Perception is more than the passive response to stimuli. When you focus your visual attention on something, it looks different. This is not about gaze, the orientation of the most sensitive part of your retina (the fovea), here I mean that somewhat intangible ability we have to devote our mental processing to an object in, or area of visual space without directly regarding it. Experimental data in the form of human verbal reports and the activity of single cells in the brains of monkeys demonstrate that this is quite concrete: visual attention makes you and the cells in your brain better able to distinguish a variety of properties such as color, the angle of lines and small distances1.
This viewpoint, observation as both active and reflexive, highlights a dichotomy present in debates concerning brain function in general. That is the distinction between seeing such neural exciation as occurring “bottom-up,” driven by the responses to incoming stimuli, versus “top-down,” controlled by higher level cognitive (dare I say conscious) processes.
For example, there is a Dalmatian (somewhat) hidden in figure 1. Without your knowledge of its presence, it is perhaps more natural to simply see a collection of dots, but once you’ve found that dog, its impossible to miss. This demonstrates that your perception of the object is not purely bottom-up. The image impinging on your photoreceptors alone doesn’t necessarily lead to the experience of seeing the object. Other examples include seeing faces in clouds, a result of our overactive face recognition areas, or hearing words in the sounds produced by a gaggle of geese. These are both top-down examples, your existing perceptive mechanisms imposing themselves on the incoming information.
Charles Gilbert, a professor of Neurobiology at Rockefeller University, has been researching visual perception by recording from single neurons in the brains of monkeys for quite some time. A recent paper from his laboratory was aimed at quantifying the role of attention in the perception and cortical processing of a specific visual stimulus: long contours made of small individual line segments3. Figure 2 contains examples of these contours made from more (A) to fewer (C) subsegments .
There are single neurons in your visual cortex (area V2) that will become active when exposed to these larger, constructed contours in certain parts of the visual field. This response is built up from those of cells (in area V1) that are excited by exposure to small, continuous lines in particular places like the ones making up the larger edges above. The reactions of these cells are in turn shaped from the combination of many small, pixel-like bits of information, coming from the retina. This is the hierarchy of the visual system, the responses of neurons that represent progressively more complex objects are formed from earlier, simpler patterns, until we end up finally with neurons that respond best to images of your grandmother, or Bill Clinton, or Jennifer Aniston2.
This hierarchy is important to the theme of top-down versus bottom-up because it informs us as to what is at the top and what is at the bottom. It also allows us to construct a simple example that we can apply to more complicated cases.
The shapes in figure 3 are called Kanisza figures, but many scientists do refer to them as the pac-men that they bear more than a passing resemblance to. It is hard to ignore the triangle that seems to be formed by this particular configuration of polygons, despite the fact that each edge is missing a large segment. What’s happening here is that there are enough cells in V1 turned on, collinearly, by the partial edge of the invisible triangle to excite the V2 cell that would respond to a whole triangle edge in the same location. This is not the whole story however, it turns out that in addition to the bottom-up connections mediating the hierarchy I described before, there are also extensive feedback projections from higher areas like V2 to lower ones like V1. Thus, when the cell in V2 relays information up to higher areas that there appears to be a line spanning two of the Kanisza figures, it also informs all of the V1 cells that might be making up that line, including both the ones which are in this case actually being stimulated, and the interstitial ones where there is no edge to detect. The cells not receiving any actual visual stimulus are activated, to a lesser extent than they might be, by the feedback or top-down signal.
This then is the paradigm for top-down perception. Something like this is most likely happening in the Dalmatian example as well, with some high-level neuron that responds to dogs being activated and sending feedback signals down to all of the neurons that would normally activate the percept, facilitating the “segmentation” of the dog from the background.
What Gilbert and his colleagues did in order to more thoroughly understand this process was to engage a monkey in a task related to the perception of these incomplete contours while measuring the responses of the neurons in their area V1.
The monkey was presented with two images like the ones in figure 2 simultaneously, one in which some (1-9) of the line segments were oriented to form a contour, and one in which their angles were random. It’s task was to simply look at the one with the contour. Maybe because of something intrinsic to their visual system, or maybe just because the monkey didn’t understand what the experimenters wanted them to do, they required extensive training before they became proficient at this game.
Before becoming experts at performing this chore, the cells in V1 which respond to the smaller constituent segments responded with a transient increase in their activity (figure 4, left). However, once they had become skilled at this task, the transient was followed by a prolonged bout of activity whose amplitude was proportional to the number of colinear segments making up the larger contour (figure 4, right). Intriguingly, even after the training, if the monkeys were anesthetized and exposed to the images the responses again showed only a transient increase, and no difference between the number of line segments making up the contour. What this suggests is that the engaged, top-down process of performing the task and attending to the stimuli is what generated the difference in the responses, thus active perception.
The idea of active perception brings to mind a question: what aspects of sensory experience are subject to this kind of cognitive control, and what are its limitations? Extreme cases are somewhat helpful: you can’t see a circle when presented with a square, although every circle you’ve ever seen on a computer monitor or television is simply a lot of pixels, and thus not really a circle. This sort of fuzziness is certainly present in somatosensation (touch). An example, if I arrange a situation where your hand is hidden, but there is a rubber hand approximately where yours might be, I can (with a bit of “training”) evoke a somatosensory experience in you by touching the rubber hand. Multiple pairings of poking your hidden hand while simultaneously having you watch me poke the rubber stand-in will lead to the feeling that any touch of the rubber hand is a touch of your hand. Further, I have certainly had the experience of enjoying the taste of something before knowing what it was, implying that the mere knowledge of the thing to be tasted can modify the experience of tasting it.
There must be some evolutionary/developmental aspect to all of this. At least in the vision example, if our evolution didn’t provide us with feedback connections the likes of which I described above, there would be no anatomy to mediate the top-down control. Similarly, if our development didn’t equip our visual systems with automatic detection systems for faces and lines and circles, there would be no high-level percept to feed-back down to lower systems, against which our brains might try and favorably compare incoming data.
The question then becomes, to what extent is this effect mediated by sheer brain circuitry and to what extent by the nebulous mystery that is conscious experience? I would have to argue that our brain circuitry is the only basis for our conscious experience, and thus any effect that we might attribute nonspecifically to our mental being represents our lack of knowledge about the connectivity in our skulls. However, my mother taught me that it is a great thing to be wrong, because that means you’ve got something to learn. So in any case, I look forward to deeper unravelling of these phenomena.
1. Maunsell JH, Treue S. (2006) Feature-based attention in visual cortex. Trends Neurosci. 29(6):317-22.
2. Quiroga RQ, Reddy L, Kreiman G, Koch C, Fried I. (2005) Invariant visual representation by single neurons in the human brain. Nature 23;435(7045):1102-7.
3. Li W, Piëch V, Gilbert CD. (2008) Learning to link visual contours. Neuron 7;57(3):442-51.