Novel Colors Against Physicalism

Alpha Meme

More Articles

About Sascha

Dr. Sascha Vongehr [风洒沙] studied phil/math/chem/phys in Germany, obtained a BSc in theoretical physics (electro-mag) & MSc (stringtheory)...

View Sascha's Profile

Sascha Vongehr

Light which triggers predominantly the so called L-cones, which are photoreceptors for long (L) wavelengths, trigger us to experience red color. In this sense, some people hold the color red to be physical and “out there” in “reality”.

Purple (magenta) is elicited if our L-cones and also the S-cones, which are especially responsive to short (S), “blue” wavelengths, are both excited, while however the receptors for medium (M) wavelengths in between L and S (M-cones responsive to green) are much less excited. In this sense, we can represent “red” by the LMS values 100 (only L excited), while “purple” = 101. Yes, this is similar to RGB colors, however, L-cones are most responsive to green-yellow wavelengths rather than red light.

The point is that there is no single wavelength light that elicits purple, because a wavelength between L and S will trigger mostly M-cones, and thus be “green” = 010 rather than purple. Purple must be a mixture, and not just one like orange (110), but it must be a "mixture with a hole" in the more physical description via wavelengths, a zero between several 1s. Even to those who claim that pure red or blue “are out there” and all else is out there as mixtures, purple is rather the hole, the zero(s) between 1s!

L-cones are most responsive to green-yellow light, not red.

Does the above imply that color is "only in my mind" and not “really out there”? Yes and no:

Our color space is two dimensional (2D). It is spanned by a 2D surface in the 3D intensity space of three different kinds of wavelength receptors, namely L, M,and S types. This simple dimensional color space has been put into question by “Novel colours” [1] like reddish-green [2,3]. But let us nevertheless stick with the dimensional color space, because then we can get to “novel colors” in a systematic manner via my “generalized purples”.

The system of four “pure” hues (yellow, red, green, blue) in spite of three different types of receptors is a mostly cultural phenomenon. We have three dimensions, L, M, and S, which leaves a 2D color surface after taking intensity out as a mere distance to the origin at 000. You can traverse a color circle around the white point on that 2D surface, and therefore, there is a “purple”, i.e. the circle closes and starts again after having gone through the spectrum of wavelengths.

Follow the spectrum from red over orange and yellow all the way to blue, then close the circle around the white point in order to hit purple along the lower edge, then thank the magic of qualia for "purple".

Most birds and even perhaps 3% of female humans have four different types of color receptors, namely one more for ultra-violet (UV), a condition called Tetrachromacy. Their color space is a patch on a 3D hyper-sphere hyper-surface in 4D space: L, M, S, UV. They have different and more “purples”: While we have only LMS=101, they have a probably similar 1010, but also 0101,1001, 1101, and 1011.

Those visual systems do see those “hyper-purples”; they are “in their minds”, but they are not, have never been, in my mind. Thus, they are in other minds in the same (= indistinguishable) way as I would see it, too*, if I could, so these purples are “out there”! With more types of photoreceptors, there are ever more of my hyper-purples waiting to be discovered – how many are there with Pentachromacy?

Just like red could possibly be described in a book titled “How red looks like, finally satisfactorily explained and thus taken from the ineffable” (likely to be an audio book popular with those blind from birth), so also the UV-purples are a distinct and potentially describable to the extend of widespread inter-subjective consent, and in that sense, they are “out there”, to be discovered somehow, just like the red is out there relative to red-green color-blind people, reachable with technologically modified eyes plus training of our visual systems, out there “existing” outside of my mind although not directly physical.

Would they be different for different wavelengths (IR instead of UV), for different emotions (green blood), for different cultures?

-------------------------------

* Except you think that everybody has a personal red.

** My argument is similar to that of Nagel [4] about Bats and Frank Jackson’s [5]. Jackson imagined a vision researcher called Mary [6] who gets to know everything there is to know about color while being in a black-and-white-only room. When she steps out for the first time, there is something still to be discovered: color. Some “Physicalism” is false because the complete physical account does not tell us everything there is to know.

--------------------------------

[1] Evan Thompson: “Novel colours.” Philosophical Studies 68(3), pp 321-349 (1992)

[2] Hewitt D. Crane and Thomas P. Piantanida: “On Seeing Reddish Green and Yellowish Blue.” Science 221 no. 4615 pp.1078-1080

[3] Billock, Vincent A.; Gerald A. Gleason, Brian H. Tsou (2001):"Perception of forbidden colors in retinally stabilized equiluminant images: an indication of softwired corticalcolor opponency?". Journal of the Optical Society of America A18 (10): 2398–2403. Direct Link

[4] Thomas Nagel: "What is it Like to Be a Bat?" Philosophical Review 83: 435-450 (1974); reprinted in his Mortal Questions (Cambridge: Cambridge University Press, 1979), pp. 165-180, p. 175: "If mental processes are physical processes, then there is something it is like, intrinsically, to undergo certain physical processes. What it is for such a thing to be the case remains a mystery."

