Mixing Memory

I am deathly afraid of brown recluses like the one in the picture, the nasty little bugger. I never put on a pair of jeans without shaking them first, ’cause they like warm, dark places (and I know someone who was bitten by one hiding in her jeans), and when I recently found one in my medicine cabinet, I strongly considered moving. Other spiders? Who cares? Some of them are even kind of cute. But if you want to see me jump, show me a brown recluse in a jar (if you want to see me run, open the jar).

I tell you this to assure you of my deep, visceral appreciation of the need for an innate spider detection mechanism, even if I’m entirely agnostic about the existence of one. As a result of this appreciation, I was intrigued, if a little amused, when I saw that there is a brief article by Rakison and Derringer in press in the journal Cognition titled, “Do infants possess an evolved spider-detection mechanism?” (which you can read here, no subscription required).

The paper describes three “preferential looking” studies designed to test for the existence of the “Evolved spider-detection mechanism.” Preferential looking studies are based on a simple principle: infants tend to look longer at stuff that interests them. So if you show an infant a bright, noise-making toy on one side of a computer screen, and your tax returns on the other, chances are they’re going to look longer at the toy. It’s a simple, effective, and widely used paradigm in infant research. The idea, then, is that if infants have an “evolved spider-detection mechanism” (I just like the ring of that), they’re going to look longer at spiders than at other stuff. The problem, of course, is figuring out the relevant “other stuff,” and it turns out that this study may not covered all of the “other stuff” bases, but I’m getting ahead of myself.

i-4022d1e3f6359d64388e34413345dd2b-Rakison&DerringerFig1.JPG

In their first study, Rakison and Derringer showed infants between 4 1/2 and 5 1/2 months old the three figures pictured above (from their Figure 1a, p. 3) for up to two minutes (they stopped the trial if the infants looked away from the screen for 5 seconds prior to the two-minute mark), and timed how long they looked at each figure. In case the figure captions aren’t clear, the left-most figure is a (decidedly un-brown-recluse-like) “schematic spider,” the one in the middle is a spider reconfigured by turning the legs around, and the one on the right is a spider with all its features thrown about in a semi-random order (or ” completely scrambled,” as an egg, or a smushed brown recluse). And consistent with the hypothesis that infants have an innate spider detection mechanism, infants looked longer at the schematic spider (mean = 24.1 seconds) than at the “reconfigured” (m = 15.5 seconds) or “scrambled” (m = 17.3 seconds) spiders.

In their second study, they presented infants with similar figures, but this time the spiders’ bodies were rectangular (making them look like spiders from an old Atari 2600 video game), instead of like the curved bodies of actual spiders. Since in this experiment, even the “schematic spiders” didn’t look like real spiders, Rakison and Derringer predicted that infants would look at all three for about the same amount of time, and they did. This suggests it’s not just the configuration, but the spider-likeness of the figures in Experiment 1 that caused infants to look longer.

The third study used a slightly different method. As I said above, the preferential looking paradigm utilizes infants’ tendency to look longer at stuff they find interesting to determine what it is that infants find interesting. It turns out that if you have infants look at the same thing for a while, they get bored with it, and look longer at new stuff. This is called habituation, and it’s a great way to study things like infant categorization, to give one example. If you present infants with a bunch of different examples from the same category, and then with a new example from that category and an example from a different category (to which they haven’t been habituated), and they look longer at the latter, you can conclude that they’ve correctly categorized the new example of the habituated category. So habituation is a really powerful tool.

Rakison and Derringer utilized habituation by showing infants several examples of real spiders (in color photos, none of which were of brown recluses, but one of which was of a black widow… yuck), and then presenting them with the three figures used in their first experiment (the figures presented above). They argue that if, when presented with the schematic, reconfigured, and scrambled spiders after being habituated to real spiders, infants look longer at the reconfigured and scrambled spiders (implying that they consider the schematic spider old and boring now), it will suggest that they are generalizing from real spiders to the schematic spider. This in turn would imply that the infants have a “perceptual template” for spiders that serves the “evolved spider-detection mechanism.”

And that’s what they found. The infants only looked at the schematic spider for 6.7 seconds on average, while the mean looking times for the reconfigured and scrambled spiders were 12.1 and 13.4 seconds respectively. Thus the infants were able to generalize from the real spiders to the schematic one, and found it boring as a result.

