Do ravens have a theory of mind? A new experiment suggests “yes”

May 16, 2016 • 9:00 am

A new paper in Nature Communications by Tomas Bugnhyar, Stepehen Reber, and Cameron Buckner (free download; reference below) examines the question of whether ravens, already known to be really smart, have a “theory of mind.”

What do we mean by that? Well, according to the authors, a “theory of mind’ means that animals can have a representation of an action performed by other animals, and then act on that representation rather than having to see the action itself. In other words, the animal imagines that another individual of its species is around, and then, without using visual cues, acts according to what it thinks the other animal is thinking and will do.

To be specific, in this case the researchers studied the behavior of ravens in caching food, and how that ability changed if the ravens thought another individual was around but couldn’t see it.

It’s been known for a while that when corvids, like ravens and scrub-jays, are caching food—burying it underground to hide it—their behavior changes if they know another individual is around. They may then re-cache it, burying it in another place so their food isn’t purloined. Other ways they protect their caches are hiding them behind obstacles, interrupting their caching when they see another bird, and staying away from already-cached food so as not to attract the competitor’s attention to the location.

In such cases, the birds (mostly scrub-jays in these studies) clearly are reacting to the stimulus of seeing another bird, and, through either learning or evolution, behaving in a way to hide their food, which is clearly an adaptive act.

According to the authors, previous experiments like this don’t really address whether the birds have a “theory of mind”, that is, an imaginative representation of what the other bird might be thinking, because they can actually see the other bird. As the authors say, these experiments suffer from the problem of being “unable to empirically distinguish representations of directly observable cues from a genuine representation of ‘seeing’.”

In this experiment, the authors tried to get around this problem by watching ravens cache when they didn’t see any competitor ravens, but thought they were around because raven noises were played. There was thus no chance that the ravens were acting on “gaze cues” sent by another bird. The authors did this by making peepholes between rooms represent the assumed presence of competitor ravens.

Birds were allowed to cache bits of cheese and dog food in one of two adjoining rooms. The rooms were connected by a window that could be closed, as well as a peephole that could be blocked. The ravens were also trained to know what the peepholes were for, as they learned to look through them and see a human caching food in the next room, and then were allowed to go into that room to recover the food. They thus learned that the holes were a way to see into the adjacent room, even though they couldn’t see through them unless their eyes were right up against those peepholes.

The initial predictions of the experimenter were, if ravens had a representation in their minds of what another raven in the next room would be thinking or would do, that:

  1. Ravens would cache their food more rapidly if the peephole were visible (and they could hear raven noises) than if the peephole were not visible. (Peepholes were obscured with a board.) That would mean that they carried the idea that another raven might be looking through the peephole at them, watching them cache. In such cases they’d finish up caching as soon as possible. They could not see a raven through the peephole, as there were none in the next room, but they could hear raven noises. Their caching behavior is expected to resemble the behaviors performed when the windows were open and they could see and hear a raven in the next room.
  2. In the same way, a visible peephole was predicted to make the ravens less likely to “improve” their cache: go back and recache or change the layout. That behavior calls attention to possible competitors (as noted above), and so a reduction in “improvement” means that the birds thought another raven could see them through the peephole.

This protocol, then, involves three experiments, as shown below: bird caching while seeing another raven observing it (a); bird caching with another bird in the next room that it couldn’t see, but could hear (b); and peephole with no bird in the next room (c). Raven noises were played to the caching bird in all three conditions. All ravens were hand-reared and, while trained in cognition studies, had no experience with caching, either with or without other ravens nearby.

(From paper): Figure 1 | Sketch of experimental setup. (a) Observed (Obs) condition: The cover of the window is open (white bar) and the focal subject (storer, st) caches food in the visual presence of a conspecific (observer). (b) Non-observed (Non) condition: The cover of the window is closed (grey bar) and the focal subject caches food in visual isolation of a conspecific (non-observer). Both observers and non-observers make sounds in the experimental chamber, which are audible to the storer. (c) Peephole (Peep) condition: The cover of the window is closed (grey bar) but one of the two peepholes (small white square) is open; the focal subject caches food in the absence of any behavioural cues, whereas the presence of conspecifics is simulated via playback of sounds recorded from non-observed trials (symbolized by loudspeaker).

To ensure that ravens weren’t operating because they’d seen another raven watch them before (even before they reacted to the peephole), trials b and c were given first. Birds were then exposed to a (open window) and then b and repeated.

