Vision is a 'Reflex'

kernkopje writes "A recently publicized book by two neurobiologists at Duke University introduces an interesting emperical theory on vision. Rather than postulating a visual system that generates a picture of what actually exists in front of the viewer, they theorize that evolution, as well as life experience, has created a visual system in which perceptions represent what a given visual stimulus has typically signified in the past. Admiring the view from a high building, staring at a beautiful woman, shapes, colors, textures, it's basically all the same knee-jerk response... The news release is here, read more about the concept here. Their lab work & results can be found here."

Comments

Looking at prOn...

[darkov]

... it's not exactly a knee jerk reaction...

Honest, Dear

[Zerth]

It's just reflex to look at hot chicks.... I mean, I look at you, don't I.

Somehow that never works.

To simplify

[dpilot]

Vision is whatever helped our ancestors survive. If that were a pure image of objective reality, that would be how we see. If it were an interpretation based on past events, and that that predator is going to eat me and deprive me of my right of reproduction if I don't move, pronto, then that would be how we see.

How well this fits into modern life and jury trials is another issue.

Syntax vs Semantics

[SmileyBen]

Whilst this article is interesting, and raises some interesting points, they seem to come to some very radical conclusions, based on what can only be called a mistaken belief that vision is 'about' something. They claim that rather than perceiving what is out there, we perceive what we've been trained to see, by past experience individually and of the species.

Well, we can take this two ways. We could say that we've grown eyes and a complete visual system purely based on previous stimuli, in which case it's obvious that new stimuli will only be interpreted in terms of past stimuli. Or we can say that we currently have an apparatus, and despite the fact that something we're immediately seeing might be ambiguous and look like something we've previously seen we could have further sense data that tell us that it isn't what we'd seen before - so what looks like a famous statue could be revealed to just be a photograph of a famous statue when you move around and see that it doesn't change corresponding to how it should.

But either of these interpretations - the past-centric and future-centric ones - are just that, interpretations. They still aren't 'about' anything.

What this article seems to be claiming is that anything you're currently seeing is actually 'about' things in the past, but clearly this isn't true. Vision isn't 'about' anything. When you have a robot with sensors you don't say that when they 'see' an object in their way what they're seeing is 'about' their programming, so they're not actually seeing anything in the present.

Sorry, I'm not putting this very well. Basically I'm trying to say that either their claim is banal - the obvious fact that our eyes are only capable of creating objects in our 'minds' based on things that have been saliant in the past (such as emphasising red objects - i.e. berries - in a green field - i.e. leaves - more that they actually contrast), since obviously our visual system has evolved based on what has been saliant in the past. Or they are making an incorrect claim that current vision is just 'seeing the past' because we don't actually get new ideas from our visual fields, we just try to fit it into previous sets of sense data. The reason that this must be incorrect is that past sense data is no more 'real' no more 'seeing actual objects' than current sense data. So if I see an apple now, it isn't fair to say I only know it's an apple because I've seen apples before because the only sense in which I've seen apples before is the sense in which I'm currently seeing apples. There isn't a 'good old days' when we really saw things, and which we're just reminding ourselves of every time we try to see again...

Re:Syntax vs Semantics

[Bazzargh]

If you program a computer to 'see' (using cameras or whatever) it can tell that two colours in an illusion are the same if they are 'really' the same.

Now do the same to a human. You might /know/ that you are seeing an illusion, but you /cannot/ see the two colours as being the same. Thus the eye isn't a simple camera, and we certainly don't have access to the vision 'bitmap' from the cornea in our concious brain, or we could train ourselves not to see the illusions.

The writeup says that the conventional explanation involves (something like) colour bleed within the neural system to get the effect; for which there is no direct evidence and no explanation for /why/ our eyes would have such a defect (why didn't illusions disappear as we evolved better sight?).

The idea that there might be some evolved preprocessing mechanism at the fundament of our vision which prevents us from seeing like cameras (as opposed to simple 'mechanical flaws') is actually recent and controversial (theres a chapter on exactly this in The Blank Slate thats worth reading), so I don't think it quite qualifies as 'banal'.

