There's a very well-known experiment in developmental psychology called the "A-not-B task." The experiment goes something like this: you, the experimenter, are seated opposite a human infant. Within the reach of both you and the child are two boxes: box "A," and box "B." You hide a toy in "A," in full view of the infant. As expected, the infant reaches for "A" to retrieve the toy. You repeat the process several times. Each time you hide the toy in "A," and each time the infant reaches for "A" to find the toy. Experimental set-ups like this are extremely common in infant and animal studies. When trying to determine how a young baby - barely able to interact with you or the world - thinks about the world, you've really got two options: design an experiment that relies on the infant's ability to direct his or her eye gaze to a given location, or one that relies on the infant's ability to (somewhat clumsily) reach towards a given location.
So you keep hiding the toy in "A" and the baby keeps searching for the toy in "A." Simple enough. But what happens if you suddenly hide the toy in "B"? Remember, you're hiding the toy in full view of the infant. An older child or an adult would simply reach for "B" to retrieve the toy. But not the infant. But, despite having just seen the object hidden in the new "B" location, infants between 8 and 12 months of age (the age at which infants begin to have enough motor control to successfully reach for an object) frequently look for it under box "A," where it had previously been hidden. This effect, first demonstrated by Jean Piaget, is called the perseverative search error or sometimes the A-not-B error.
The A-not-B error is one of the most replicated findings in developmental psychology, but it seems almost as if for every time the experiment and results have been replicated, there has been a new explanation presented for why the error itself even occurs. Piaget thought that the existence of the object under A is causally related to the search response itself. That is, independent of where the object was hidden, searching under A would result in the object being found at A. In other words, he thought that the error reflected the immaturity of the child's understanding of object permanence, which is the understanding that objects continue to exist even after they become hidden. When a ball rolls under a couch, according to Piaget, an infant with incomplete object permanence would behave as if the object ceased to exist since it is no longer visible. Other more recent explanations have suggested that the infants are unable to inhibit a previously rewarded motor response, perhaps reflecting the immaturity of the prefrontal cortex, or that the error is due to limitations on working (short-term) memory. Others have suggested that infants are unable to switch their attention from location A to location B, presumably also due to underdeveloped executive functioning in the prefrontal cortex. This would reflect perseveration in attention rather than in search behavior, per se. Yet others have implicated the putative mirror neuron system to explain the error.
When experiments get replicated, even under the best of circumstances, there are minor tweaks in the way the experiment is conducted, like in the game telephone. What is especially remarkable, then, about the A-not-B error is that despite the methodological differences in conducting the experiment, the results are extremely consistent. Even if developmental and cognitive psychologists can't agree on why the error occurs, that it occurs is certain. But a finding without a solid explanation isn't particularly useful, is it?
One way to better determine the reason for the A-not-B error would be to find a way to break it. This isn't necessarily a ground-breaking idea: the common errors children make when learning to read, for example, tell us important things about the process of learning to read. When Dan Simons and Chris Chabris discovered inattentional blindness - essentially a breakdown in the attention system - that told researchers something really important about how attention works in the first place. So, are there any circumstances under which an infant under 12 months of age would reliably pass the test, and search for the toy at location B on the B trials?
A group of researchers from Hungary, including Hungarian rockstar developmental psychologists GyoÌrgy Gergely and Gergely Csibra (confusing, I know...), began by identifying the one feature of the experimental methodology that was certain to be included in every A-not-B experiment ever conducted: baby-talk. Even the most stodgy, grumpy, curmudeonly human adult can't help but use baby-talk (or, formally, infant-directed speech) when interacting with a baby. More generally, the task always occurs in a social-communicative context. When the experimenter hides the object each time in location A, it is accompanied by things like eye-contact, addressing the baby by name, looking back and forth between the infant and the hiding location, and baby-talk. These are called ostensive and referential signals. Is it possible that over fifty years of replicated results on the A-not-B error comes down, essentially, to baby-talk?
This isn't as far-fetched as it might sound. Previous findings have indicated that these ostensive-referential communicative signals could be interpreted by the infant in a special sort of way. Gergely and Csibra suggested that these sorts of social cues could activate within the infant what they called a "pedagogical learning stance." In other words, they could prepare the infant to go into "learning mode." Infants, after all, learn just about everything about their culture from their parents and other caretakers. But given how much sensory and social input they get, they must have a way of determining what social interactions are designed to confer generalizable information about the world as opposed to episodic information that relates only to the immediate situation.
