06.09.2021 Views

Mind, Body, World- Foundations of Cognitive Science, 2013a

Mind, Body, World- Foundations of Cognitive Science, 2013a

Mind, Body, World- Foundations of Cognitive Science, 2013a

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

For our purposes, though, the above Sudoku example illustrates how constraints<br />

can be propagated to solve problems <strong>of</strong> underdetermination. Furthermore, it shows<br />

that such solutions can be fairly mechanical in nature, not requiring higher-order<br />

reasoning or problem solving. For instance, the “there can be only one” constraint<br />

could be instantiated as a simple set <strong>of</strong> interconnected switches: turning the 5 on<br />

in Figure 8-4 would send a signal that would turn the 5 <strong>of</strong>f at all <strong>of</strong> the other greyshaded<br />

locations.<br />

The natural computation approach to vision assumes that problems <strong>of</strong> visual<br />

underdetermination are also solved by non-cognitive processes that use constraint<br />

propagation. However, the constraints <strong>of</strong> interest to such researchers are not formal<br />

rules <strong>of</strong> a game. Instead, they adopt naïve realism, and they assume that the external<br />

world is structured and that some aspects <strong>of</strong> this structure must be true <strong>of</strong><br />

nearly every visual scene. Because the visual system has evolved to function in this<br />

structured environment, it has internalized those properties that permit it to solve<br />

problems <strong>of</strong> underdetermination. “The perceptual system has internalized the most<br />

pervasive and enduring regularities <strong>of</strong> the world” (Shepard, 1990, p. 181).<br />

The regularities <strong>of</strong> interest to researchers who endorse natural computation are<br />

called natural constraints. A natural constraint is a property that is almost invariably<br />

true <strong>of</strong> any location in a visual scene. For instance, many visual properties <strong>of</strong> threedimensional<br />

scenes, such as depth, colour, texture, and motion, vary smoothly. This<br />

means that two locations in the three-dimensional scene that are very close together<br />

are likely to have very similar values for any <strong>of</strong> these properties, while this will not<br />

be the case for locations that are further apart. Smoothness can therefore be used to<br />

constrain interpretations <strong>of</strong> a proximal stimulus: an interpretation whose properties<br />

vary smoothly is much more likely to be true <strong>of</strong> the world than interpretations<br />

in which property smoothness is not maintained.<br />

Natural constraints are used to solve visual problems <strong>of</strong> underdetermination<br />

by imposing additional restrictions on scene interpretations. In addition to being<br />

consistent with the proximal stimulus, the interpretation <strong>of</strong> visual input must also<br />

be consistent with the natural constraints. With appropriate natural constraints,<br />

only a single interpretation will meet both <strong>of</strong> these criteria (for many examples, see<br />

Marr, 1982). A major research goal for those who endorse the natural computation<br />

approach to vision is identifying natural constraints that filter out correct interpretations<br />

from all the other (incorrect) possibilities.<br />

For example, consider the motion correspondence problem (Ullman, 1979),<br />

which is central to Pylyshyn’s (2003b, 2007) hybrid theory <strong>of</strong> visual cognition. In the<br />

motion correspondence problem, a set <strong>of</strong> elements is seen at one time, and another<br />

set <strong>of</strong> elements is seen at a later time. In order for the visual system to associate a<br />

sense <strong>of</strong> movement to these elements, their identities must be tracked over time.<br />

The assertion that some element x, seen at time t, is the “same thing” as some other<br />

Seeing and Visualizing 375

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!