The Duke guys don't seem to be quite saying that we see the past at a macro level (leaves, trees, sheep) so much as at a micro level. Something is translating raw colour info into shape, distance and colour /cues/ that we internally reconstruct into the image using past knowledge. This layer is where the visual illusions arise.

Anyway, thats what I thought they were saying, I could have completely misunderstood. It would be more convincing if they could construct a computer model from their ideas that had the same vision defects we do.

-Baz

Re:Syntax vs Semantics

[SmileyBen]

Sure. You're absolutely right. A computer can tell us that the wavelengths of light hitting two sensors are the same. If what these researchers are saying is 'Isn't it bizarre that that doesn't happen... that we can have illusions?' then the simple answer is that if that was what eyes were trying to do (as much as they're trying to do anything) that would be a problem, but they're not.

Frequently we don't see two instances of the same wavelengths of light as the same colour, but that's because there are obvious evolutionary advantages of not seeing the same 'colour' as the same colour quale (the 'mental image of the colour'). Such as when there are red berries amongst green leaves, where it is advantageous for the red to look more red and the green to look more green, or such as edge detection, where if you have alternating strips of light and dark the light seems lighter and the dark seems darker where they meet, emphasising the edge.

I assumed that everyone had accepted that these were the products of evolution, and didn't think that illusions were illusions in a pejorative sense (i.e. not defects) - but it seems like these guys are presenting this as something new and exciting, and I just can't see how...

Re:Syntax vs Semantics

[erichbox]

Yes, illusions occur at the level of perception. The researchers in the article even speak briefly about the currently accepted theories on visual neurobiology involving the retinas, nerve cells, etc. The current theories even explain some of the types of visual illusions that people fall into. However, what is novel here is that the researchers have isolated certain types of visual illusions that current theories can't readily explain, and the researches have come up with a theory that *can* account for these specific types of visual illusions, as well as all the others, *and* their theories _may_ prove to be generalizable to the other sense perceptions occuring in the cerebral cortex. In no way is this banal observation.

There was an artile like this

[Unknown Poltroon]

in Discover, or scientific american a while back. It was a bout a man who had been given sight back after losing it as an infant. He went form being blind, to having decent vison at teh age of 40? or so. IT was fascinating reading. Depth perception for him was something he had to think about, optical illusions didnt work. Catching a ball was a new experience. He learned to ski while blind, so skiing with vision was actually harder in ways, he had to lear to interpret what his eyes were telling him, versus what his legs were telling him.

I could have told them that!

[hackwrench]

Took that story long enough to wind up on slashdot.

I could have told them that what you perceive as being there is based on your previous experience as opposed to what is actually there. The data from your eyes is just far too noisy, and my brain has guessed wrong often enough that its become obvious that what I was seeing was based on previous experience. By the way, hearing works the same way. Calling it a reflex, however doesn't seem right.

I doubt this will revolutionize anything...

[vidnet]

Why? Because nothing revolutionary has ever had given me the message "Flash error: version 6 or better required."

Seems like common sense to me

[zejackal]

It seems like common sense to me to say that the initial perception of an object is like a reflex. Think about speach or hearing in general. If your mother speaks to you, you don't have to listen to her to know that it's her, you just know. Your brain makes the determination as to who is speaking and you simply know.

Now let's say you are talking to someone for the very first time. Do you instantly recognize their voice? Of course not, you've never heard it before. Do you recognize it as a human voice? Yes you do, unless they've had a trachiotomy or something. The brain of course makes this determination and tells you that it's a person not a dog or a cat, etc.. Now the next time you speak with this person, the voice may be more familiar. Each successive time you speak with them, your chances of accurately identifying the voice without having to think about it go up.

Certainly some aspects of our visual system are inherited, the fact that we can distinguish between red and green even thought there is little contrast in their brightness. Or the fact that our vision is best suited to naturally lighted (read sunlight) situations. These are basic adaptations of our visual system, both sensors and processing, to the environment in which it evolved.

