In the original Harvard experiment, a gorilla walks through a crowd of students passing a ball around. While you're busy counting the throws (as instructed), you don't even know you've seen a guy in a gorilla suit (try it here on Youtube thanks to Daniel Simmons). In the new one, researchers stuck a picture of a gorilla onto an X-ray of lung cancer to see if doctors would notice. Only 4 of 24 did. (I hope my doctor's one of that 17%).
The writers go on to describe the similar Low Prevalence Effect, in which you fail to identify something you are looking for but rarely see (such as airport carry-on X-ray workers looking for guns in handbags). They go on to say that the latter effect can be greatly reduced by training in which, for instance, every handbag has a gun, and then fewer and fewer do, until they are accurately identifying them in very low percentages.
Great! My problem is their conclusion that "people can train their visual systems to work more efficiently." Poppycock, says I. I find it ironic that in the same issue there's a review of a book called "Brainwashed: The Seductive Appeal of Mindless Neuroscience," which decries the rushing to false conclusions based on neuroscientific data. Because their visual systems are fine. What's becoming more efficient is their relevance realization.
Did they see the gun before? Yes, just as you "saw" the gorilla in the video. The problem was you didn't notice it. Noticing things has little to do with vision and everything to do with conscious attention. The trainees in the studies discussed in the SciAm article did not have their vision improved, they had their ability to recognize (re-cognize) target items enhanced by getting their neurons to fire in a certain sequence (caused by viewing the target item in context) repeatedly, so that when they saw that item again, they would spot it.
This training also works in reverse, by the way. Psycho-magic guy Goldmind does this great trick where he makes an artist drawing the Toronto skyline, which she has done many times before, totally forget to include the CN Tower, which is the middle of her field of vision. He basically untrains her from noticing something she has seen a thousand times before--and astonishingly quickly, at that.
Improving relevance realization is no mystery, it's just like training a dog to smell drugs. The dog's sense of smell doesn't needs improvement; in fact, if it was sub-par, you'd get another dog. It just needs experience.