Metacognition in Dogs: Do They Know What They Don't Know?

1. Introduction

1.1 The Question Behind the Question

There is a real difference between an animal that responds to the world and one that monitors its own knowledge of the world. The first only needs stimulus and response; the second needs some internal signal that says, in effect, "I don't have enough to go on here." If dogs have even a rudimentary version of that signal, it changes how we think about their learning and their confidence. If they don't — if what looks like self-monitoring is a trained habit — then the richer story collapses into ordinary associative learning. Keeping those two possibilities honestly in play is the whole discipline of this topic.

1.2 How to Read the Evidence

Metacognition research in animals carries a built-in skeptical default, and for good reason. Because a dog cannot report its doubt in words, every claim rests on a behavioral proxy — and almost every proxy can, in principle, be produced by simple associative learning without any inner monitoring. The mainstream position in comparative psychology is therefore cautious to the point of deflationary: the burden is on demonstrating that a behavior cannot be explained by learned rules, and that burden is rarely fully met. This article adopts that same caution. It treats suggestive findings as suggestive, not decisive, and it defaults to the leaner explanation until the richer one is earned.

1.3 Why This Question Is Worth Asking Carefully

Metacognition sits at a point where the evidence is thin and the intuition is strong. Anyone who lives with a dog has seen it hesitate, glance back, or search again, and each of those looks like an animal registering that it does not know.

Those impressions are not worthless — they are what motivated the experiments — and they are also exactly what the experiments were built to test rather than to confirm. Holding both at once is the whole difficulty of the topic.

1.4 What This Article Is Not About

Two adjacent questions are frequently mixed in. Whether dogs understand what a human knows is theory of mind, a separate literature with separate paradigms. Whether dogs recognize themselves is self-awareness, addressed later in this article and equally distinct.

What follows concerns one narrow question: whether a dog registers the state of its own information and acts on that registration (defining a capacity in measurable terms).

2. What Metacognition Is

2.1 The Components

"Thinking about thinking" is the slogan; in the lab it decomposes into several measurable abilities: uncertainty monitoring (recognizing when one is unsure), confidence-based decision-making (acting on how certain one feels), information-seeking (looking for clues when knowledge is incomplete), and error detection (noticing a mistake before any external feedback). Dogs are studied almost entirely through the third of these, information-seeking, because it produces a visible action.

2.2 The Proxy Problem

The core difficulty is that none of these internal states is directly observable in a non-verbal animal. Researchers must infer "uncertainty" from behavior — a check, a hesitation, a search — and any such behavior is open to a leaner reading. This is not a minor caveat; it is the central obstacle of the entire field, and it is why careful researchers hedge their language so heavily.

2.3 Monitoring and Control Are Separable

Metacognition is usually decomposed into monitoring — registering the state of one's own knowledge — and control, acting on that registration. An animal could in principle have the first without the second, and a behavioral task can only ever observe the second.

That asymmetry runs through everything here. Every result is a control behavior from which monitoring is inferred, and the inference is where the disagreement lives.

2.4 Why the Word Carries Baggage

In human psychology, metacognition arrived attached to conscious deliberation: knowing that you have forgotten a name, deciding to study a topic longer. Transferred to animals, the term keeps that connotation while the evidence supports only the functional core.

Some researchers accordingly prefer "uncertainty monitoring" or "information seeking", which describe the observable without importing the rest. This article uses the common term and means the narrow thing.

3. The Knowledge-Seeking Paradigm

3.1 The Logic

The dominant method is elegant. A reward is hidden in one of several locations. The subject either sees where it goes (the informed condition) or does not (the uninformed condition). Before choosing, it can seek more information — look through a gap, move closer. The question is whether behavior tracks the state of knowledge: does the animal check when uninformed and choose directly when informed?

3.2 What Would Count as Evidence

Strategic information-seeking that scales with uncertainty is the signature researchers look for. But the paradigm has a well-known vulnerability: a dog could learn the simple rule "if I didn't see the hiding, check first," which produces exactly the same behavior with no monitoring of any internal state. Distinguishing genuine uncertainty-tracking from this learned heuristic is the make-or-break problem, and it is the reason the same behavior can support opposite interpretations (much as with the associative-versus-causal debate in canine reasoning).