Do infants have an “evolved spider-detection mechanism,” then? Eh. These studies are suggestive, but there’s a glaring alternative explanation. We already have a bunch of evidence that humans have an innate agency-detection mechanism (there are even books based on that research). If you look closely at the figures above, only one of them looks like it could really be an animate agent: the shematic spider. The reconfigured one might be a dead spider, and the scrambled one might be a spider on the bottom of someone’s shoe, but neither of them looks like it’s a possible agent. So it’s not really surprising that infants look longer at the a agent than the non-agents. So the study lacks the proper controls that would allow it to conclude that infants really do have an innate spider-detection mechanism. To draw that conclusion, you’d need to pit images of spiders (schematic or otherwise) against other, perhaps harmless agents (say, bunny rabbits, as long as they’re not from Caerbannog).

UPDATE: I neglected to mention a 4th study in the paper, which, in a comment, Wes Anderson suggests addresses my point about the lack of proper control. The fourth experiment is pretty much like the first, but instead of looking at spiders, infants looked at a schematic, reconfigured, and scrambled drawings of flowers. This experiment is meant to control for the possibility that infants are looking longer at schematic spiders because they are biologically plausible, while the reconfigured and scrambled spiders are not. Turns out, they look the same at all three, suggesting that it’s not just biological plausibility that causes them to look longer at the schematic spiders.

I left this study out, perhaps unwisely, because I think it’s utterly pointless. The proper control is another animate agent, not a flower (which is clearly not an animate agent), and if “biological plausibility” is a concern, then you can use a biologically plausible drawing of another type of agent. Until someone does a study with that control, all the present studies have shown is that infants find agents more interesting than non-agents, which is something we already knew.

Comments

#1 Derek James
October 3, 2007

I always feel like researchers trying to demonstrate an “innate” function are in an inherent Catch-22. By the time an infant is old enough to use this paradigm (in this case about 5 months old), how can you possibly control for their exposure to similar stimuli? In other words, isn’t it highly likely that 5 mo. olds have been exposed to insect/spider figures a great deal by this point (via toys, picture books, TV, on their pajamas, etc.)?

I read a study which used newborn chicks which were hatched in darkness, and their first exposure was to the experimental stimuli. They demonstrated a preference for biological motion over scrambled stimuli. But it’s unethical to carry out those kinds of controls with human infants, so in a sense, why even bother?
#2 Chris
October 3, 2007

Derek, good question. There are a lot of problems with “innateness” hypotheses, in large part because it’s so difficult to test them. But by using a bunch of different paradigms that produce results that ultimately converge to support a hypothesis, you can at least get a pretty good idea that something is, or is not, innate.
#3 Wes Anderson
October 3, 2007

Derek, the researchers asked the parents of the children if they had had frequent exposure to spiders, which included asking about toys, TV, films, and so forth. The parents reported that their children had not been exposed to spiders (or images of them) very often.

Chris, there are four studies (not three) and the fourth, adds a bit more of the proper control that you have criticized the overall research for not having. Instead of pitting images of spiders against bunny rabbits, they showed the children images of flowers. Instead of agency, they were looking to see if it was merely “biological plausibility” that triggered their visual preference for the schematic spider. They found that when looking at the flower images (schematic, reconfigured, and scrambled), the infants did not vary in their looking times, i.e., they looked equally long at all of the flowers, whether or not it was schematic, reconfigured, or scrambled. “The results of this experiment [the fourth experiment] suggest that infants� behavior in Experiment 1 did not result from a general bias to attend to biologically plausible stimuli.” Moreover, “[t]he data from Experiment 4 suggest that the visual preference in Experiment 1 did not result from prior experience with spiders; that is, 5-month-old infants in all likelihood have greater exposure to flowers than to spiders but nonetheless they did not look longer at a schematic flower than at scrambled versions of the same stimulus.” Perhaps further studies with rabbits and spiders and children (oh my!) would be highly beneficial nonetheless.