What the article seems to be saying is that all less basic aspects of our visual system are learned. For example, you may not know what type of car is driving in front of you in traffic, however after you get the latest Road & Track and read about model X, you begin to recognize the car without having to think about it. I'm not saying the article is banal, I'm simply saying that what it talks about seems like common sense. That said, I think it is important to scientifically explore things that seem like common sense, because if we were to simply dismiss a line of inquiry because the answer seemed like common sense, we would often miss out on some very interesting discoveries.

Re:Seems like common sense to me

[MrGrendel]

Not only does it seem like common sense, but it isn't even a new idea in neurobiology. The only thing that might be new about this is attaching the label "reflex" to it. Evolutionary psychologists and vision researchers have been talking about this stuff for at least the last decade, and probably longer. There is nothing new about the idea that visual illusions exist because visual stimuli are ambiguous and the brain interprets them according to the most likely (most frequent) source. I hope that this is just a case of an uninformed PR rep oversimplifying their theory.

The article also misrepresented some of the research that has gone on in the past. Visual circuits have been traced and analyzed in an effort to learn how the brain sees, not what it sees (which is what their theory is really about). Researchers want to know how the brain is able to detect motion against a noisy background, how people are able to represent 3-D objects in their heads, how the brain assembles thousands of visual cues into a coherent representation of the world. (Some brain injuries disrupt these functions -- there are people who cannot see motion - moving objects are percieved as stationary objects that jump around discontinuously - some people are not able to construct an image of everything that their eyes detect - they can only see one or two distinct objects at a time). The observation that vision is a kind of reflex does not even begin to answer questions about how the visual system works. I don't see how they can call it a theory of vision if it doesn't offer any explanation for fundamental problems that visual systems must solve.

Also the comment about detecting the source of sounds is incorrect. The neural circuit that takes care of that has been known for a long time. It localizes sounds by comparing phase differences between sounds arriving at each ear. In fact, it has even been demonstrated that the circuit trains itself to do this task (a great example of how Hebbian learning works).

I'll give them the benefit of the doubt and assume that the person who wrote the article didn't understand a bit of their theory.

'the eye isn't a simple camera'

[oliverthered]

Well, it kinda is. It's the bits that do the processing that arn't.

Suffering from frequent migrains and having some other bad seritonine based illnesses I can tell you what it's like to have diffency or enhancement in processing what you see.

When you see words/letters your brain see's words and letters, not a bitmap.
How do I know this, well sometimes I have problems being able to see shapes properly, and have to consesly work out what things are.

Patterns are picked out as textures for the shape, sometimes I see the patterns 'waving' around and blury, it's very hard to tell what the shape is.

Speed and motioned is also determined using some of the elements, often when a pattern becomes wavey it feels kinda like the floor is moving, and my sense of distanse goes tits up.

Well, it's all fun, and I can get by, it just takes a little more concentration now and again.
Oh and did I say it's fun, like being on Acid for free.

Duke Magazine need a course in perception

[exp(pi*sqrt(163))]

Didn't anyone tell them that having a vertical grey bar through black text on a white background is painful to read? It seems any idiot can be hired to do graphic design these days.

White balance.

[TheLink]

The colour thing is useful for humans - coz it's good to perceive how ripe a fruit is even if the lighting changes.

It's probably similar to automatic white balancing which you have in many modern colour cameras. But the reason for that is kind of circular - most pictures are viewed by humans. If a computer ever needed to guess something's objective colour under subjective light it'll be useful.

As for the flaws/bugs/illusions etc, hey that's evolution for you, perspective, autocontrast/balance. If I want to complain about anything, I'd complain about the damn blind spot practically next the the fovea. Octopuses don't have such blind spots.

This article is bullshit.

[t]

If you examine figure 4, the rubix cube, with gimp, then the so called brown squares are NOT the same color.

Compare this to an illusion that actualy passes this test: checkershadow

The question is why make a bogus illusion when a good ones exist? NIH? (Or perhaps they don't want to support MIT?) Is all their "research" of this quality?

The reason why the brain confuses colors and such is because it quite sensibly corrects for shadows. It very sophisticatedly determines the light source, and compensates for it. Why is that so hard to grasp?