Most researchers (Piaget included) have treated the hiding behaviors in the A-not-B experiment as indicative of episodic information about the location of the toy - something like, "the toy is under box A." But if researchers always provide ostensive-referential communicative cues while conducting the experiment, Gergely and Csibra, along with JoÌzsef TopaÌl, AÌdaÌm MikloÌsi, and AÌgnes ErdoÌhegyi, hypothesized that the infants are actually interpreting the information as generalizable rather than episodic - something like, "objects like this toy tend to be found under box A." If this is in fact the kind of information that the baby is receiving, then it is indeed reasonable for the baby to continue searching for the object at location A, even if she has seen it hidden at B. The hiding of the toy at location B is more like a statistical outlier, in this case. This would make the A-not-B error not an error at all, but a feature of the pedagogical learning system!
In order to test this, the team recreated the standard A-not-B task, but they added two important conditions. The standard condition included the ostensive-referential signals as usual: the researchers made eye-contact with the infant, greeted him or her by name, used infant-directed speech, and continually shifted her gaze from the infant to the hiding location and back. In the non-communicative condition, the researcher performed the experiment exactly as usual, but without any of the ostensive-referential cues. While all of her actions were completely visible, she never made eye-contact and never spoke to the baby. A third condition removed the social partner entirely from the set-up. The movement and hiding of the object was the same, but the experimenter conducted the entire experiment from behind a curtain. In each condition, there was a four-second delay between the hiding of the object, and the infant being permitted to search for the object. In each condition, the toy was first hidden four times at A, and then three times at B. Different ten-month-old babies participated in each different condition, with a total of fourteen per condition, for a total of 42 infant participants (these numbers are typical for infant studies).
If the pegagogy explanation for the A-not-B error is correct, then there should have been fewer errors made in the non-communicative and non-social conditions, compared with the standard ostensive-referential communicative condition. Indeed, this is what was found! In the ostensive-referential condition, infants succeeded in finding the toy on only on 14% of the B-trials (the black bars in the figure below). In other words, 86% of infants committed the standard A-not-B error in this condition, as expected. In the non-communicative condition, in which the experimenter was present but did not socially engage with the infants, far fewer of the infants made the A-not-B error. In this case, 57% of infants successfully found the toy in B trials. In the non-social condition, in which the social partner was entirely absent, 64% of infants successfully found the toy on the B trials. Not only was the proportion of successful B-trial searches statistically significantly higher in the non-communicative and non-social conditions, compared with the ostensive-referential condition, but in the non-social condition, there was no longer a statistical difference between the proportion of successful A-trials and successful B-trials!
The interpretation of these results is pretty straightforward: by removing the social-communicative cues and not talking to babies using baby-talk, Piaget's classic A-not-B error can be reduced or eliminated. To be clear, these findings are not consistent with the generally-accepted explanations for the error: lack of object permanence, reduced working memory, or the inability to inhibit previously learned motor responses. These findings are also inconsistent with the mirror neuron explanation, because both the ostensive-referential condition as well as the non-communicative condition included the same amount of visual and motor experience. Instead, the A-not-B error is not an error at all, but a fundamental feature of a cognitive system designed for the rapid learning of generalizable information.
The researchers are quick to point out that the four-second delay included in each condition of the experiment likely made it so that the infant was more likely to err. They write, "in our study, the perseverative error was reduced but did not completely disappear in the [non-communicative] and [non-social] contexts, which suggests that infants' search behavior also depends on their inhibitory, information processing, and memory skills." However, while these accounts can explain why infants might not search at the correct location, they do not explain why the infants perseverate at the wrong location. In other words, immaturity of inhibitory, information processing, or memory skills predicts random searches, rather than perseverative searches. It is specifically the perseveration of the infant at location A, then, that is due to the "pedagogical learning stance." The researchers frame this particularly elegantly:
Human infants are highly social creatures who cannot help but interpret the ostensive communicative signals directed to them. Although such a disposition prepares them to efficiently learn from adults, in certain situations (e.g., the A-not-B task) it can also misguide their performance.
The take-home message, it seems, is this: all those parents who incessantly babble at their babies in that singsongy baby-talk? That could actually a highly adaptive trait: the combination of adults' propensity towards baby-talk, combined with infants' interpretation of information provided in baby-talk as generalizable, allows infants to rapidly learn important information about their world and their culture from their caregivers.
TopÃ¡l J, Gergely G, MiklÃ³si A, Erdohegyi A, & Csibra G (2008). Infants' perseverative search errors are induced by pragmatic misinterpretation. Science (New York, N.Y.), 321 (5897), 1831-4 PMID: 18818358
Csibra, G., & Gergely, G. (2009). Natural pedagogy Trends in Cognitive Sciences, 13 (4), 148-153 DOI: 10.1016/j.tics.2009.01.005
Wow. That is a really clever experiment; I like the experimental designs where you have to sit back and go, hey, why didn't anyone think of doing that before?
