology: Ge neral, 124 (4), 391 408. doi:10.1037/0096 3418.104.22.1681 Sole, L. M., Shettleworth, S. J., & Bennett, P. J. (2003). Uncertainty in pigeons. Psychonomic Bulletin & Review, 10 (3), 738 745. Sternberg, R. J. (1984). What should intelligence tests test? Implic ations for a triarchic theory of intelligence for intelligence testing. Educational Researcher, 13 (1), 5 15. Sternberg, R. J. (1986). Inside intelligence. American Scientist, 74, 137 143. Sternberg, R. J. (1998). Metacognition, abilities, and developing e xpertise: What makes an expert student? Instructional Science, 26 (1 2), 127 140. doi:10.1023/A:1003096215103 Suda King, C. (2008). Do orangutans (Pongo pygmaeus) know when they do not remember?. Animal Cognition 11(1), 21 42. Tarski, A. and Vaught, R. 1 Compositio Mathematica 13: 81 102. Wells, R. S. (1993). In K.Pryor, K. S.Norris(Eds.), The role of long term study in understanding the social structure of a bottlenose dolphin community