[5] Frank Jackson: "Epiphenomenal Qualia." Philosophical Quarterly 32: 127-136 (1982)

[6] Frank Jackson: "What Mary Didn't Know." Journal of Philosophy 83: 291-295 (1986)

Comments

What about synesthesia? Is it real or not? Same debate. Some Asperger people have the capacity to calculate large numbers via colours or landscapes, so there must be something real about their representation of the reality, but does it compromise the necessary agreement we must establish about what reality is or is not, or about the method to consider what is within or without the range, if we want to progress, and therefore you could distinguish between this and that :

Other languages have various concepts for "reality" or the fact "to be", and these concepts produce aspects of the worlds. Take the French with its "reality" and its "real", both substantive, and you could say the entire XX°th century french literature is looking forward to answer questions about what "the real" is opposed to common ground "reality". Culture produce effects. In Spanish you have two verbs, "ser", which would consider "the being as being", and "estar", which would consider the quality of "being such or such", you can bet this distinction also produce strong effects about everyday metaphysics in the spanish world. So in English you might get confused about what reality is, whether you think of it as "reality" or "Reality" with a capital R maybe. Interesting fact to remember, contemporary realism comes from the US and Australia, it doesn't mean there's a link between this and that, but you could wonder about that fact. You could scrutinize German, Latin, Greek, Chinese languages, you'd find peculiarities too leading to their respective cultural system of thinking.

So here are we in the contemporary world, we have a method, or a series of criteria to admit or reject methods in their legitimacy to assess about what reality is. While we must also admit that, reality as a bold thing... well we do have theories so far.

Where to place the cursor? Some will be happy with an internal realism, or even with an Internal Realism, some will want to add some bits of constructivism or constructionism, some will delight themselves with the abrupt walls of the integral strongest realism, or the strongest claims of absolute relativism, and so on. You see the contemporary field light up on the basis of the ancient and medieval problem.

I'd favour to use a realism/antirealism selector function of a pragmatical aim. In all situations where there is no critical problematic of nomothetics or social power attached to the description of the world for what we have to do there, common ground realism a minima will just be fine. The world is separate from us, we have the average sense data to capture an fairly representative sample that we can share outside of any major stake, and we can build detectors and artificial senses to explore what we don't feel. In short, it's the physicist 's reality. You see immediately that, at the edge of the present time the hard debate about what even the physicist's reality might be, is uncertain. But this is not to be confused with the situation discussed below, and after all this is for common ground that there are standard models to sum things up from time to time.

But there are situation where calling a cat a cat is not neutral, because it takes the social world with it, and it has political implications that would seem "natural" to a conservative realism, and "constructed" to a nominalistic progressist, to put it in caricature. We have to switch the selector, simple rule, the social, human, political stake is too high to leave the naming or labelling of the world to irresponsible naive scientists or dishonest partial politicians, experts, lobbyists, and we have to lay stress on the fact that many aspects of the world in its social dimension, is not transparent at all to its users and parties.

You could as well keep on refining on the selector switch positions, like does subtly, for example, a mild physicalist and a mild anti-realist philosopher at the same time Ian Hacking.

The appropriate use of the selector switch would indeed recognise the purple as a real nuance, unless there would be an insane arbitrary policy at risk in some place of the world to order to eliminate those who do or do not perceive purple as a proper distinct nuance. And you see such policies operating everywhere, so halas! we're not finished at all with this silly dissertation about what the real world really is or not...

Any one who claims absurdly this reasoning is but anti-science propaganda, btw, is a (rather thick) lobbyist, hello lobbyist!

Mathieu Baret (not verified) | 01/28/13 | 11:25 AM

Link

Fine article, Sascha. I have seen on TV a claim that ducks have five visual pigments and can see into the far red (near infrared), and that they can distinguish different shades of green in vegetation more clearly.

Here is a higher resolution version of the same figure of Bird Visual Pigment sensitivities.

Regarding perception of colour, what about things like this: Hues and views: A cross-cultural study reveals how language shapes color perception?

There is an interesting book on this subject Through the Language Glass: Why the World Looks Different in Other Languages by Guy Deutscher. In the 19th Century, people wondering about the paucity of colour words in Classic Greek and other languages of the time, and puzzled by expressions such as “wine-dark sea” in Homer, speculated that colour vision might have evolved very late in human history.