I really wanna see the 'with and without 4 second delay' graph.
Strictly speaking, the 'inability to inhibit previously learned motor responses' isn't entirely ruled out- it would just mean that motor responses learned under social conditions get learned *differently* than motor responses learned under non-social conditions.
Which would be very cool.
All this is making me wonder if I need to reconsider the feedback I give my toddler during puzzle time. Unlike when he first got his puzzles, he almost never gets frustrated as long as I leave him alone. When I interact with him more, there are times he repeats the same (incorrect) action multiple times and gets very frustrated indeed. It could be that he's already sensitive to seeing himself as 'succeeding' in front of me (which is rather disturbing in a sense). It could be that when he does them alone, he can actually ask me for help and I might be inclined to help him, so when he would get frustrated he has a work-around. (whereas when I am sitting there watching him, I am inhibiting some desire to help him so he can do it himself, and he *knows* that *I* know he is getting frustrated and am doing nothing, and that itself is more frustrating).
I think there have also been times when I was trying to perform a motor activity that I had an audience for, and once I did it wrong I *kept* doing it wrong in the same way. I attributed this to self-defeating nervousness. And indeed, that's true in a sense. But it may be that it was this (likely innate) learning system getting hijacked as well.
very curious if you think that similar effects-- both the original A-not-B error and the social aspect-- might/do occur in dogs.
@Sam (#3): excellent question! check back on wednesday for the answer, in part 2 of this week's series of posts on pedagogy.
@becca (#2): right, the authors and I indicated that pedagogy doesn't explain everything. but lack of inhibition alone doesn't explain perseveration, specifically. as for doing the task without a short delay, the rate of success on B-trials would likely be significantly higher. the way the task is designed in the first place involves a brief delay.
Did they try it without the delay? Adult humans with known perseveration, in frontal lobe dementia for example, would be far far far less likely to perseverate after a 4 second delay.
Very interesting results though. Will be cool to see it repeated
Very interesting. Thank you!
I think we should replicate the experiment with Members of Congress.
One quibble: are the findings really not consistent with the generally accepted explanations? Consider that a human adult would have an intact pedagogical learning stance, yet we would still know that an object hidden under Box B must be under Box B, yes? There must be some cognitive skill that adults possess that takes precedence over the pedagogical learning stance and allows us to perform the task correctly. Therefore, one would think that it is this skill - whatever it is - that is not yet developed in infants.
This is interesting. Their findings are consistent with the recent work done by Bonawitz. She that showed when infants and children are showed how to complete a task (in this case, where the ball is) they follow direction of the "teacher", and stop thinking as much on their own to find the ball. The teacher baby-talked to the infant in the first scenario, the infant took the teacher's direction and saw the ball was in A, so when the teacher switched it to B, the infant just went back to A, without problem-solving. The infant was already "shown the correct answer", so at that point, problem-solving ceased. Then he made the error in determining where the ball is.
In the second scenario, the teacher was present, but not giving direction, so the infant had to pay attention more and solve it on his own, thus making fewer errors. More like a hands-off approach to teaching; present, but not intrusive.
In the third scenario, the infant was basically in an independent learning state, which is consistent with Bonawitz's claim that being given too much direction with teaching inhibits independent exploration and subsequent problem-solving.
Great find, and nice analysis, Jason!
On the surface, this seems (to me) to be a good argument for allowing your child to play independently for periods of time, as well as parent-child interactive play. While learning quickly (ie memorization) from baby talk/parental interactions is desirable for the large amount of information babies need to absorb, independent play seems to be equally beneficial for "higher level" learning.
Experimenter's cues are facilitating the "error", but are not the cause of it. After all, "error" is not completely gone in the third setting (no experimenter).
I would actually support the view that this is not an error at all. It is just a way of learning / generalising things. E.g. at first babies learn "object is in A" rather that "object is where experimenter puts it". So they just need some more examples for learning the "correct" pattern. Which seems to lead to the original Piagetian interpretation.
It would be interesting to see statistics about how many "B examples" reverse the pattern of looking for object in A, compared to the number of initial "A examples". Also, how babies act when presented a different initial pattern, e.g. A-B-A-B-A-B. Would they look into A and B in turns or just get lost?
WOW! Thanks a lot for posting the article! I have not heard of this research before.
Besides, here you can find a paper by GyÃ¶rgy Gergely, in which he answers some criticisms on their results: