[Originally posted on February 20, 2006]
Here's a picture of our daughter Nora at about 3 months of age. She looks like she's fairly aware of the events going on around her (arguably more aware than she sometimes appears now, at age 12). However, as our knowledge of how infants begin to perceive the world around them has increased, we've learned that the world of a three-month-old literally looks different to them than the world we perceive as adults. That's because vision, which seems so obvious and instinctive, is actually an active process. When we perceive the world visually, we're not just passively "seeing" what's there, we're constantly determining where one object ends and the next one begins. We're applying logical rules to help break objects into groups and understand how the two-dimensional image on the inside of our eye corresponds to a three-dimensional physical world.
In the picture of Nora, for example, how do we know that the bonnet isn't part of her body? Because it's a different color, white? But the white buckle is part of the baby carrier. Clearly the set of rules we've learned are not simple. But when do we learn them? And in what order?
Yuyan Luo and Renée Baillargeon developed a set of experiments to test just one aspect of infant perception: how babies recognize when an object moves behind an occluder. They constructed a small "stage set" to test a variety of different types of occlusion. The simplest event involves an object moving behind a screen. Lou and Baillargeon followed up on experiments by Andrea Aguiar and Baillargeon which found that 2.5-month-old babies will not be surprised by an object moving disappearing when it passes behind a small screen and reappearing when it emerges from the other side. However, if you cut a hole in the bottom of the screen and repeat the experiment, 2.5 months old infants are still not surprised when an object disappears behind the screen and reappears on the other side -- "magically" moving past the window without being seen. By the time they are 3 months old, they are surprised when they don't see the object in the window in the middle of the occluder. But three-month-olds aren't surprised in a separate experiment when the middle of the occluding screen is made shorter and a tall object is passed behind the occluder, again magically moving past the window unseen. It's not until 3.5 months old that this trick is surprising.
Aguiar and Baillargeon developed the following flowchart to show how infants perceive objects passing behind occluders:
When infants are 2.5 months old, they have only learned the first step: if the object is behind the occluder, it will be hidden (blue boxes indicate when babies believe objects will be visible). By 3.0 months of age, they have learned another idea: when the lower edge of the occluder is interrupted, they should be able to see an object behind it. It is not until 3.5 months of age when infants expect that they will be able to see tall objects behind a short occluder.
But other researchers suggested that there might be flaws in Aguiar and Baillargeon's system. Perhaps babies don't develop rules, but instead file "videotapes" of events they have seen before, and if they see a new event that appears to match the "tape," they aren't surprised. Younger babies haven't had as many experiences, so they probably just haven't seen the more complicated occluders before. Another explanation suggested that babies were just focusing on different parts of the objects as they passed behind the occluders.
Luo and Baillargeon's new experiments addressed these concerns by presenting the opposite scenario. Instead of testing whether infants were surprised at impossible events, they tested for surprise at possible events. They worked with 3-month-old infants, which Aguiar and Baillargeon had previously found were surprised when an object did not appear in a window at the bottom of an occluder, but weren't surprised when an object did not appear in the window at the top of an occluder. This time they showed two events -- the impossible event from Aguiar and Bailargeon's study, and the possible event, where the object did appear in the window as it passed behind the occluder. Here are the results:
The light blue bars correspond to the original Aguiar and Baillargeon study: as before, 3-month-olds are surprised (measured by looking at the display longer [see this article for an explanation of this method]) when an object does not appear in the window at the bottom, and not surprised when it doesn't appear in the window at the top. The new data tested the reverse: when the object does appear. Now infants were not surprised when the object appeared in the window at the bottom of the occluder, but were surprised when the object appeared in the window at the top of the occluder. In a similar experiment, they found a corresponding effect for 2.5 month-olds.
Luo and Baillargeon argue that these results offer additional support for the model that Aguiar and Baillargeon developed in the original experiments: as infants get older, they learn more and more rules about how objects behave in relation to one another, which allows them to develop a more sophisticated representation of the physical world.