—

Robert H. Olley Quondam Physics Department University of Reading England

Robert H Olley | 01/28/13 | 12:11 PM

Link

I have seen on TV a claim that ducks have five visual pigments and can see into the far red (near infrared), and that they can distinguish different shades of green in vegetation more clearly.

A mantis shrimp has 16. So imagine how awesome this mantis shrimp looks to another mantis shrimp:

—

Want more no-nonsense, independent science? Buy Science Left Behind

Hank Campbell | 01/28/13 | 12:22 PM

Link

It may seem to readers that I implied it, but I did not claim that "awesomeness" goes up with the number N of different receptor types along the frequency parameter. The human ear has thousands, yet auditory phenomena have no purples, but instead harmonies etc. If you are not an audiophile, such is not as awesome as greenish-red or yellowish-blue. So, depending on how the response curves of the 16 receptor types overlap in frequency space, and depending on how much associative neural capacity is involved in the visual phenomena, a mantis shrimp may look dull to other mantis shrimps, yet smell like heaven.

Sascha Vongehr | 01/29/13 | 01:05 AM

Link

With 16 receptors, one could go some way to colour coding Bertie Bott’s Every Flavour Beans without relying on patterns.

—

Robert H. Olley Quondam Physics Department University of Reading England

Robert H Olley | 01/31/13 | 04:17 AM

Link

I believe to have covered all that via

The system of four “pure” hues (yellow, red, green, blue) in spite of three different types of receptors is a mostly cultural phenomenon.

Yes, well, not that this is a lot of detail, but then I can be all Wittgenstein and claim that I dealt with the topic more comprehensively than anybody else by not mentioning it.

Sascha Vongehr | 01/29/13 | 01:29 AM

Link

Have you read what Dennett has to say about this? I won't try to summarize him, but he doesn't seem to agree with you. He calls the black and white Mary experiment an intuition pump. It relates to the "know everything" about color without experiencing it.

Other thoughts on this:
1. What would "know everything" about color and human color perception mean? If it means knowing precisely how your brain would react to seeing red for the first time, then such knowledge can only be obtained by actually computing what your brain would do, i.e. unless you believe in P-Zombies, such computing is actually experiencing it. In that case "know everything" cannot even be defined without whatever is meant by "experience" also being included.

2. Practically speaking if you had kept someone in a room with just black/white all their life, the signals from their red receptors could fuse with the green etc so when they left the room, they would just still see grey. Differences in signals may (when stepping out the room, red/green are not always the same intensity anymore) then have their brain gradually rewire itself so greys would start by feeling slightly different, then discern themselves into red/green. That would be something worth experiencing.

Regarding having more receptors, what would it be like if you have not 5, but more like >80 or so overlapping receptors?
Wouldn't that be just unbelievably mindbogglingly incredible?

However you have this for hearing, and you could make up an 80 dimensional plot to show that the sounds you can hear are sooo much richer than the colors you see, but it sure doesn't feel like that. If you take a spectrum, there is high correlation between these different dimensions - the structure/correlation of the data as much as the number of dimensions you choose to describe it in determines how it feels.
You can also of course collapse the sound signal down to 1D without losing any data.

Because of that, I am not convinced that someone with 5 color receptors would have that much richer an experience, especially if the filters were close together and gave little extra uncorrelated information.

—

Thor Russell

Thor Russell | 01/28/13 | 15:50 PM

Link

@Thor

I understand what you are saying, but you are begging the question by presuming from the beginning that "such computing is actually experiencing it". If computing were the same thing as experience than that would mean that performing the same computation with pencil and paper, or abacus beads, or a Leibniz Stepped Recokener, would, even if it took 100 years, turn something red to someone.

Do you think that it would? Do you think that at some level of complexity computations begin to itch and squeak? To invent green out of nowhere? Why would they, and how could they is my question. If you have computation, you don't need qualia.

Here's my take on why new colors can't be imagined: http://s33light.org/post/40296309313

Craig Weinberg

Craig W (not verified) | 01/28/13 | 20:44 PM

Link

Since conscious experience (there are no other sorts!) is not observable by an external observer, it is quite impossible to model it in objective language. When I say "Look for the red-wave complex in the subject's EEG" I merely draw on past discovery that, when I provide a red stimulus, a certain response is created. I do not provide an explanation of why that response can be felt - or *is* a felt thing if you want to avoid the recursion of a "felt feeling". The "red-wave complex" is a surrogate experience. It allows a scientist to talk about something red without saying what the experience of redness is. If the scientist wishes to assert that the experience exists, then he is reliant on a conscious observer to say when the experience occurs, otherwise there is only response. We take it for granted that brains and minds are consistent but p-zombies are always lurking ready to jump up and bite us if we get over-confident. and then there is the reporting problem. A subject may say she experiences redness and may be perfectly sincere when she says it, but it is not impossible she is deluding herself that she has some mystic capacity. It is the old "hard problem" applied to qualia instead of consciousness per se.

It is therefore impossible to argue coherently about whether it can emerge from a mechanistic model whilst staying within that paradigm. For this reason I depart from mechanism. I surmise that there is some "law" of the universe which goes beyond physics and forces the emergence of conscious experience in my brain and by reasonable (?) extrapolation, in the brains of other human beings and probably other animals. I do not have a metric for the intensity of consciousness, it seems reasonable to equate it with the level of mental activity but I cannot prove it, it's the best I can offer. A religious view might shift the problem away from the brain and say that humans are unique in having eternal souls but that does not answer whether non-human brains have non-eternal souls, i.e. conscious minds. Neither does the infamous "all information is conscious" claim help very much, as this would make a thermostat conscious. Sure, one could reckon a single bit to be conscious but a trillion times less conscious than you and I are. However there is a lot more information in one place in a big computer and it seems reasonable to assume that it needs to be organised in a specific way for it to become conscious. Anything else is science fiction.

In case you do not know already, my personal views are Christian, so you might expect me to be sympathetic to the soul-dualism theory. I am not. I can distinguish between the eternal or resurrection aspects of the soul - as required by my faith - and the simple fact of consciousness existing which is required by my knowledge of my own experience. I think the latter undermines materialist reductionism beyond repair, so I am not at all sorry about it, nevertheless, for a working theory I just assume that the consciousness of an entity is precisely formed on what it does, no more and no less.

Thus, at long last, I get to the point! We must give machines the benefit of the doubt. The more like us we make them, the more they are likely to feel the same as us. I have no worries about a giant computer suddenly becoming a feeling machine and taking over the world. If it has any consciousness, it is the consciousness of whatever alrorithms it is running and, frankly, I see no reason to think a Fast Fourier Transform is in any way associated with any secondary experience ("Wow! What fascinating spectral peaks!"). Self-programming systems might stumble upon the possibility though, I suppose. But this view does allow a giant abacus to have experiences if it is correctly structured. Indeed it then becomes hard to see where the line should be drawn between actually performing a conscious computation and performing a simulation - working out how the registers (or a system's state) would change, step by step, without actually having such a system running.

This sounds crazy, but it is the consequence of not having an objective theory and defaulting to the next best thing - the assumption that the laws governing human consciousness are universal. As for new colours you would have to have the information processing structures that correspond to the new colour. As far as we can tell, we are wired according to our three primaries, though perhaps tetrachromics grow four sets of circuits - brain tissue is very adaptable. So maybe new colours are a possibility though it's a bit of a stretch to think that LSD can do the re-wiring job, complete with all the conscious perception-y stuff in a few minutes.

derek_potter (not verified) | 01/29/13 | 06:35 AM

Link

he doesn't seem to agree with you.

I defend no position here but that about language games being language games, so how can he not agree with me? I agree with him in as far as he makes sense to me.

what would it be like if you have not 5, but more like >80 or so overlapping receptors? ... However you have this for hearing, and you could make up an 80 dimensional plot to show that the sounds you can hear are sooo much richer than the colors you see, but it sure doesn't feel like that.

Precisely! You get the "best comment" award again. See my replies above about that. Also: that a color space is 3D does not tell us how it looks like. The blue-yellow and green-red 2D sub spaces are clearly different from each other! How much is due to objective overlap shapes of response curves and how much is due to "subjective" association work (red blood, blue sky, ...)?

Because of that, I am not convinced that someone with 5 color receptors would have that much richer an experience, especially if the filters were close together and gave little extra uncorrelated information.

Yes, although, looking at that L and S overlap so much with the M response curve, it seems that you have the influence of the closeness of the filters upside down (?).

Sascha Vongehr | 01/29/13 | 01:21 AM

Link

Absolutely. There is something slightly odd about colour vision. Put it this way, the intensities of three primary colours or, better, the intensities of stimulation of the three types of receptor, span a 3 dimensional volume. This can be mapped to a 2 dimensional chromaticity diagram - there is a 3rd dimension, overall intensity, but colour perception does not depend very much on brightness at normal daylight levels so it can be omitted for many purposes. Even so, two psuedo-colours and brightness make three dimensions, so they span the same colour space as red-green-blue.

However, the eye makes absolutely no concession to spectral composition. A colour is always a single point in hue-saturation-intensity space. In contrast, sound remains fully analysed with a useful resolution of several hundred divisions for music, requiring a space of several hundred dimensions. An organism with a set of narrow band receptors might take the opportunity to see in colour "chords".

A mixture of red light and green light produces the same stimulation as a yellow, so there is no way the spectral information can be recovered. There is good reason for this. Three dimensions is three dimensions. Try as you may, there is no way you could ever separate the red-green components. Mapping red-green-blue to hue-saturation-intensity keeps all the information available, given the receptors. The alternative would be to have no overlap in the response and a blind spot around yellow - and/or no gradations of hue in that region, just as we have very poor colour resolution in the far red.

The nearest to we have to colour "chords" is the colour we see when we get a mixture of red and blue light in our eyes. The red and blue responses do not overlap very much so you might think there would be benefit in seeing the two primaries separately. However a mixture of just two primaries can be characterized as a single dimension, the relative proportion of each, plus overall intensity and this is indeed what our colour vision does: it creates a new range of colours we call purple. This is the straight line on the bottom of a chromaticity diagram. There is no such thing as monochromatic purple light (unless you believe in violet light).

So the big question is this: given narrow-band receptors, would there be any advantage in increasing the dimensionality (N) of colour space? It may be useful to allocate two dimensions - for brightness and saturation, leaving N-2 dimensions for hue. Unfortunately I am not a mantis shrimp and I have no way of knowing whether it uses its 18 primary colours to create a 16 dimensional colour chord or whether they simply provide hue resolution with less processing. If they are smart enough to train it might be possible to see whether they distinguish between bichromatic and monochromatic spectra. That should be worth a PhD for someone, not to mention a caper in a nice holiday location for some lucky post-grad or, more likely, her supervisor.

derek_potter (not verified) | 01/29/13 | 08:10 AM

Link

A mixture of red light and green light produces the same stimulation as a yellow, so there is no way the spectral information can be recovered.

You basically hold the position that the reported greenish-red is simply not to be called a new "hue" rather than some other, pattern-of-hue-induced emotion.

The red and blue responses do not overlap very much so you might think there would be benefit in seeing the two primaries separately.

Yes, much of the usefulness of the M detector is due to that S and L overlap too much. Another question is: Would the purple feel different in case green would be truly 0,1,0 and not like 0.5,1,0.5?

given narrow-band receptors, would there be any advantage in increasing the dimensionality (N) of colour space?

An evolutionary advantage can always be conceived of, some just-so story. Did not red-green come about especially because of fruit and foilage?

I have no way of knowing whether it uses its 18 primary colours to create a 16 dimensional colour chord or whether they simply provide hue resolution with less processing.

An interesting and almost depressing question (hue resolution with less processing = perhaps they see no colors). How do you get the number of primary colors (how defined) to 16+2=18?

Sascha Vongehr | 01/30/13 | 02:18 AM

Link

You basically hold the position that the reported greenish-red is simply not to be called a new "hue" rather than some other, pattern-of-hue-induced emotion.

If greenish-red is a satisfactory description to the subject then she must consider it to be a single hue. The map from spectrum to chromaticity diagram does not create any such point but a modified one would require another layer (another degree of freedom) and we could label the colour even if we could not print it.

I am not going to be drawn on the ontology of hues - you know I have an aversion to qualia :)

Yes, much of the usefulness of the M detector is due to that S and L overlap too much. Another question is: Would the purple feel different in case green would be truly 0,1,0 and not like 0.5,1,0.5?

I don't see why it should, although, of course, we cannot say what it "would" be like if our colour vision were different. Purples are a way of using up an otherwise unused segment of the chromaticity diagram to express the red:blue ratio. It would still be available if the 3 spectral responses were triangular or even rectangular.

given narrow-band receptors, would there be any advantage in increasing the dimensionality (N) of colour space?
An evolutionary advantage can always be conceived of, some just-so story. Did not red-green come about especially because of fruit and foilage?

No it was the other way round. Plants decided to colour up when animals started seeing colours. Before that everything was 50 Shades Of Grey :)

I am not talking about what advantage was actually involved in (our) evolution. I am talking about the a priori advantage of retaining all the information you have. A huge amount of information is thrown away if N>3 dimensions are projected onto 3 (hue, saturation and intensity).

How do you get the number of primary colors (how defined) to 16+2=18?

My bad. I should have said 16 primaries (separate channels with different spectral sensitivities). And 14 dimensions of colour left over if processing is done to create intensity and saturation.

An interesting and almost depressing question (hue resolution with less processing = perhaps they see no colors)

If we can credit a crustacian with enough of a mind to see colour at all, then it shouldn't really matter how or where the resolution occurs. If it chooses not to create a single hue out of its 16 channels, then perhaps it enjoys complex colour chords instead. You will have to ask the shrimp.

derek_potter (not verified) | 01/30/13 | 08:18 AM

Link

information is thrown away if N>3 dimensions are projected onto 3 (hue, saturation and intensity).