The results are given in the graphs below, and confirm both of the authors’ predictions:

  • Ravens subject to the peepholes and windows (“peep” and “obs” respectively) cached for significantly shorter times than ravens who could not see the bird in the next room, even though raven noses were played. That’s shown in (a) below.  Conditions connected with a line show that those comparisons were either significant (*) or not significant (NS). The duration of caching was, on average, about 14 seconds when they couldn’t see the neighbor raven watching, but were significantly reduced to 7-8 second when either the peephole was visible or the window was open. There was no difference between the peephole and open-window conditions, suggesting that ravens construed the peephole as if it denoted “another raven is watching”.
  • As predicted, ravens “improved” their caches less often when the peephole was visible, or when the window showed another raven next door, than when the window was closed. Plot (b) shows this difference. Again, there is no difference between “open window” and “peephole” conditions.

Note that raven noises were played in all conditions, so the differences seen can’t simply be ascribed to hearing a competitor. No, there must be something about ravens thinking they’re being watched that explains the peephole results.

(From paper). Figure 2 | Effects of condition on caching behaviour. (a) Mean time to finish a cache and (b) mean number of revisits with improvements, in the non-observed condition (Non, total of 4 trials per 10 ravens), observed condition (Obs, total of 4 trials per 10 ravens) and peephole condition (Peep, total of 2 trials per 9 ravens). Box plots represent 25th and 75th percentiles, centre line indicates the median, whiskers represent non-outlier range and dots are outliers (Friedman test, post hoc Wilcoxon signed ranks test; *P < 0.05; NS = non-significant).

Does this show that ravens really do have a “theory of mind”? Well, it shows that ravens can act on the basis of assumed behaviors without having to see other animals. They think they’re being watched because they associate the peephole with vision, and so are acting on assumptions rather than observed behaviors. If that is a theory of mind, then ravens have one. Draw your own conclusions.

Let me add one more thing: the ravens altered their caching behaviors without ever having to suffer the loss of their cache to a competitor. Indeed, they altered their behaviors without ever having seen a competitor raven in the next room. To me this suggests that the caching behavior, and the way it changed under the assumption of a nearby competitor, are evolved rather than learned traits. These were naive ravens, and had no experience in nature before the experiment.

UPDATE:  A reader suggested that maybe evolution of this type isn’t involved, and that the ravens may be generalizing from their own behavior. That is, perhaps their behavior stemmed from watching through the peephole and seeing the humans cache food, and then going into that room and retrieving the food. They could then assume that if another raven is around, it would do what they did—retrieve food. That is still a theory of mind, but doesn’t necessarily involve evolved propensities to behave differently when you think a conspecific is around.

But how to determine if re-caching is evolved? I think one can expose naive ravens to another raven without any training. Having it cache food (presumably that is evolved), and see if it recaches if another raven is visible compared to the window open with no raven visible. It could be using behavioral cues, but those cues would have to act on an evolutionary program if they caused naive ravens to change their caching behavior.

Smarter than you think.


Bugnyar, T., S. A. Reber, and C. Buckner. 2016. Ravens attribute visual access to unseen competitors. Nat Commun 7. Article number:10506doi:10.1038/ncomms10506

48 thoughts on “Do ravens have a theory of mind? A new experiment suggests “yes”

  1. This may be a naive question, but how is this necessarily indicative of them having a theory of mind? Couldn’t the sound cues cause the same “instinctive behaviour” to manifest that the visual cue would have?

    1. The sounds cues were played in every treatment, so that cannot be the cause of their behavioral change.

    1. Perhaps the paper in question should have been entitled “Deep Into That Darkness Peering.”

      Merely this and nothing more.

  2. I have often wondered whether the reasoning ability of animals flickers on an off like a faulty fluorescent tube. We humans maintain a steady glow (although sometimes not that bright) but moving through the animal kingdom the flickering becomes increasingly rapid till at the microscopic level there is total darkness.
    Just a thought.

    1. Well, as an idea, it should be testable. But … on the assumption (and it is an assumption) that intelligence is generally beneficial to organisms, then you would expect there to be a “ratchet” that selects for individuals whose intelligence is more often on compared to others of the same species/ society. So you wouldn’t have the flickering state for long. There is probably an equation relating the degree of benefit to the number of generations for the population to become dominant in the population. Say, the new character gives a 1% increase in number of fertile offspring, then you might expect it to take ~100 generations to become the norm for the population. Call that 500 years for ravens. So the odds of actually observing that transition are pretty slender (I’m a geologist ; it shows.)
      Can you think of another character that comes and goes in a natural population in the way you describe?