I have more difficulty with how they measured the time the children looked at the images and whether or not there was an order-effect in the experiments.
#4 knd
October 3, 2007

I also have trouble with the ultimate function: why is it that we have evolved an innate mechanism for spider detection? Do this really increase survival fitness (not to mention reproductive rate)? I know of no evidence for such claims (on the other hand, some controversial one has been gathered in favor of a snake-fear module in primates, see http://johnhawks.net/weblog/reviews/behavior/snakes_isbell_2006.w )

Despite my inclination to massive modularity hypotheses, I remain skeptical regarding the actual mechanism targetted in the paper and — in cauda venenum — I don’t really buy your agency-detection alternative neither, at least not before the mere effects of differences in spatial frequency content have been controlled.
#5 Chris
October 3, 2007

Wes, you’re right, I’d forgotten the flower study (wrote the post from my notes, and excluded that one as methodologically pointless), but honestly, that just makes it worse, because a flower is nothing like the control I’m looking for. Flowers aren’t animate agents, and generally aren’t mistaken for ones. In order to draw any conclusion about the spiders, you’d need the animate agent condition.
#6 plover
October 3, 2007

So it sounds like a generalised “biologically plausible” seems like too large and varied a category. Wouldn’t testing for other threats be a good approach? Use a rat (or whatever other small mammal was probably biting early hominids and causing disease) instead of a bunny. Also what happens with other arthropods: a centipede or cockroach or butterfly? What happens if you remove a few legs from the schematic spider?

From the descriptions in the post, I was imagining that something like “biologically plausible spider” might be the relevant category. Widening the scope of “biologically plausible x” to include flowers seems too wide, but there are obviously a lot of other possible scopes for which “biologically plausible spider” is a sub-category, and it seems likely there would be some degree of fuzziness for what “counts as” a spider.
#7 Greg
October 3, 2007

We need to be careful not to create a false dichotomy between evolved and learned. It is one question to understand what actually comes out of the box, so to speak, and another to examine what reliably develops. Many systems that unquestionably have some genetic component, and reliably developing design features, require some exposure to the relevant stimuli. In the case of something like spiders, we should expect learning to be minimized (those could be some costly learning trials), but even the most “innate” systems likely require some input (i.e., learning) in order to boot up properly. It might be wise to not concern ourselves too much with what innate means, and instead focus on what reliably develops. The developmental outcomes are what selection acts on. What is (or is not) needed to develop properly is a matter of empirical detail, not a factor in the determination of what is a product of natural selection.
#8 plover
October 4, 2007

Chris, I think I misread your original remarks on the agency-detector the first time through. My ideas were intended to point in more of less the same direction as what I think you’re saying. Though I’m also curious as to whether “agency-detector” is a concrete isolable phenomenon or is just the current black box that is a convenient description of the data. In other words, what are the parameters of this agency-detector? What can it do and not do? (E.g. if one of things it can not do is distinguish between the interestingness of a spider and that of a bunny, what (if anything, in the case of an infant) performs that function? And does that relate to any of this data?)

I hope I’m not too far off in left field here.
#9 Wes Anderson
October 4, 2007

I agree. Although I still think that the flower images were not completely pointless. Perhaps we can take that experiment as a step in the right direction-a place that further studies need to tend to-and for now we can remain agnostic.
#10 Nigel Thomas
October 4, 2007

Why do you think that the spider image “looks like it’s a possible agent”? Surely it looks like an agent because it looks like it’s a spider (or, at least, like some sort of bug). Frankly, it is not at all apparent to me that the “reconfigured spider” image (and even, with a bit of imaginative effort, the scrambled one) might not depict some sort of agent, although it is not be quite so easy to say what sort.

I am not quite sure what studies you are referring to on agency detection (the book link you give doesn’t help much) but I was under the impression that agency detection depended on movement: e.g., the fact that some blobs in a simple animation might seem agent-like because they are moving around freely, and perhaps seem to be doing things, such as pushing or “eating,” to other purely “passive’ blobs. If that is what you are talking about, it does not seem relevant to this study, which is concerned with what can be detected purely on the basis of shape, without movement cues. If, on the other hand, there are studies that claim to show that agents can be distinguished from non-agents purely on the basis of shape (I would be interested to see references), I would hazard a guess that these are almost certainly better interpreted as showing not that we have a generalized agent-from-shape detection mechanism, but rather that we have animal detection mechanisms (presumably a bunch of different ones for detecting different types of animals, such as people, mammals, snakes, spiders . . .).