Aguiar, A., & Baillargeon, R. (2002). Developments in young infants' reasoning about occluded objects. Cognitive Psychology, 45(2), 267-336. DOI: 10.1016/S0010-0285(02)00005-1
Luo, Y., Baillargeon, R. (2005). When the ordinary seems unexpected: evidence for incremental physical knowledge in young infants. Cognition, 95(3), 297-328. DOI: 10.1016/j.cognition.2004.01.010
- Log in to post comments
First, this experiment is needlessly complicated by the description in this post. The third case, for example, is described in several different (and not necessarily consistent) ways: the occluder is shorter than the object; the occluder has a notch cut out the top; the object does (or does not) appear in the window of the occluder.
Second, the form of this experiment is that of affirming the consequent. By selecting for surprise at possible results, the form of the argument is, essentially, 'If infants develop rules they will be surprised at violations of the rules, they are surprised at violations of the rules, therefore, etc.' This is, as you know, an invalid form of argumentation.
Third, the underdetermination points to the false dilemma used to set up the argument: the choice between forming rules and storing 'videotape'-like memories of events. Clearly there are alternative possibilities, including the one that probably actually characterizes the process: the creation of (non-rule-like) patterns of connectivity between neurons caused by successive viewings of these events.
This is significant because it points to an alternative explanation of the phenomena. Non-surprise may occur when the perceived pattern of events is relevantly similar to some (set of) previously viewed events. Surprise only occurs when the new event is relevantly dissimilar to previously viewed events.
Comparing similarities of patterns of connectivity is very different from matching videotapes. In the latter, two events must be physically similar in order to match. However, if we are matching connections, two events need not be physically similar in order to produce a match. These are the cases where we suggest that a rule is being followed. But the putative 'rule' is epiphenomenal, having everything to do with how the experimenter is interpreting the results, and nothing at all to do with what is happening inside the infant's head.
Second, the form of this experiment is that of affirming the consequent. By selecting for surprise at possible results, the form of the argument is, essentially, 'If infants develop rules they will be surprised at violations of the rules, they are surprised at violations of the rules, therefore, etc.' This is, as you know, an invalid form of argumentation.
Actually, this is called inductive reasoning, and it is used in any science experiment.
The results support the hypothesis (but do not prove it). Results inconsistent with the consequent statement would go against the hypothesis. Experiments can be used to compare the explanatory power of competing hypotheses. This is how science works.
...storing 'videotape'-like memories of events. Clearly there are alternative possibilities, including the one that probably actually characterizes the process...
Stephen, this is called a MEH-TA-FOR. It is a way of describing a complex phenomenon by analogy to a simpler, more familiar process.
Interesting experiment. But I wonder how do the experimenters pull of this 'magic trick'?
> Actually, this is called inductive reasoning,
No, that isn't how induction works. Look it up. n inductive inference proceeds from a series of similar instances p1...pn to infer to the generalized case, pz. That is not what is happening here.
> this is called a MEH-TA-FOR
Not sure what the purpose of the pseudo-phonetic spelling is - presumably it's an ad hominem of some sort.
Whatever. I know it's a metaphor. D'oh. Nobody supposes that there is actual videotape inside anyone's head. I mean, really now.
But there is a theory of perception we'll call the 'image' theory of perception. Usually it is represented (as in Fodor, for example) by caricature. The central presupposition of this caricature representation is that mental phenomena are bound by physical laws, or in other words, mental phenomena are isomorphic to physical phenomena. Like - say - a videotape.
When you work with this caricature, you tend to represwent menory as though it were image mappings. That is, you represent the images as wholes, or at best, composites or image fragments of composites of image fragments. Thus, memory becomes a 1-to-1 (or isomorphic) match to those fragments. It's a pretty easy theory to disprove, and so is the favorite target of people supporting rule-based theories of cognition.
But you can create non-rule-based theories of cognition that do not rely on isomorphism (and hence, do not describe memory as some sort of identity-based matching process). That was my point in my response. Hence, I described a mechanism where similarity, rather than identity, would form the basis for remembering.
Rather than imagine how my response might have a point, though, you were too busy thinking of a good way to spell metaphor phonetically. But this isn't one of those political discussions, and merely making fun of the opposition doesn't carry any weight whatsoever, and really, just distracts from the discussion at hand.
I think the violation-of-expectation method is one of the more ingenious techniques for investigating Piaget's notion of object permanence, and more importantly, is at the very heart of elucidating the stages involved in building our memory system and its corresponding mental representations. It was originally one of the more challenging concepts to explain in my developmental psychology class. I have a less abstract idea of what might be happening in the stages described above.
