3.1 - Why study Vision, and What is Vision for?

all right let's go on let's talk about vision but first let's ask why are we studying vision why is that an important thing to look at it's not just that that's what I study although that's

relevant more relevant is that we are just spectacularly visual animals if you ever doubt this try blindfolding yourself for 15 minutes and try just

doing basic stuff and you will learn why in what senses we are such visual animals we lean on vision for pretty much everything we do another way of

looking at that is that vision occupies a very large percent of the cortex pretty much roughly all of this stuff back here does vision in various ways

okay about five billion neurons so we allocate a lot of machinery division and that's both because vision is hard an ill-posed problem in many ways and it's because we lean on vision for most of

what we do okay another reason to study vision is it is arguably the best understood system in the human mind and brain and that's for

all kinds of reasons it's in part because the study of vision in humans has had this as the study of the neural basis of vision especially in humans has

had this a huge benefit from decades of research on the visual system in monkeys which seems pretty homologous to the system in humans and from which we've learned a great deal my personal view is

that the reason that functional MRI studies of vision in humans are just much better in general method logically there's a lot less kind of stupid garbage in functional MRI vision than

there is in some other domains which I'll remain nameless right now I think that's in part because there were a lot of ways to reality check based on the neuroscience of vision and monkeys which have been going on since

the 50s and so you couldn't just kind of make stuff up because there was all this kind of hard data to relate it to right from the get-go and that I think instilled kind of methodological rigor

and studies of vision which is another reason we'll start with it here okay and another reason to study vision is that until recently AI vision systems machine

vision couldn't touch us right we were just way better and so there's you know all this magic going on in human brains that the people across the street don't know how to do in machines now that story has gotten a

little more complicated because now those guys all of a sudden can do all this stuff they're really good at pattern recognition in some cases rivaling human abilities with visual

pattern recognition however there's still a lot that we can do with our visual systems that the AI guys can't and so there's still lots of secrets in the human brain that they can learn

those secrets are moved into a different zone not just pattern recognition but other kinds of things that we'll talk about later in the course to do more with visual understanding that is not just being able to label the objects in

front of you but there's people in desks and computers and a street and cars and so forth but understanding their relationship to each other how they are interacting with each other how they depend on each other what the whole just

conceptually of the scene is and that goes far beyond pattern recognition and AI systems are not so good at that and we'll talk about all of that in a month

or so okay so let's start again with the kind of low-tech version of Mars computational theory what is the problem of vision what do we use it for

well vision is about telling us what's out there in the world so there's a world out there it casts images on our retina and we figure out what's going on out there and this mystery box is vision

and that's what we want to understand okay so that's a problem how do you go from the rich input that we get in the world through our visual systems to some kind of understanding of what's out

there okay okay so I gave you examples before here's one example well that comes in you've identified that's Julia or that's Brad or that's a rabbit and so

we're really good at that just to show you how good we are at this here's a little low-tech demo I'm just going to flash up I totally ran a bunch of pictures that have nothing to do with each other and you're just going to look

at them and your amazing visual system is going to tell you what each of those things are here we go okay everybody got most of them that's awesome those things we're going at about four

or five images per second and it there's another important clue here there was no way to predict what the next image would be from the previous one and that tells

you that the idea of your expectation of what will happen next important as it is it is not is not necessary right your visual system can figure out the identity of an object even if you have

no prior predictions whatsoever as you couldn't have here okay and you're fast so we want to know how does all that work okay how are we going to

think about this well that's what we've been talking about for the last bunch of meetings and the key insight is we're going to think about vision as

information processing as a series of computations done on the input to enable us to get from this to that okay and so one of the key things that you do to try

to understand an information processing system is figure out what its steps are what are the steps in that computation what are the stages and what is represented and what is computed in each

alright so the model is really not like that there's gonna be a whole bunch of processing steps and what we want to know is what's going on at each of those all right and so looking at the brain is

just a particularly convenient way to get a sense of that because actually some of those steps are laid out spatially and we can look at each of them and say okay what's going on here what's going on there what's going on

there it's just a very literal way to try to infer them okay all right so let's look in the brain and let's try to follow that pathway and see what we can learn by characterizing the neural

mechanisms that are going on it at least the first few stages everybody on board with this agenda