You mean if hue is kept 1D. On the other hand, the brain's information can likely always be expressed in terms of bits, so a coarse resolution in 4D would still be less than a very fine resolution in 2D for example. But how does it feel (you know, your favorite mystic qualia)?

I should have said 16 primaries (separate channels with different spectral sensitivities).

But they say that trichromatic vision has 4 "primary colors": yellow, green, red, blue.

chooses not to create a single hue out of its 16 channels, then perhaps it enjoys complex colour chords instead. You will have to ask the shrimp.

I would first look at whether the spectrum covers more than one octave and whether overtones (f' = 2f) are often excited. Perhaps one can claim: No overtones, no "real" scotsman music.

Sascha Vongehr | 01/30/13 | 22:02 PM

Link

On the other hand, the brain's information can likely always be expressed in terms of bits, so a coarse resolution in 4D would still be less than a very fine resolution in 2D for example.On the other hand, the brain's information can likely always be expressed in terms of bits, so a coarse resolution in 4D would still be less than a very fine resolution in 2D for example.

I rather doubt whether a bit count is all that matters. A lot of the brain is dedicated to extracting meaningful patterns and creating objects out of sense data. Whilst you might represent colour by repeating the intensity scale hundreds of times, once for each resolved hue, this would appear to be a backward step as far as picking out objects is concerned. After all, the name of the game is getting the job done, not encrypting the data in as awkward a way possible just because you can :)

But they say that trichromatic vision has 4 "primary colors": yellow, green, red, blue.

I was referring only to the three colour receptors.
I would first look at whether the spectrum covers more than one octave and whether overtones (f' = 2f) are often excited.
Overtones are often created with sound as components of the wave but never with light except in physics labs! Nobody knows why we find particular intervals harmonious with the possible exception of octaves and even then if the second harmonic is a natural interval for sound, then so should third harmonics be. Whilst there was a fad for fancy tuning schemes it turns out that we are perfectly happy with non-harmonic intervals and skilled singers and violists will alter the precise note according to context, quite often away from the mathematical ideal. Evaen pianos are routinely tuned with about a semitone drift from the bottom to the top! To sound "right".

derek_potter (not verified) | 01/31/13 | 11:04 AM

Link

I rather doubt whether a bit count is all that matters.

You talked about "information being thrown away", so the ball is in your court to come up with a measure that is more suitable. If it is not bits and qualia (as labels of the ineffible) you despise, what is it?

Overtones are often created with sound as components of the wave but never with light except in physics labs!

Which is why I hold the phenomena to be different and not that audio is automatically how a high dimensional color-space feels like. The harmonies relate to the overtone spectrum that is characteristically absent from dangerous sounds. This started the opportunity to increase interest by breaking the simple harmonies ever further with higher order harmonies and other deviations. That is completely absent from color phenomena.

Sascha Vongehr | 02/01/13 | 02:56 AM

Link

You talked about "information being thrown away", so the ball is in your court to come up with a measure that is more suitable.

I did retract the blanket statement that information is thrown away. However, according to the Principle of Charity which a certain contributor here recently mentioned , that puts the ball in your court to see whether the point can be restated better. I suppose I shall have to be charitable myself and spell it out for you :) [Just kidding]

Do not be mislead by the additive property of information counted in bits. I doubt whether the organism cares how many bits would be needed to encode the data if it had an analogue-to-digital converter handy. To encode, say, 100 hues, 20 levels of saturation and 50 levels of brightness would require 100,000 levels in a single dimension. That is asking a lot of a system which seems to like its information in quasi-analogue form. Even musical pitch has only a few hundred discernable levels - and it starts with data already split into a huge number of channels from the cochlea.

Then there is the problem of how to map a 3D space into 1D. The map of RGB to hue-saturation-brightness is smooth and robust in the sense that small changes in the RBG spectrum map to small movements on the chromaticity diagram. So the system is tolerant of noise and small errors. As far as I know you can't reduce the dimensionality of a simple data space smoothly. So if you mapped the 3D canonical box to a single dimension then the smallest error due to noise or encoding/decoding errors woulkd mean that instead of seeing a blue fish in the shadows get a little bit darker as it turns you might see it flash dazzling scarlet and then go black or green.

Furthermore, similarity of colours is important for other reasons - most colours are created by absorbtion from ambient - the intrinsic colour of an object (absorbtion spectrum) is much more important than the spectrum of the light reaching the eye. Computing a metric for colour distance would require at least as much processing as unpacking the data back into its original three streams. Chromaticity (the polar plot of hue and saturation) leaves this vital information immediately useable.

