So in this second lesson, I want to discuss vision as feature detection. And as I said, if one were to ask a large sample of vision scientists today what the way in which they thought most plausible for the operation of vision, I think feature detection would be the leading candidate, and that's because there's a lot of evidence for this idea dating back to the 1950s. And let me begin by describing a bit of history that concerns three of the figures in this photograph. The first of these is Stephen Kuffler. In the 1950s he was a leading vision scientist working at Johns Hopkins. And in the late 1950s he came to Harvard Medical School and he brought with him two postdoctoral fellows, David Hubel, who's pictured here, and Torsten Wiesel, who's pictured here, so Kuffler, Hubel, and Wiesel. And their work, for which they won a Nobel prize in 81, Kuffler was by then, unfortunately, dead or he probably would've shared it with them, they established a very good case for vision as feature detection. Let me say a little bit about what this picture is. This is the cover of a book that Hubel and Wiesel published in 2005 by Oxford University Press. And I've said many times to students that if they had one book to have in their collection of neuroscience volumes, this would certainly be the one. It's a remarkable book that collects their papers written over a period of 25 years during which they collaborated beginning, as I said, in the late 1950s and going on to the early 1980s. And the title of the book is Brain and Visual Perception. We'll say more about that in a minute, but what's so remarkable about this book is that it's not just a collection of papers, it's a collection of papers with commentaries. And in forewords and afterwords they say why they did the work, and they say it in a very straightforward way, without trying to remove the warts that are involved in any scientific endeavor. And then they have afterwords that say how the conclusions of the paper fared over the years and other comments about some of the problems they encountered along the way. So this is a very unusual book in that it tells it like it is. Most books, of course, gussy things up to make it seem all sort of scientifically inevitable, but this is certainly not the case in this book, and that's, I think, the main reason, or one of the main reasons it's so remarkable. The other reason, of course, is that this work set the stage for understanding vision that began in the late 1950s but persists to this day. So what's the nature of what they did? Why has this been such an impressive and field-determining body of work? Well, this takes us back to something I mentioned early on in the course but we need to come back to it now and talk about it in detail, which is the concept of receptive fields. And this way of doing an experiment is really the standard way of thinking about the generation of visual perceptions, and I want to describe it to you in a little bit of detail. They started with cats as an experimental animal in the late 1950s and on through the 60s, and then turned to monkeys, rhesus monkeys in particular, which is an animal that's much more like the human visual system, but it's the same idea. And their students and many of the students' students, a whole line of progeny over the last 50 years, have carried on work like this, defining the receptive fields of individual neurons, individual system. The idea being that if you could identify the behavior of individual neurons and how they connected to each other, you could understand not only how the visual system worked but how perception would fall out of that. So let's make sure we are all on the same page about what's going on here. So the cat is anesthetized, of course. It's looking at a screen that's placed in front of the animal, and stimuli are shown in that screen. That can be determined in any way the experimenters wanted to do. In the example here, there are bars of light that are oriented in one orientation or another. And recording from an individual cell, cell by cell with a microelectrode which doesn't go inside the cells but is an extracellular microelectrode that records when you place it in the cortex, the primary visual cortex, in this case, of the cat, records from one cell or another. And you can ask, well, what does this particular cell that I've isolated with a microelectrode in this way, what is it responding to? And again, in the example here the question being asked is what is the location on the screen that it's responding to? And what's the nature of the stimulus that it responds to? What are the characteristics of the stimulus that it likes to see, that it likes to respond to? And that's being judged in terms of the number of action potentials that the cell fires. So you can hear this on an audio amplifier. And you can actually find this on the Internet. And I suggest for those of you who register that you find the Hubel and Wiesel video of their actual experiments and see how this was done and how [LAUGH] different it was 50 years ago from the way it's done now, more than 50 years ago now, than the way it's done now. But the point has been the same over this period of decades, and the first point is that there's a location in visual space that any particular cell is going to respond to. So in this case the cell doesn't respond, the cell that one's recording from here. Remember, it's just a single cell being record from the primary visual cortex of the animal that, although anesthetized, the visual system is still working at a reduced but adequate level to make these kinds of experiments. But there's only one place on the screen that the cell is going to respond to. That's the location of the receptive field. The receptive field is the locus in space that the cell responds to. That's the sort of primary definition of a receptive field as a location. You put the stimulus here, you put the stimulus here, you put the stimulus here, you put the stimulus here, no response. So there's a particular locus, a particular location of the receptive field, for any cell that one wants to test in this way. The second characteristic, and it's equally important, is, as I said, what kind of stimuli, not just in the placement of stimulus, but what kind of stimuli does the cell like? And again, you hear this by an increase in the frequency of action potentials that you listen to or you see on your oscilloscope that I'll tell you about in a second. So these are called the receptive field properties, the characteristics of the stimuli that cells like. And there's a long list of characteristics that can be tested. In this example, Hubel and Wiesel were testing the orientation of the stimulus. So this bar can be oriented in this direction, in this direction, in this direction, in this direction, again, in the appropriate locus in the receptive field. But obviously you can orient the stimulus as diagrammed here in a variety of different ways, and in that way determine what's the orientation that the cell likes. Can the stimulus be in any orientation, or is there a particular orientation that it is tuned to, that it likes best? And what you see here is typical of cells that are recorded in primary visual cortex of basically cats, monkeys, presumably us, that if you orient the bar in one way, there'll be no response. There are no action potentials generated in response, again, and just this example, in a cell where the orientation is horizontal. If you orient the stimulus vertically, there is a strong response. If you orient the stimulus obliquely, there is an increasing but reduced response up to the maximum that's vertical. So it's not that all cells respond to vertical orientations. Of course, they respond to different orientations. The point is that any one cell likes a particular orientation, and that's a way of defining its receptive field properties. And you can, in this way, define a tuning curve for any cell. What is the stimulus that it likes best? And of course, it's not just orientation. You can do this for luminance, you can do it for color, you can do it for motion, you can do it for direction of motion, you can do it for a large number of properties, and the list goes on. And over the last 50, 60 years, people have done this in extraordinary detail, and one understands the receptive field properties of animals, cats, monkeys, other species, very well on this basis. Now what's the upshot, and how does this relate to the idea that vision is about feature detection? Well, you can see from what I've just said that the implication is very strong here, it's not an implication, it's just a fact, that the neuronal activity represents in some sense image features. The idea is that, okay, there are image features of the retina, the orientation in this case of a stimulus, and that's a feature that the cell responds to and there's a particular orientation that it likes. So one would imagine from that that vision is all about detecting features, measuring the features in the world, reporting them to the sensorium, to our visual brains, by virtue of the features that are on the retina that are translated into features in our perceptual sense, and that those features correspond in some way to reality.