Prior to 3 months, when an infant sees one object disappearing behind another one, they may actually believe that the entire object literally vanishes. Consider a real-world scenario where we see someone holding a newspaper in front of their face. We don't think that person's face has disappeared whereas a 2.5 month old might. So when the person puts the paper down, we would be shocked if that person's face wasn't actually their (violates expectation), while a 2.5 month old would not be shocked since it's obvious to them that objects disappear when not in view. This might be related to an early forming visual system that cannot keep images stored in mind for more than a couple of seconds--literally an out-of-sight, out-of-mind phenomenon.
The additional scenarios are a bit trickier to hypothesize about, but I'm guessing there's probably a sequence in which information disappears from the visual system. From the data, the infant visual system may be constructed in such a way that info about the lower half of an object will disappear before info about the top half. So using the newspaper example again, a 2.5 month old wouldn't be surprised if the person reading the newspaper lifted it up until the lower half of their face was uncovered, and the lower half of their face wasn't their; however, 3-month-olds and adults would be surprised since their visual system can store lower half info longer than a 2.5 month old. Alternatively, 3-month-olds would not be surprised if the person reading the newspaper pulled it down so that the top half of their face is exposed and find that it isn't their. Presumably because top half info disappears quickly in 3-month-olds. However, 3.5-month-olds (and adults) would be surprised as their memory system can hold all the information about a face for relatively longer periods of time.
This is fairly consistent with a famous experiment conducted by George Sperling, who identifed that the duration of information in visual sensory memory ranges from 0.5-3 seconds.
Their hypothesis is "if A then B" and they observed B. It's only affirming the consequent if they conclude that A must be true because they observed B. Given that they were open to testing alternative explanations in the second paper, it doesn't seem to me that they are saying that. Rather, they're saying "we cannot reject the possibility of A", which is entirely valid. They may also be saying that B offers evidence for A, which is Bayesian-valid if not first-order-logic valid.
Stephen, I didn't need to think too hard about how to spell metaphor phonetically. It really just came to me in a flash.
This comment isn't political, but see comment #6 re: inductive/scientific reasoning.
In addition to that, there's research out there specifically about VISION (this was more like interpretation of vision). Maybe 15 years ago, when my sister was born, I remember being told she couldn't see our faces from a distance like we could. I think it was mainly eyes and nose that stood out. I don't know if that still holds valid...
@8:
Newborns are nearsighted at birth and can only see clearly if objects are less than a foot away, which is just about the distance a caregiver will automatically hold a child when it's time for feeding. Before 2 months, when looking at faces they will stare at the edges of a face (chin, hairline, ears); after 2 months they will start looking at internal facial features (eyes, nose, mouth). The theory behind this developmental transition is that prior to 2 months they are attempting to locate objects (looking at edges helps to do that); after 2 months they are identifying objects (e.g., is face mom or dad?). Adult-like vision is probably acquired around 6-7 months when they acquire depth perception and crawling emerges. I think the 'magic' trick in the experiment might also depend on their not having fully developed depth-perception.
Another interesting study (Walton, Bower, & Bower, 1992) I came across was one designed to test whether newborns could identify in a video, their mother or a person dressed up to look just like their mother. Infants could do this since they stared longer at mom. What's interesting about this is that newborns are staring at the edges of a face at this age, which suggests that they are easy very sensitive to mom's movement, or, (dare I venture into parapsychological territory) they may be detecting differences in auras.
>> Actually, this is called inductive reasoning,
>No, that isn't how induction works. Look it up. n inductive inference proceeds from a series of similar instances p1...pn to infer to the generalized case, pz. That is not what is happening here. (Posted by: Stephen Downes).
Okay, that just made me laugh. The process described here (using a series of similar instances to infer the generalized case) is the definition of of "mathematical induction", or just "induction", which is a form of argument that can be inductive OR deductive, depending on the conclusions drawn.
induction != inductive reasoning
Deductive reasoning is basically any form of reasoning that is rigourous. Inductive reasoning, on the other hand, is "the process of reasoning in which the premises of an argument are believed to support the conclusion but do not ensure it", as quoted from wikipedia. outlier is correct.