3.3 The Two Families of Task

Two designs dominate. In uncertainty-monitoring tasks, an animal faces a discrimination and may decline it — opting out of a hard trial for a small guaranteed reward. In information-seeking tasks, an animal may gather information before committing.

They test different things. Declining a trial requires assessing a judgment already available; seeking information requires recognizing that a judgment is not yet possible. The canine work is almost entirely of the second kind.

3.4 Why Opt-Out Tasks Are Hard With Dogs

Opt-out paradigms require an animal to learn that one response yields a small certain reward while another yields a large uncertain one, over many trials. That is a lot of training, and after it the animal's behavior is shaped by exactly the reinforcement history that the deflationary account invokes.

The more training a task needs, the weaker its evidence for anything beyond learning — a tension the field has not resolved in any species.

4. What Studies in Dogs Actually Show

4.1 The Central Study

The most direct test is Belger and Bräuer (2018). Dogs had to find a reward behind one of two V-shaped fences and could either commit to a fence or first check through a gap. The headline result was positive but modest: dogs checked more often when they had not seen the baiting than when they had, suggesting some sensitivity to their own perceptual access. But the study's own details temper that reading sharply. Dogs checked rarely overall. They did not show the "passport effect" — the flexible scaling seen in great apes, who check more when the stakes are higher — as dogs did not check more for high-value rewards or after longer delays. And crucially, dogs likely used their nose to check, meaning the behavior may reflect olfactory searching rather than the management of visual uncertainty. The authors themselves concluded that the results did not allow them to say whether dogs possess metacognition, only that there was some evidence for it.

4.2 The Earlier Null Result

This is not the field's only data point, and the other one cuts the other way. In an earlier study, dogs did not check more in the uninformed condition than the informed one, leading the authors to suggest that dogs might not have reliable access to their own perceptual and knowledge states at all (Bräuer, Call, & Tomasello, 2004). Together with the social-informant study described in 4.6, the canine record is one partial positive, one null and one mixed result — not the accumulating, one-directional evidence sometimes implied.

4.3 The Balance

Put together, the fair summary is that dogs show some tendency to gather more information when they lack it, that this tendency is weak, inconsistent across studies, and confounded by smell, that it appears more readily when the informative source is a person (see 4.6), and that it falls well short of the flexible, stakes-sensitive pattern that would make a strong metacognitive case. "Metacognitive-like behavior" is the most the current data support, and even that label is contested.

4.4 What "Checking" Has to Look Like

For a checking response to count, it has to be selective. A dog that peers through a gap on every trial is doing something, and it is not monitoring its knowledge; only a dog that peers more when it did not see the baiting is.

This is why the null results in this literature are about the difference between conditions rather than about whether dogs look at all.

4.5 Sample Sizes and What They Permit

The canine studies here involve modest numbers of animals across a handful of experiments. A null result from such a sample is compatible with a real effect that the study lacked the power to detect, and a positive result from one is fragile.

Neither direction has been established with the numbers that would settle it, which is worth stating before either is quoted with confidence. The honest summary of the canine literature is that it is small, that its results lean negative, and that a small literature leaning negative is a weak basis for any confident statement in either direction.

4.6 The Social Informant Study

A third canine study is worth reporting precisely because it is a borderline case. Across three experiments, the first found no evidence of information seeking in the narrow sense: the dogs did not move deliberately to obtain additional information about where the reward was (McMahon, Macpherson & Roberts, 2010).

In the second and third experiments the dogs preferred a human informant over a non-informing person, even where that choice brought no additional reward advantage. The authors read this as information-seeking behavior in a social context (McMahon, Macpherson & Roberts, 2010).

4.7 Why That Result Is Contested

The dispute is not about what the dogs did but about what it means. Dogs preferentially selected a human informant, which suggests context-dependent information seeking; whether this reflects metacognitive awareness or a learned reliance on human social cues remains unresolved.

The split runs along exactly the line this article draws. A dog that has learned, over a lifetime, that people indicate where things are will approach a person when it does not know something — without ever registering that it does not know. That reading is available for every result in this chapter, and it is strongest here, where the informative option is a human being.

