The next possibility that I want to consider with you is vision as efficient coding. And I need to explain what efficient coding means. And it really stems from the work of a person, Claude Shannon, who's pictured here, who's not a neuroscientist or a vision scientist at all. But was a tremendous intellect who, working at Bell Labs in the 1940s and 1950s, invented the concepts, the principles of information theory. And he did this as a way of understanding how you can compress information in telecommunications or in computer systems to make them vastly more efficient. And you may or may not have heard of Shannon, but much of what we depend on today when we Google something instantly or use anything on our iPhones or the Internet, stems in great part from the work of Shannon. So as I said, Shannon worked at the Bell Laboratories. The Bell Laboratories was closed a few years ago, but it was a fascinating place that had a major impact on science in the 20th century. It was started in the 19th century by Alexander Graham Bell as a laboratory where people could do research on telecommunications. And of the many things that happened over the 20th century at Bell Labs, the list is amazing, transistor was invented there by Bardeen and Shockley. The detection of the microwave radiation that established the reality of the Big Bang Theory was worked on at L labs. And information theory which has guided not only telecommunications, but all sorts of computer applications and messaging applications is due to the work of Shannon. What he basically determined was a way of compressing the enormously redundant information that potentially exists in any telecommunication system or any information processing system. And the visual brain is an information processing system. Certainly lends itself to that kind of thinking. So let me just tell you why, in a nutshell, the operation of information processing in the neuro system learns itself so readily to this kind of interpretation. And many of you probably know this already, but for those of you who don't, I just thought it worth taking a minute to describe the logic of efficient coding in the nervous system, why it's an appealing idea in the first place. And that is because neurons and this is a diagram of the spinal cord and a motor neuron and the spinal cord that's connected to a muscle. Neurons, whether they're in the spinal cord or the visual brain, or any part of the brain, are using action potentials to transfer information from one place to another, to distribute the information in the system. Much in the way wires in a telecommunications system or now fiber optic cables distribute information today. The action potential is a very brief more or less digital all or none, not really all or none but it's more or less all or none, in that it either occurs or doesn't occur and is very brief. It lasts only a millisecond or two. So these are basically kind of like dots in a Morse code, an on or an off signal that lends itself to the idea that information must be coded in some way. And as the information travels down an axon from a neuron to a muscle, or from any neuron to another neuron in the brain, or visual brain or from any other neurons who would target cell in our bodies. That information is being coded in something that's, more or less, like digital form and being decoded by the target whether it's another neuron in muscle cell or what have you. Well, the principles of information theory that Shannon established in the 1940s and 50s was quickly recognized by vision scientists within a few years as being another way to think about the operation of the visual system. And the person, whose name is most frequently associated with this is Horace Barlow. Horace Barlow is in his 90s today, but was another major figure in the second half of the 20th century in vision science. And among many thing that he did he recognized that the concepts of information theory could reduce redundancy in images. Well, why are images redundant? Well, in what sense are they redundant? Well, let's go back and look at the surfeit of information that arises in natural images. We've seen these images before but think of the tremendous amount of information that's in any natural image that falls on the retina. Well there are a limited number of detectors in the retina. There are a limited number of cells in the thalamus. There are a limited number of cells in the cortex. And it must be the case, and this is what Barlow realized, that the full amount of information that's in any one of these natural images, or any natural image, has got to be somehow compressed. And the compression in information theory that Shannon pioneered, seemed to have a useful application in the visual system as a conceptual framework for thinking about how you can take the surfeit of information in an image and reduce it to a sparse representation as it's called today. That is still useful just as compression of digital signals and telecommunications and computer applications does the job that we kind of take for granted today. So fine, that's really a good idea. And what's the problem for understanding vision in this way? Well, in one sense, there is no problem. Compression occurs in the visual system. It occurs because of the redundancy in vision, or any sensory system for that matter, of the information that's out there in the world. What's the problem with it? The problem in this case is that the mechanisms of transmission are not those of perception. So this is a fine way of thinking about the efficiency of processing in the visual system. And lots of good things incidentally have emerged from that. So, I'm not trying to dismiss the value of this way of thinking about vision. But with respect to thinking about perception, it's not going to get you where you want to go. Why? It's simply because efficient transmission over a phone line or over the neural circuits of the visual system is not the same as, or really even pertinent to, the message. They're just two different things. In phone lines, you can have a very efficient system as telecommunication today shows, but that doesn't tell you what the messages are. The messages go over that system are entirely different, separate from the efficiency of a system that sends them around and distributes them to us in our telecommunications or Internet systems. So the idea of efficient coding is a very good one in vision. It's definitely applicable but it's not going to tell you about the messages, it's not going to tell you about visual perception despite its impact and importance in efficiency. Efficiency, obviously, being a goal of vision, a goal of visual evolution, you want to be everything as economically as you can. That is being done by evolution, but not that’s telling you anything about the message perception.
