How much do eyewitnesses really see?

How much of what an eyewitness "sees" represents inputs from the outside world, and how much of it is constructed by the brain?

My layman's grasp of new brain science currently revolutionizing views of human cognition remains spotty at best, but much of what's coming out on the topic appears to have implications for the criminal justice arena. Take, for example, the debate over the best procedures for gathering eyewitness identification, long considered the gold standard (or perhaps silver, after a confession) for gaining a conviction in court. Scientists are learning that to a great extent what we see is constructed by the brain from memory instead of reflecting an image of the outside world.

In an extensive article in the June 30 New Yorker on the brain and itching ("The Itch" - not online but see discussions here and here), surgeon and author Dr. Atul Gawande writes that:

[T]he more we examine the actual nerve transmissions we receive from the world outside, the more inadequate they seem.

One assumption had been that the sensory data we receive from our eyes, ears, nose, fingers, and so on contain all the information that we need for perception and that perception must work something like a radio. It's hard to conceive that a Boston Symphony Orchestra concert is a radio wave. But it is. So you might think that it's the same with the signals we receive - that if you hooked someone's nerves to a monitor you would watch what the person is experiencing as if it were a television show.

Yet, as scientists set about analyzing the signals, they found them to be radically impoverished. ...

The images in our mind are extraordinarily rich. We can tell if something is liquid or solid, heavy or light, dead or alive. But the information we work from is poor - a distorted, two-dimensional transmission with entire spots missing. So the mind fills in most of the picture. You can get a sense of this from brain anatomy studies. If visual sensations were primarily received rather than constructed by the brain, you'd expect that most of the fibres going to the brain's primary visual cortex would come from the retina. Instead, scientists have found that only twenty percent do; eighty percent come downward from regions of the brain governing functions like memory. Richard Gregory, a prominent British neuropsychologist, estimates that visual perception is more than ninety percent memory and less than ten percent sensory nerve signals.

So what we see in the heat of the moment is a "radically impoverished" version of reality and our minds fill in the blanks from memory. No wonder faulty eyewitness identifications turn out to be the single leading cause of wrongful convictions among DNA exonerees! This also helps explain why witnesses are more reliable identifying someone they knew before the crime - their memories have more data from which to construct the scene. It's when identifying strangers, primarily, that limitations on eyewitness identification come most starkly into play.