Encoding it to a single dimension would all be a great deal of processing just to make the system fragile.

If it is not bits and qualia (as labels of the ineffible) you despise, what is it?

I am sure there are metrics for the robustness of information but I don't know anything about them.

As for qualia, I have no objection to them as labels for the ineffable, I despise theories which claim that by labelling something you have explained it. And I despise even more theories which seek to explain something which is otherwise inexplicable by invoking yet another unexplained entity with dubious ontological status. God of the Gaps has nothing on Qualia of the Gaps.

Especially when they sprout fairy wings and start screwing with quantum probabilities. :)

derek_potter (not verified) | 02/01/13 | 10:14 AM

Link

Slow down - this is alpha meme - make it to the point! Of course the problem is not about some useless resolution that makes no difference to the organism. Still, given a neural network, you can estimate the bit number and processing speed by informed guestimates about time resolution for example (first wave theory will of course give you finer resolution than average firing rate, and connectionist calculation via dendrite trees is difficult to estimate, but surely, the visual nerve as a bottle neck is a good place to start, because once information is thrown away, it is gone). Certainly there is less qualiagoo if the bit rate processed is very small.

Sascha Vongehr | 02/01/13 | 23:37 PM

Link

Uh? Well what actually is the point? I made a case that the map from our 3 colour signals to 3D qualia with 1D of hue is robust and useful. What more do you want?

derek_potter (not verified) | 02/02/13 | 19:35 PM

Link

Sascha:

You asserted:

But they say that trichromatic vision has 4 "primary colors": yellow, green, red, blue.

Or, as you put it within your article:

The system of four “pure” hues (yellow, red, green, blue) in spite of three different types of receptors is a mostly cultural phenomenon. ...

Who is the "they" or the "culture(s)" that "say" this? I have never seen this assertion made anywhere. The closest I have seen is Newton's sevenfold colors of the rainbow: red, orange, yellow, green, blue, indigo, and violet (AKA red, orange, yellow, green, cyan, blue, and violet). (Personally, I would say there are more like six to nine colors, depending on contrast, and how vivid the rainbow/spectrum is.)

David

David Halliday | 02/09/13 | 13:10 PM

Link

Though they aren't felt as chords, and they don't have the same relation to frequency as auditory chords, our visual system does sense dissonances and harmonies. Red and green create a 'beat' just as surely as 400 and 402 Hz.

I accidentally created an odd 'chordal' effect in a little animation yesterday. Pure red stripes and pure blue stripes (rgb=100 and rgb=001) move across a box in such a way as to make an illusory bulge and sag in the box. (They also make purple, unsurprisingly.)
http://www.polistrasmill.blogspot.com/2013/01/why-ohm-met-resistance.html

polistra (not verified) | 01/31/13 | 05:35 AM

Link

harmonies should not be too loosely divorced from overtones (harmonics of a frequency), otherwise this is no longer science but art talk - like all is vibration and energy - no it is not - red and green do not create a beat

Sascha Vongehr | 01/31/13 | 07:01 AM

Link

I don't think there has to be consciousness (or qualia) in order to perceive these "non physical colors". A modern machine learning algorithm naturally learns to associate stimulus from different sources and compress them in to a "higher order sense". A deep artificial neural network would learn colors as features in the lowest layer. The color purple would be learned, simply by making the network process a lot of images.

Eirik (not verified) | 01/31/13 | 15:41 PM

Link

And why would that not be conscious?

Sascha Vongehr | 02/01/13 | 02:43 AM

Link

And why would that not be conscious?

Because it wouldn't.

derek_potter (not verified) | 02/01/13 | 10:03 AM

Link

Is that because you agree with your BFF under the white hat (see Helen's corkboard) that we may torture you to death once you lost your ability to communicate? It'll be fun - for us. ;-)

Sascha Vongehr | 02/01/13 | 23:28 PM

Link

No, I was calling you on your woolly reply. The pp had showed that perception of colour need not be concious, your response was to challenge him to say why his proposed expert system would not be conscious. There is a vast difference.

derek_potter (not verified) | 02/09/13 | 15:13 PM

Link

He proposed a lot more than a mere expert system!

Sascha Vongehr | 02/11/13 | 01:45 AM

Link

As far as the English language experience of color goes, the major colors work off of Pascal's Triangle - 1 full, 3 primaries, 3 secondaries, and 1 empty. Tetrachromats might have 1 full, 4 primaries, 6 secondaries, 4 tertiaries, and 1 empty. Of course, that also doesn't get into additive color versus CMYK subtractive color, nor that some languages use the same word for black and blue, or have two different colors for colors we would call blue (except that when you really start exploring the language, English has so *MANY* names for colors... How many people would argue that pink really is a separate color, and not just a shade of red....)

