AI: The Technological Trickster
By ai-depot | October 26, 2002
Vision and/or Vision
A fundamental activity of the brain is to interpret incoming perceptions and to attempt to incorporate them into a known worldview. Much research in virtual reality has begun with attempts to provide computers with sensory experience so they will have something to manipulate that is amenable to machine intelligence.
At Sandia Labs, researchers aren’t talking about AI much any more. For one thing, the attempts to duplicate what even the eye does is just too many orders of magnitude above what even the most sophisticated computer imaging can do. For another, too many unknowns are involved.
To image a mountain, for instance, a grid is developed on the screen, then it is manipulated to resemble the shape of the mountain. Each polygon of the grid is then coloured to represent the trees and rocks, then shaded depending on where the sunlight is coming from. This process results in fairly accurate representations, and may be something like what our eyes do when we look at the mountain. But the differences are quantitatively huge, if not qualitative.
A computer capable of imaging a scene may be able to work with a grid structure of two or three thousand polygons. To do what the eye does “…would take at least 80 million polygons to represent the same view…” Even if this were possible, we would still need the “…form, shape, colour and depth of field information… extracted from the image…” which the eye does “…without us recognizing the effort.” (Pimental and Teixeira, p. 105)
In conjunction with the image is the revision of it that takes place as the brain acts on it. Many of the processes are unknown. Stereoscopic vision, for instance, allows very accurate depth perception. But as I learned from a former one-eyed pilot, the ability to see depth-of-field is possible with only one eye. Sandia researchers have discovered the eye’s ability to read parallel lines plays some part in depth perception, among other possibilities.
Computers are nevertheless capable of detecting and representing an astounding amount of detail in any given image. But there seems to be much more to perception than simply the ability to provide a detailed analysis of the scene. Some larger sense of the totality of the scene, where it fits within a larger context, is necessary for a proper interpretation. Campbell notes that “…perception… depends not on….detail, but on a set of properly ordered perspectives…” (Campbell, p. 210) The brain’s ability to observe a given scene from a variety of perspectives, a trait that occurs through selection, is only in its infancy in the computer world.
Jerry Slambrook of Sandia Labs considers the emerging realization that we cannot build computers to imitate precisely what even the eye does is nevertheless generating a whole new paradigm as researchers switch to making machines that compliment, rather than replace, the brain. (Pimental and Teixeira, p. 171)
Tags: none
Category: essay |