      1. Interesting comment.
        No, I cannot think of such a characteristic but I do think that something like what is called “reversion to the mean” is a strong factor in evolution so today any such characteristic will be the consequence of perhaps considerable variation in previous generations which one for one reason or another have been dead ends and so overall there has been little “progress” and a transition has never be made.

        1. “reversion to the mean” doesn’t work like that. If you’re taking independent samples from a distribution, then the mean of your early set of samples will regress towards the (population) mean as you increase the number or size of samples.
          The concept isn’t relevant to generations where properties are controlled by discrete (binary ; other bases are theoretically possible) variables and particularly where one generation inherits it’s controlling variables from a preceding generation.
          True, Darwin did indeed worry about exactly this point. If only he’d actually had time to read the paper sent to him by Gregor Mendel, then this concern would have been put to bed permanently by about the 5th or 6th edition of Origin.

  3. Very interesting. The last point Jerry made is on point. That these ravens were hand reared and had no experience at caching does suggest instinct rather than reasoning. More studies needed! With both naive and experienced wild subjects.

    1. Huginn ok Muninn… (Odin’s two ravens, that bring him news every day, variously translated as ‘thought’ and ‘memory’, ‘mind’ and ‘desire’, etc):

      Mind and Heart/ make their flight every day, over the limitless lans./ For Mind I fear/ he may not come back,/ though I am more anxious for Heart.

      (Ursula Dronke’s translation.)

  4. If the Ravens hope to make it to the NFL playoffs and Super Bowl this season, they better develop a Theory of Mind real fast.

  5. I found this link:
    Which is a recap of some of the work Nicky Clayton has done in theory of mind with corvids. Clayton is a Professor of Comparative Cognition at Cambridge University and did earlier work at UC Davis. The article provides an overview of some of her research in corvid brains. According to Clayton some corvids think forward in time and show complex food stashing including stashing food in a cage in which it once went hungry. Interesting reading.

  6. Very good, although the authors are using a theory of mind that seems very stringent. I thought T of M applies if one knows that another of your species can see you, and so you change your behavior accordingly. It is cool, though, that ravens exhibit T of M if their information is in the abstract — hearing a conspecific nearby is enough to make them behave deceptively.

    1. The sounds were always present, so they weren’t using those alone. Even more impressive to me, they seem to reason that with the peephole present, another bird might be watching, since they themselves could watch through the peephole earlier.

    2. Well, I could argue that seeing another raven, and then losing your food, causes you to become more conscious of your cache when you see another bird, simply out of conditioning. No theory of mind is involved in that case–or at least anything I’d call a “theory of mind.”

  7. The experiment shows that the ravens act the same whether they know they are being watched or whether they could be being watched. That seems reasonable because in nature, presumably, they cannot know they aren’t being watched, if they know competitors are in the vicinity. What is curious to me is that in the Non case, the ravens “know” they have privacy even though they can hear other ravens.

    I am inclined to think ravens have a theory of mind, but I look forward to reading about follow up experiments.

  8. In some cases, I find it hard to see the line between evolved behaviour and learned behaviour. The idea that ravens may be generalizing from their own behaviour suggests that they are learning. But, if they are, what enables them to reflect on their own behaviour, to generalize it, and then to project it onto others? It must be, as the very last paragraph of the post suggests, “an evolutionary program”.

    1. Well said. I also find it difficult – as I find it difficult to see where precisely the line comes between ‘conditioned’ and ‘learned’ behaviour. Conditioning must involve a propensity to be conditioned – as Karl Popper pointed out in connexion with Pavlov’s experiments with dogs.

  9. Smarter than you think… very interesting. Science is finding more evidence every day that mental functions are shared with other creatures and are not unique to “created humans.”

    On Mon, May 16, 2016 at 7:01 AM, Why Evolution Is True wrote:

    > whyevolutionistrue posted: “A new paper in Nature Communications by Tomas > Bugnhyar, Stepehen Reber, and Cameron Buckner (free download; reference > below) examines the question of whether ravens, already known to be really > smart, have a “theory of mind.” What do we mean by that? We” >

  10. And now for some shameless anthopomorphizing!

    Certain birds seem pretty darned conscious/aware/intelligent to me — at least at the level of d*gs and cats.

    I have on numerous occasions in the mountains of western North America been alone with wild ravens. I have almost always been able to initiate and continue a “conversation” of vocalizations with them: Me mimicking the raven. It’s that chortling call they do.

