The fallibility of human memory was one of the first things covered in my undergraduate psychology class (or at least it was to the best of my recollection). However, the brain is a mysterious thing, and there's still much we don't fully understand when it comes to figuring out how it stores and processes information.
The current PNAS1 features a paper from a team of scientists at MIT who have been probing the limits of visual memory. Previous studies have demonstrated that visual memory has an impressive capacity for storage; in studies where volunteers are shown 10,000 images (each for a few seconds), they were able to determine which of two images they had previously seen at a rather high level of accuracy. However, it had been thought that this visual memory was light on details, instead providing just the gist of the image.
In the PNAS study, the volunteers were shown 2,500 images, each for 3 seconds. In contrast to prior research, the images were stripped of any background details. The subjects were then shown a pair of images, one of which was previously seen and one that was new. The paired images were shown in three ways; novel, where the image was paired with an image of something from a completely different category (for example, false teeth and a DNA double helix), exemplar, where the image was paired with a different, but similar image (two slightly different starfish for example), or state, where the images were of exactly the same image, but in different conditions (such as a telephone on and off the hook).
The results of all three tests showed that visual memory is surprisingly detailed. In the novel test, subjects correctly identified the correct image 93 percent of the time. The exemplar and state test conditions were handled with slightly less accuracy but, at 87 percent and 88 percent, respectively, the margin wasn't large. The test subjects were also very accurate in their ability to detect repeated images, with 96 percent of repeat images being identified, and only a 1.3 percent false positive rate.
This work comes on the heels of some other studies on the limits of human memory, published within the last month in Nature2 and Science3. Those studies focus on the depth of visual memory when it comes to remembering data from images.
These studies suggest that, when it comes to remembering several details about an image, visual memory is dynamically allocated, in contrast to prior dogma that suggested that humans were limited to remembering only between two and five details at a glance. This form of visual memory is much like RAM though, as pointed out in a prior post. Although we can take in details and store them, that process is happening constantly as we process vision.
What these studies do all add up to is the realization that the limits of visual memory are much further than we previously thought.
1: PNAS, 2008. DOI: 10.1073/pnas.0803390105
2: Nature, 2008. DOI: 10.1038/nature06860
3: Science, 2008. DOI: 10.1126/science.1158023