Developing Intelligence

My friend Geoff once said that “all cognition is social.” Smugly, I reminded myself that the conclusions of cognitive psychologists are drawn on evidence where social cues are kept constant. But even in the absence of confounding social cues, perhaps the underlying cognitive processes themselves are caused by social factors.

A great example of this comes from today’s issue of Science, in which Topal et al describe how a well known “cognitive” phenomenon – perseveration – may be dramatically influenced by social cues.


Topel et al focus on a situation in which very young infants will mistakenly search for a hidden object in an old hiding location – as though they cannot maintain or update their representations of where the object has been most recently hidden. This is known as the AB error, since the toy can be found in location A, but not when that toy is hidden in B.

Previous work has shown a variety of interesting factors which contribute to the AB error. Yuko Munakata and others have shown that the direction in which children look can reveal sensitivity to the object’s new location, even when their reaching fails. Linda Smith and others have shown that the error is sensitive to spatial and motoric manipulations – when weights are placed on infants’ arms, or when their spatial perspective on the scene is altered while the toy’s hiding location is being changed, the children perform much better on the task. These and many other manipulations contribute to our knowledge about the nature of memory, “object permanence knowledge” and inhibitory skills.

But does this “error” reflect a cognitive failure on the part of children? Topal et al predicted that the AB error reflected an “interpretive bias” by which infants adopt a “pedagogical learning stance” based on the multiple social cues provided by the experimenter. Basically, to get infants to cooperate at all in the task, you need to make eye contact, direct their attention towards the first location, and generally scaffold them so that they will reach for the hidden object in the first place. The argument is that these social cues cause infants to infer generalizable properties about the hidden object: it’s as though the experimenter is saying “this kind of object usually hides over here.”

To test the idea, Topal et al run 14 infants each in three different versions of the classic AB task: in one condition, the experimenter established eye contact and verbally encouraged the infants to search for the object (the standard version of the task) whereas in another condition, the experimenter was oriented at a 90 degree angle from the infant and did not interact socially with the infant. A third “nonsocial” condition allowed infants to see only the experimental apparatus, as the experimenter was hiding behind a curtain.

The authors found that the standard condition yielded many more errors than the noncommunicative conditions. This was true both in parametric tests (regressing the proportion of errors on both trial type [A vs B location] and condition) and in a nonparametric chi-square for the number of children searching incorrectly in each condition.

The authors contend that the lingering bias to reach for the first, A-location is imbued by the social context in which the A-trials occur, and that in the absence of such social cues, infants merely search randomly.

While I won’t continue to brush off claims that all “cognition is inherently social,” I will note that there are some serious concerns here:

1) The authors arbitrarily define perseverative errors as more than one reach towards the incorrect location, which could be the “magic number” for providing a nice result in their chi-square analysis. We just can’t know without additional information about why this was chosen as the cutoff. And the parametric test is pretty unusual for this kind of data, which we can assume is at least somewhat non-gaussian and heteroskedastic between groups.

2) One could always claim that the reason their infants didn’t search perseveratively in location B is that they aren’t paying attention to the task – without the social cues from the experimenter which direct attention, the hiding locations might not be perceived as important. This could be particularly true in the later – and most critical – “B” phase of the experiment, once the novelty of the toy and the hide-and-seek procedure have begun to wear-off.

3) When running this task – and I’ve run around 200 infants on it – I’ve noticed that many infants very simply have a preference for one side or another of the apparatus. Usually the hiding locations are counterbalanced, of course, but with small sample sizes you always run the risk of getting infants which just “happen to like” or “dislike” location B in a particular condition. 14 infants is within that sample size, in my opinion. In this case, 9 infants appear to have made “mistakes” in the initial A trials, which could reflect reaching towards the B location (even in the absence of a toy being hidden there!) Critically, all of these infants were in the “nonstandard” conditions which showed improvements. I would prefer to see an analysis which includes only those infants that reached with 100% accuracy in the A trials, or at least one with a more consistent distribution of errors during A trials. Note that one would expect precisely this imbalanced distribution of errors on A trials if infants in the nonstandard conditions were not adequately encoding the hiding events, as I predicted in caveat #2 above.

So, maybe all cognition is social – but at least in this case, I’m going to need a little more data before I believe it.