4.8 What the Pattern Across the Three Experiments Suggests

Taken together, the study found information seeking where the source was a person and not in the classical spatial version of the task. That asymmetry is itself a finding, and it points in the same direction as the rest of the canine literature on social versus physical cognition.

It also makes the study a poor foundation for a claim about metacognition and a good illustration of why the question is hard to settle in this species.

5. The Deflationary Alternative

5.1 The Simpler Reading

The reason for caution is not mere fussiness; it is that a complete, unglamorous explanation is available. A dog that checks more when it did not witness the hiding may simply have learned the rule "when I didn't see, checking pays" — an ordinary product of associative learning that requires no introspection whatsoever. This is the standing critique of the whole animal-metacognition enterprise: the most widely used uncertainty tasks can, in general, be solved by simple associative mechanisms (Smith, Shields, & Washburn, 2003). On top of that, perceptual features of the setup — a visible barrier, the experimenter's posture, incidental cues — can drive "checking" independently of any inner state. Until a study rules these out, the deflationary account is not a fringe objection; it is the default hypothesis that the richer claim must defeat.

5.2 Why the Simpler Account Wins by Default

Where two explanations predict the same data, the one requiring fewer assumptions is preferred until something distinguishes them. That is a working rule rather than a claim about what is true, and it cuts against the richer interpretation here.

It also cuts against the reverse error. Preferring the simpler account is not the same as having shown that the richer one is false, and comparative cognition has a long history of treating the first as the second (where what was learned turns out not to be erased).

6. The Sign-Tracking Account

6.1 A Named Alternative

The deflationary reading described above is not merely a philosophical worry. It has been developed into a specific proposal in a review covering information-seeking across species, including dogs (Roberts et al., 2012).

Having a named alternative matters. A vague appeal to "associative learning" is hard to test; a specific mechanism makes predictions that can fail.

6.2 What Sign-Tracking Is

Sign-tracking describes an animal approaching and engaging with a stimulus that predicts reward, because the stimulus itself has acquired value through its association with the outcome. The animal is drawn to the predictor, not reasoning about what the predictor tells it.

Secondary sign-tracking extends this to a stimulus that predicts the availability of information which in turn predicts reward — a chain of associations, each of which is ordinary learning.

6.3 The Proposal

The review suggests that the choice of information in pigeon, rat and dog experiments may be a form of secondary sign-tracking, and therefore different in kind from the metacognition-like processes attributed to primates (Roberts et al., 2012).

That is a substantive claim about which mechanism is operating, and it explains the same behavior without any monitoring of internal states.

6.4 Why It Fits the Canine Data

The proposal predicts that an animal will make the informative choice when forced to choose between an informative and an uninformative option, because one has become the better predictor. It does not predict spontaneous checking when no such choice is offered, because nothing has been established as a predictor.

That is close to the pattern the dog studies report, which is what makes the account uncomfortable rather than merely available.

6.5 What Would Distinguish the Two

Sign-tracking and genuine uncertainty monitoring diverge on transfer. An animal tracking a learned predictor should need to learn it again in a new configuration; an animal monitoring its own knowledge should check whenever it lacks information, whatever the setup.

Testing dogs on a novel arrangement after training on another is therefore the informative experiment, and it is not what most of these studies have done. The reason is practical rather than negligent: transfer designs need twice the sessions, they lose animals to attrition, and a null result in the transfer phase is hard to interpret when the training phase already produced one.

6.6 Holding Both Accounts at Once

Nothing here shows that dogs lack metacognition. It shows that a simpler mechanism accounts for the observations, which shifts the burden onto anyone claiming the richer one.

That is the ordinary state of comparative cognition rather than a special problem with dogs, and it is why the field moves slowly. Every capacity attributed to a non-verbal animal has to survive the same challenge, and most of the progress in the last thirty years has come from designing tasks that the simpler account cannot pass rather than from finding animals that perform better.

7. Why Primates Are the Comparison

7.1 The Benchmark Behavior

The reason the primate literature keeps appearing in canine metacognition research is that it established what the target behavior looks like. Apes and monkeys will look down opaque tubes or under opaque containers to locate a hidden reward, and they make those looking responses less often when other information already indicates the location — visible baiting, transparent containers, or logical inference (Roberts et al., 2012).

That contrast is the crux. Looking is not impressive on its own; looking selectively when one lacks information is.

7.2 What Non-Primates Do Instead

Studies of information seeking in pigeons, rats and dogs indicate that they do not readily show the looking responses seen in primates (Roberts et al., 2012).

Three species, three research traditions, the same absence. That consistency is more informative than any single null result — and it remains an absence in a particular kind of test, not a demonstrated absence of the capacity.

7.3 The Forced-Choice Result

The picture is not uniformly negative. Given a forced choice between stimuli that do and do not yield information about the location of a reward, these non-primates make the informative choice (Roberts et al., 2012).

Dogs will take information when it is offered as an option. What they do not reliably do is generate the search themselves.

That distinction is easy to lose in a headline. A study reporting that dogs choose the informative option is accurately described as showing that dogs prefer information, and inaccurately described as showing that dogs know when they do not know.

7.4 Why That Distinction Carries the Argument

Choosing the informative option when it is one of two available responses is exactly what a value-based account predicts. Spontaneously interrupting a task to gather information nobody offered is harder to explain that way.

The canine evidence sits on the first side of that line, and the popular version of this topic sits on the second.

7.5 Not a Ranking of Species

It would be easy to read this as dogs coming off worse than monkeys, and that reading does not follow. The paradigms were designed around a primate behavior, in a modality primates rely on, using an apparatus primates manipulate readily.

A dog that does not look down a tube may be a dog for whom looking down tubes is not the natural way to resolve uncertainty. Whether an olfactory version of the paradigm would produce different results is untested and obvious.

8. Metacognition vs. Self-Awareness

8.1 Two Different Claims

A distinction prevents a common overreach. Even the strongest possible reading of these findings would not imply that dogs possess human-like reflective consciousness or an abstract self-concept. The abilities in question — monitoring uncertainty, seeking information, perhaps detecting error — do not require that lofty machinery. They would require only a functional, low-level monitoring system that helps an organism act efficiently under incomplete information. So the interesting hypothesis is modest by design: not that a dog ponders its own ignorance, but that its behavior can be regulated by the state of its information. Conflating that with self-awareness inflates the claim far beyond anything the evidence could carry.

8.2 Why They Get Conflated

Both claims involve the word "self", and both are attractive for the same reason: they promise that the dog has an inner life resembling ours. Neither the evidence nor the concepts support treating them as one question.

Monitoring whether one has enough information to act is a narrow, functional capacity that many systems could have. Awareness of oneself as a subject is a far larger claim with a separate and equally contested literature.

9. Neural Considerations

9.1 The Neural Picture

In humans, metacognitive monitoring and error detection are associated with the prefrontal cortex and the anterior cingulate cortex (Fleming & Dolan, 2012). The canine prefrontal cortex is present and functional but proportionally smaller and less elaborated than the human one, and there is no direct neural evidence for metacognition in dogs. Awake-fMRI work has shown reward-related activity in response to hand signals and to praise or food (Berns et al., 2012; Cook et al., 2016), but it is important to state plainly what this does and does not establish: it demonstrates that reward-related regions respond to learned signals and preferred outcomes, not that dogs engage in metacognition. The imaging shows the hardware could in principle support flexible monitoring (how the dog brain underpins behavior more broadly); it does not show the software running. The prefrontal contribution to self-control and flexible decision-making is itself an active canine research question (the prefrontal cortex and canine self-control).

9.2 What Awake Canine Imaging Has and Has Not Done

Awake unrestrained canine fMRI is established (Berns et al., 2012) and has produced real findings about reward and social preference (Cook et al., 2016). None of that work addresses metacognition.

The prefrontal correlates of metacognitive accuracy come from human imaging (Fleming & Dolan, 2012), and applying them here is an extrapolation across species and across tasks. No canine study has related brain activity to performance on an information-seeking task.

9.3 Why the Imaging Route Is Harder Here Than It Looks

Even a well-funded attempt would face a problem the human work does not. Human metacognition studies rely on confidence reports, which is precisely the response a dog cannot give. Without that, there is no metacognitive judgment to correlate brain activity with.

The imaging would have to be paired with a behavioral measure that already settles the question, which is the question the field has not settled. Neuroimaging can show where something happens; it cannot establish what that something is when the behavioral criterion is itself in dispute.

10. What Would Settle It

10.1 Transfer to a Novel Setup

The single most useful test is transfer. Train a dog on one information-seeking arrangement, then present a structurally different one in which the same logic applies but no learned predictor does.

Uncertainty monitoring predicts immediate appropriate behavior; sign-tracking predicts starting over. Few canine studies have run this comparison.

10.2 Graded Uncertainty

A second approach varies how uncertain the animal should be rather than treating knowledge as present or absent. If checking behavior scales with the degree of uncertainty, that is harder to produce with a single learned association.

Doing this requires a task where difficulty can be titrated, which is straightforward with a perceptual discrimination and awkward with hidden food. Most canine paradigms use hidden food because it needs no training, and that convenience is precisely what rules out the graded version.

10.3 Cost Manipulations

Making information costly — a delay, a detour, a smaller reward — tests whether an animal seeks it in proportion to how much it needs it. An animal that pays the same cost regardless of whether it already knows is not monitoring anything.

This is the design that produced some of the more convincing primate results and it has barely been attempted in dogs. Cost is also the variable most easily arranged in an ordinary training hall, which makes its absence from the canine literature harder to explain than the absence of the imaging work.

10.4 Using the Dog's Own Modality

Every paradigm discussed here is visual. Dogs resolve most natural uncertainty by sniffing, and a task in which the informative response is olfactory rather than visual would test the same question in a form the species is built for.

That no such study exists is the clearest single gap in this literature. Such a study is not technically difficult to build; the obstacle is conceptual: a task in which sniffing resolves the uncertainty makes the informative response so easy that distinguishing knowledge monitoring from ordinary searching becomes the new problem. That is a real difficulty and it is a different one from the difficulty the field currently has.

10.5 Reporting Individuals

Group-level results in this area conceal wide variation, and a paradigm this demanding is likely to produce a minority of animals behaving one way and a majority another.

Reporting the distribution rather than the mean would tell readers whether they are looking at a species-typical capacity or at a handful of unusual dogs. It would also make replication meaningful. A study reporting that four of sixteen dogs behaved selectively is a different finding from one reporting a small group effect, and the two are currently indistinguishable in most write-ups.

10.6 What a Positive Result Would Have to Survive

Suppose a canine study reported selective checking. Before that counted as metacognition it would have to survive several ordinary objections: that the checking response had been differentially reinforced, that the conditions differed in difficulty rather than in the dog's knowledge, that the experimenter's behavior differed between conditions, and that the effect was carried by a few individuals.

Those controls are demanding and routine. Their absence is the usual reason a promising result in this area does not persuade the field, in dogs as in every other species.

11. Research Gaps and Methodological Challenges

The limitations here are severe enough that they define the state of the field rather than merely qualifying it.

The associative confound. The central problem — learned "checking" rules mimicking genuine uncertainty-monitoring — has not been decisively ruled out in dogs, and arguably cannot be with the current paradigms.

Sensory confounds. In dogs specifically, olfaction contaminates visual-checking tasks: a dog may "check" by smelling, which is not the visual metacognition the task was designed to probe.

A tiny, inconsistent literature. The canine evidence rests on very few studies with small samples and pointing in different directions — one partial positive, one null, and one mixed result that found information seeking only toward a human informant (McMahon, Macpherson & Roberts, 2010).

No neural signature. There is no direct neural marker of metacognition in dogs; the imaging establishes plausible substrates, not the process.

Definitional disagreement. Comparative psychologists do not agree on what would even count as demonstrating metacognition in a non-verbal animal, which makes any single study hard to interpret (a recurring measurement problem across canine cognition).

A named alternative accounts for the data. Secondary sign-tracking explains the forced-choice results without any monitoring of internal states (Roberts et al., 2012), and no canine study has been designed to distinguish the two accounts.

Every paradigm is visual. Dogs resolve uncertainty by sniffing, and no olfactory information-seeking task has been run. A null result in the wrong modality is weak evidence about the capacity.

Transfer has not been tested. The decisive comparison — training on one arrangement and testing on a structurally novel one — is straightforward and largely absent from the canine literature.

No canine imaging bears on this. The neural account is drawn from human work on confidence judgments (Fleming & Dolan, 2012), a response dogs cannot produce.

12. Practical Implications

These implications are worth drawing out, but with an explicit "if": they follow if dogs have even a functional capacity to register uncertainty, which is unproven. Read conditionally, they still point toward good practice.

12.1 Training as Collaboration

If a dog can register when it is unsure, then training is less a one-way conditioning process and more a two-way exchange: an uncertain dog may look to its human for guidance (drawing on its strong reading of human cues), raise its attention, or adjust its strategy in real time. Treating the dog as an active problem-solver, rather than a passive recipient of signals, is sound practice whether or not it turns out to be literally "aware" of needing help.

12.2 Stress, Uncertainty, and Rigidity

There is a firmer link worth making. In the species in which it has been studied, chronic stress impairs prefrontal function and cognitive flexibility, and a nervous system flooded with stress hormones is worse at evaluating information, adapting decisions, and tolerating ambiguity (the neurobiology of chronic stress). An anxious dog that seems "stubborn" or cognitively rigid may be working with compromised monitoring systems (which is closer to anxiety than to defiance). Whatever the truth about metacognition, the evidence from other species links stress and flexible cognition.

12.3 Building Confidence — and Not Punishing Uncertainty

If uncertainty can be registered, confidence is partly a cognitive state, not only an emotional one — which argues for letting a dog explore and gather information at its own pace, and rewarding effort rather than only success. The sharpest practical point is a caution: punishing a hesitant or "unsure" response may do more than suppress that action. It can suppress the dog's willingness to engage and problem-solve at all, pushing it toward a passive, helpless state — one more reason the costs of aversive methods run deeper than the immediate behavior they target.

12.4 Why the Answer Does Not Change the Training

It is worth being clear that nothing practical hangs on how this question resolves. A dog that hesitates at an ambiguous cue should be given clearer information whether or not it is monitoring its own uncertainty, and a dog punished for hesitating learns that hesitation is dangerous either way.

The value of the question is in what it teaches about reading evidence, not in what it changes on the training field. That is a legitimate reason to write about it and a poor reason to overstate it.

It is also worth saying what a negative answer would not license. A dog that does not monitor its own knowledge is not thereby a dog whose hesitation can be ignored, or whose confusion is a training failure. The behavior is informative regardless of what generates it.

13. Summary at a Glance

The benchmark behavior is selective looking — Apes and monkeys look into hidden locations and do so less often when other information already indicates the answer (Roberts et al., 2012).

Dogs have not reliably demonstrated it in current paradigms — Information-seeking studies in pigeons, rats and dogs indicate they do not readily produce the looking responses seen in primates (Roberts et al., 2012).

They do take information when offered — Given a forced choice between informative and uninformative stimuli, these non-primates choose the informative one (Roberts et al., 2012).

A named alternative explains that — The pattern may reflect secondary sign-tracking rather than the metacognition-like processes attributed to primates (Roberts et al., 2012).

The canine evidence is thin on both sides — A small number of studies, mostly with modest samples, and no strong positive result that survives the simpler reading.

Absence of evidence is not evidence of absence — The paradigms were designed around a primate behavior in a primate modality, and no olfactory version has been tried.

Transfer is the decisive test — Uncertainty monitoring should generalize to a novel arrangement; a learned predictor should not.

None of this bears on how dogs should be trained — The practical recommendations in this article follow from what dogs demonstrably do, not from whether they monitor their own knowledge.

14. Conclusion

Do dogs know what they don't know? The honest answer is possibly, but the evidence currently supports only a limited functional interpretation. In the central study dogs checked more when they had not seen the baiting, but they checked rarely, did not scale checking with the stakes, and probably used their nose (Belger & Bräuer, 2018); an earlier study found no difference at all (Bräuer, Call, & Tomasello, 2004); and a third found information seeking only where the informant was a person (McMahon, Macpherson & Roberts, 2010). Across non-primates, the selective looking seen in apes and monkeys has not been reliably shown, while the preference for an informative option when one is offered is explained just as well by secondary sign-tracking (Roberts et al., 2012). The effect is therefore weak, inconsistent, confounded by the dog's own nose, and fully open to a deflationary reading in which a learned "check when I did not see" rule does all the work without any inner monitoring (Smith, Shields, & Washburn, 2003). What the evidence does not support is the confident claim that dogs introspect on their ignorance; what it does not rule out is a capacity that visual, primate-designed paradigms have failed to reach, since neither transfer tests nor olfactory versions have been run. The defensible position is that dogs are adaptive, dynamic cognitive systems rather than simple stimulus–response machines, that they sometimes act as if they are unsure (which is a statement about behavior, not a proven statement about inner experience), and that the deeper question remains genuinely open. That uncertainty is not a weakness of the article; it is the current state of the science, stated plainly.