    Often they would respond in other ways as well, orbiting me and doing rolls in response.

    And another kind of bird: African Gray Parrot. The parents of a friend had one at home (they were about the third set of owners). I visited and the parrot flew right over to me and started “talking” to me (in Flemish!) and walking back and forth on my arms and shoulders. This was very disconcerting to me, not knowing birds well at all (those are big, sharp bills!).

    They said I probably looked like a previous owner.

    When the friend’s Mom passed away, the bird went to yet another set of owners.

  11. I wonder whether the raven noises are relevant to the theory of mind behavior – after all, when the ravens were trained to know what the peephole was for, there were humans caching the food, not conspecifics. If the “storer” ravens had observed humans taking food from raven caches (not just caching food for the ravens to retrieve), would they change their caching behavior if only a human were visible in the window, or if no raven noises were used with the peephole? I guess this is an interspecies theory of mind question – there are other animals (squirrels, dogs, etc.) in the ravens’ environment that cache or bury food, and I’d assume that occasionally there’s interspecies cache-raiding.

  12. And here’s another comment just to put in a plug for your science posts.

    I really enjoy the science posts (but rarely have anything to add to them — I look at them more as opportunities for me to learn).

    You analysis of published work, and the comments on the thread, have helped me understand the work, and especially, the methods used and statistical analysis much better than I would otherwise have done. It has been more widely applicable to other reading as well.

    Plus you make us aware of interesting research that I likely would not have encountered otherwise.


  13. It seems to me this puts an upper limit on the ravens’ reasoning abilities. They see the unobstructed peephole and imagine another raven looking through it at them. But they fail to realize that if another raven were actually watching at the peephole, they wouldn’t be seeing daylight through it. So the visibility of the peephole actually means the opposite of what they think it means.

      1. Fair enough, but it also means that when they were doing the peeping, with their eye right up against the hole, they failed to imagine what that would have looked like from the POV of the person or bird being spied upon. Their theory of mind is incomplete in that sense.

  14. I once read a quote by the author, Tomson Highway, who described Raven as “right there laughing with you”. He was more right than he thought!

  15. Well, I guess that settles Poe’s question

    what this ominous bird of yore—
    What this grim, ungainly, ghastly, gaunt, and ominous bird of yore
    Meant in croaking “Nevermore.”


  16. “UPDATE: A reader suggested that maybe evolution of this type isn’t involved, and that the ravens may be generalizing from their own behavior. That is, perhaps their behavior stemmed from watching through the peephole and seeing the humans cache food, and then going into that room and retrieving the food. They could then assume that if another raven is around, it would do what they did—retrieve food.”

    That also seems quite plausible to me. It does imply that the ravens must be quite ‘self-centred’, as in ‘what would *I* do’. Being a self-centred human myself, I naturally think this applies to all other humans. Whether it naturally applies to any intelligence capable of reasoning I can’t say, but it seems likely. The rationale being, “I think this way, that thing looks exactly like me, therefore that thing will think the same way”.


  17. If I recall correctly, this kind of study has already been done on other corvids (I immediately think of jays, but I could be wrong). In that study / those studies, they found that the birds were more likely to re-cache those caches that are more visible to other birds (from an observation window iirc), and that only birds that themselves had in the past raided other birds’ caches had a tendency to re-cache.

    All of that seems pretty strong evidence that the behaviour of re-caching is dependent on experience and dependent on risk-estimation related to other birds’ potential actions (while basing those potentional actions on the mere possibility that other birds could be watching).

  18. I read this in my email yesterday… 🙂 just sayin’!

    My question – were the ravens related? i.e. is there any kin selection – more likely to share with family? Is there more sharing when there is plenty to go around? were they kept well-fed or a little hungry?

  19. Reminds me of Santino the deceptive and forward planning swedish chimp.


    Probably he needs some “theory of mind” to conceal his stones.

  20. Describing this behaviour as having some connection to Theory of Mind is way over the top. Many species have developed extremely complex competitive strategies in contests of survival. Evolutionary Game Theory describes and to a great degree justifies such evolved strategies. The question is – do these evolved strategies represent action of mind, or just elaborate evolved “mental programs” – mixed strategies as they are thus classified? In all but the case of humans I seriously doubt there is anything but instinct at work in these strategies. One can of course ascribe all sorts of sophisticated human-like attributes to these sophisticated strategies. Bacteria have been shown to act co-operativly in experiments/models of Prisoners Dilemma. Does this mean that bacteria have empathy? Rubbish!

Leave a Reply