[Spiders themselves, incidentally, certainly seem to have innate spider detection mechanisms. Some even have special eyes just for that purpose: Land, M. (1990), "Vision in other animals," in H. Barlow, C. Blakemore, & M. Weston-Smith (Eds.), Images and Understanding (pp. 197-212): Cambridge University Press.]
#11 A
October 4, 2007

I’m thinking that if given a choice between an iconic smilely face and a spider, infants would show a preference for the face. This would obviously prove that Rousseau was correct (and the Puritans wrong) – infants are innately more trusting than fearful.
#12 Caledonian
October 6, 2007

How would the infants know enough to understand that a ‘dead’-looking spider couldn’t be an agent?

You seem to be projecting your understanding onto them.
#13 Shallel
October 10, 2007

Back on Orion the Giant Spiders were really nasty
and kept humans as slaves and pets.

I think there is a genetic memory of these monsterous beings
in people of the Orion lineage.
#14 greensmile
October 10, 2007

Chris:
I come to the discussion of “spider detecting” modules from completely outside any related discipline…I know a little physics, math and comp sci. After reading articles such as the report on Cheney and Sayfarth’s work in NYTimes I am, pretty much on my own way to positing principles numbered 3 and 5 in Cosmides & Tooby’s exposition on evolutionary psych.

I may have it wrong but I think I have read in this blog some pretty strong critiques of ev. psych. When I read this post and the care with which you stated your doubts it left me wondering:
Have the basic tenets of ev. psych. actually carried the day and only particular claims come in for a tweaking or a beating [much as evolution, as a broader understanding of what biology presents the modern scientist, is treated]?
Or do you or others have a fundamentally different theory to account for the organization of feats [and the odd weaknesses] we call our brain power?
#15 Johannes
October 11, 2007

My spider sense is tingling!
#16 Danniel Soares
October 18, 2007

It would be somewhat ironical if humans had a “spider-sense” against spiders *forceful laughs*.

Two points I found somewhat noteworthy: I think that for any sort of instinct to avoid spiders or dangerous tiny animals in general (which is far more likely, I think), this “mental template” of a toilet-door-plaque-like icon would be somewhat “unuseful”. These animals are often in different angles and distances where they’re more like “potentially legged dots”, bringing no much resemblance to an icon. Thus could be somewhat harder to evolve such specificity instead of something more general (not that it would prove anything wrong anyway; could be that it arose somewhat as an evolutionary “saltation”, with no competing alternative version of instinct in the version I think that it could be more likely to evolve).

And, speaking of likelihood of evolving something, the author(s) suggests that it could have evolved in as few as about 20-30 generations? Has been some major change in the estimates of possible rate of adaptive evolution since Haldane’s estimates of substitution costs that I don’t know about? (about 300 generations to the fixation of a gene, under rather favorable conditions). I haven’t read the paper yet, but I’ve read that in a report.

A third point that came to my mind by reading this post, is that it seems to prove more that the babies are able to generalize the spider silhouette, which I don’t think that tells much about an innate ability to recognize and fear spiders. Perhaps one could show the same with cars, showing pictures of cars’ profiles, and icons of cars and scrambled “non-cars”, and that would then prove that we’ve already evolved the instinct of fearing to be hit by a car?
#17 Christopher
June 7, 2009

Well, this comment is late in coming…and probably won’t be read. That being said, I’d love the idea of an innate spider-detection system present in infants…I friggin’ hate spiders (after living in a room where I was killing 6-7 large spiders a day). But the study being examined proves nothing even remotely close to such an idea. For one, the study assumes that a living organism can be equated to and perceived as a motionless, featureless black blob of a specific shape. Even adults subconsciously respond to the presence of an unseen living organism more quickly than a motionless, visible shape (motion is a HUGE component of our survival instincts; every day life exemplifies it and military strategy utilizes it)…and that’s with the layers of unnatural crud an adult psyche is burdened with.

And two, what about the image presented as a schematic spider? To us, adults who have a clear mental image of what the form of a spider looks like, it appears as a spider. But an infant? And infant has most likely been exposed to human beings more than it has been exposed to any living creature. Which is, going by shape, a head, a body, two arms on top, and two legs on the bottom…symmetrical. The schematic spider follows the same formula, except its arms and legs are doubled….but in left/right arm, left/right leg segmentation, exactly the same. If you were to get rid of 4 of the legs and thicken the remaining 4, it would just look like the silhouette of a cartoonish, overweight individual. Though the image of the flower is not presented in this article, I would hazard to guess it’s overall shape is nowhere near as humanistic.

Simply put, I believe the infants were responding to a familiar shape…not the presence of an arachnid.