Color and color perception are inextricably tied in with language processing, as well as retinal cell activation. In a sense, wouldn't every human be a tetrachromate, as not only do we have the 3 types of cones, but we have rod cells which have their own chromatic sensitivity curve?

Tara Li (not verified) | 02/01/13 | 10:02 AM

Link

wouldn't every human be a tetrachromate, as not only do we have the 3 types of cones, but we have rod cells which have their own chromatic sensitivity curve?

Good point. Why don't you go ahead and investigate about at which lighting level and which region in the field of view (not many rods in the fovea, not many cones in the periphery) rods and cones contribute about the same, then excercise with the hyper pinks (as derived from the known response curves - you only need four colored lamps behind a diffuse screen) until you get your brain to give you novel colors? Your idea - you will be first with this as far as I know. It seems like a doable project I would have interest collaborating on.

Sascha Vongehr | 02/01/13 | 23:26 PM

Link

Off, then, to the first step of good experimental design - check the literature on what's already been found out about the topic.

It does seem like a neat topic, though.

Tara Li (not verified) | 02/03/13 | 15:05 PM

Link

Sascha:

For your edification, and that of your readers, for an N chromaticity system (N greater than or equal to one [1]), there are (N-3)2^N-2 + 1 "Novel Colors".

Now, as for any concept of "physicalism", I would have to say that without an infinite dimensional color "sense", one has no actual perception of the physical "colors" that exist, even within the narrow human visual range. All we have is but a cheap imitation of the physical, a mere low resolution "stand-in" for reality (of a sort, at least).

David

David Halliday | 02/03/13 | 20:04 PM

Link

You mean the number of "hyper-purples" or novel colors including all the browns? If you just talk about "colors", it is not clear why we should for example see yellow and orange as different hues. How many hues your neural networks distinguish is not given by any such formula but must be established for every person. Some have very few greens. The formula for how many purples there are (as I defined them) is of course a trivial exercise.

Sascha Vongehr | 02/08/13 | 23:19 PM

Link

Sascha:

You appear to be only replying to the first part of my message, above.

OK. I am referring to your “generalized purples” (at least for Trichromacy) and "hyper-purples" (at least for Tetrachromacy and beyond). In fact, I am referring to what are actually classes of such "colors" that are independent of intensity, saturation, and other particulars of actual mixtures of monochromatic stimuli: In other words, your class designated by "101" for Trichromacy; and "1010, but also 0101,1001, 1101, and 1011" for Tetrachromacy.

Yes, deriving the formula is a simple exercise, but since it is not likely to be trivially obvious to most of your readers, I thought it might be of at least some interest. If nothing else, it shows how the number of such classes grow with N-chromacy. (It also shows the correct number for Monochromacy, and Dichromacy: Namely, zero "generalized purples".)

David

David Halliday | 02/09/13 | 11:35 AM

Link

but since it is not likely to be trivially obvious to most of your readers, I thought it might be of at least some interest.

I know - and I am grateful for that - and to make sure everybody is clear about what you calculated, I made sure you explain that it is indeed the number of those "hyper-purples" I asked for. Perhaps you like to write your own article about how to derive it and then we can cross link. This subject is actually interesting and could lead to collaboration and new insights - although a lot is published on it, most is non-sense as far as I can see it. Tara's question on the other hand could lead to new effects to be discovered (you favorite empirical science, ha ha). I also think that nobody has seriously investigated the connection between color space and audio phenomena.

Sascha Vongehr | 02/11/13 | 01:56 AM

Link

Sascha:

Offhand, I am not certain how many hues (saturated colors, with intensity held constant) the "average" person can distinguish (excepting color blind individuals, of course). However, I do know that the number of colors the "average" person can distinguish is upwards of 300,000 to 10,000,000 (again, excepting color blind individuals, of course).

From the 300,000 colors, we can estimate that the number of pure hues is of the order of 20 to 70 (with a maximum to 500, for "average" non-colorblind people, using the fact that we tend to be far more sensitive to variations in intensity than to hue, and somewhat more sensitive to variations in hue than to saturation). From 10,000,000 colors we get on the order of 50 to 200 (with a max. of 3,000).

This corresponds roughly with the Wikipedia statement (under Color vision) that "the eye can distinguish up to a few hundred hues".

David

P.S. I except color blind individuals because 1) they are not fully Trichromats (more close to Dichromats or even Monochromats), and 2) there is significant variation within such individuals. Of course, we should similarly except Tetrachromats, since they could easily perceive more chromaticities, but I am not at all certain such were excepted from any such studies.

David Halliday | 02/09/13 | 13:01 PM