Key Insights (Takeaways)

• Metacognition is the monitoring of one's own knowledge — registering uncertainty and acting to reduce it. In dogs it is among the least established high-level cognitive claims, and the field's default is skeptical: nearly every behavioral proxy can, in principle, be produced by ordinary associative learning.

• The central dog study (Belger & Bräuer, 2018) found dogs checked more when they had not seen the baiting — but they checked rarely, showed no "passport effect" (no more checking for high-value rewards or longer delays), and probably used smell to check. The authors drew no definitive conclusion. An earlier study (Bräuer et al., 2004) found no checking difference at all. A third study found information seeking only toward a human informant (McMahon, Macpherson & Roberts, 2010). The record is one partial positive, one null and one mixed result, not accumulating evidence.

• The deflationary alternative is the default that any strong claim must defeat: a learned rule ("when I didn't see it, check") reproduces the behavior with no introspection, and the standard uncertainty tasks are known to be solvable by associative mechanisms (Smith et al., 2003), and secondary sign-tracking has been proposed as the specific mechanism behind the forced-choice results in dogs and other non-primates (Roberts et al., 2012).

• Even the strongest reading would mean only functional monitoring, not human-like self-awareness. There is no direct neural evidence for canine metacognition; awake-fMRI (Berns et al., 2012; Cook et al., 2016) shows reward-related responses to signals and outcomes, not that metacognition is occurring.

• Read conditionally, the practical lessons still hold: treat the dog as an active problem-solver, protect flexible cognition by reducing chronic stress, and never punish hesitation or uncertainty — doing so can suppress engagement itself and push a dog toward learned helplessness, whatever the truth about metacognition turns out to be.

References

Belger, J., & Bräuer, J. (2018). Metacognition in dogs: Do dogs know they could be wrong? Learning & Behavior, 46(4), 398–413. https://doi.org/10.3758/s13420-018-0367-5

Berns, G. S., Brooks, A. M., & Spivak, M. (2012). Functional MRI in awake unrestrained dogs. PLOS ONE, 7(5), e38027. https://doi.org/10.1371/journal.pone.0038027

Bräuer, J., Call, J., & Tomasello, M. (2004). Visual perspective taking in dogs (Canis familiaris) in the presence of barriers. Applied Animal Behaviour Science, 88(3–4), 299–317. https://doi.org/10.1016/j.applanim.2004.03.004

Cook, P. F., Prichard, A., Spivak, M., & Berns, G. S. (2016). Awake canine fMRI predicts dogs' preference for praise vs. food. Social Cognitive and Affective Neuroscience, 11(12), 1853–1862. https://doi.org/10.1093/scan/nsw102

Fleming, S. M., & Dolan, R. J. (2012). The neural basis of metacognitive ability. Philosophical Transactions of the Royal Society B: Biological Sciences, 367(1594), 1338–1349. https://doi.org/10.1098/rstb.2011.0417

McMahon, S., Macpherson, K., & Roberts, W. A. (2010). Dogs choose a human informant: Metacognition in canines. Behavioural Processes, 85(3), 293–298. https://doi.org/10.1016/j.beproc.2010.07.014

Roberts, W. A., McMillan, N., Musolino, E., & Cole, M. (2012). Information seeking in animals: Metacognition? Comparative Cognition & Behavior Reviews, 7, 85–109. https://doi.org/10.3819/ccbr.2012.70005

Smith, J. D., Shields, W. E., & Washburn, D. A. (2003). The comparative psychology of uncertainty monitoring and metacognition. Behavioral and Brain Sciences, 26(3), 317–339. https://doi.org/10.1017/S0140525X03000086