Photographic Turing test

Discussion in 'Philosophy' started by billy_mabrey, Dec 10, 2006.

  1. A previous post in the forum about photography as a language, and seeing
    brilliant phographs of Mars from the Spirit and Opportunity rovers (some
    perhaps automaticly taken?) have come together into a startling thought:

    Could a robot making photographs pass a photographic turing test?

    If you are not familiar with the Turing test, it is a test which blindly pits a
    human versus a machine in a communication match. If the person can talk with
    the machine and the machine passes itself off as human (ie able to respond
    accuratly to human conversation input) it passes the Turing test. This test is
    generally the accepted confirmation of Artificial Intellegence. So far as i
    know.....no machine has passed the test yet.

    I myself belive that our own minds are nothing more then the functions of
    biological machines, as our brains are made from basic natural elements that
    work with in the bounds of physical law. I believe a mechanical brain, fasioned
    by whatever means, if it is able to accuratly recieve input and "compute"
    answers that allow it to pass the Turing test, then it too will have the
    function of a MIND.

    Is it perhaps possible that a robot could make photographs at its own
    discretion and algorithmic judgement of composition and subject, and pass those
    photographs off as being created by a human conciousness thereby passing a
    photographic Turing Test? Or would a robot be merely an over the top overly
    sophisticated shutter button, initiated by the human who turned it on?
     
  2. I dunno Billy! But those robotic vaccuum cleaners do a great job!

    They have a fixed aperture and work in really low light.
     
  3. Any human could surely make photographs without any content, subject, intent of
    composition,...in such a way that they would be seen as ' could have well been made by a
    computer ' so in that way a computer, even if it's not passing the photographic turing
    test, could easily make photographs that could have been made by a human.
    The question whether or not a computer would pass a photographic turing test starts from
    the ( wrong ) point that photography is a language with a strict set of rules to follow in
    order to understand or make use of the language but photography isn't a language.
    Sure, you can communicate through photography and express ideas but photography
    doesn't stop being photography when it's done by a computer without the intervention of
    a human, whether or not the resulting photographs mean something or absolutely
    nothing.
     
  4. Phylo's got a good point. The crux of this question is in the watertightness of the analogy. The process in the Turing test is a dynamic interaction. Are we talking about some kind of to and fro between the human tester and the (hidden) robot photographer? How would the test be initiated? I'm not saying it's not a valid idea, we just need details.
     
  5. I myself belive that our own minds are nothing more then the functions of biological machines, as our brains are made from basic natural elements that work with in the bounds of physical law.
    Could they work any way other than through natural means? For your amusement, you might enjoy reading Roger Penrose's muse wherein he suggests we should look to the nanotubules of the brain to find if they can sustain a quantum state long enough so that the brain is "in touch" with quantum reality. (Penrose is a Platonist). Others have chimed in with the same thought, for example Stuart Hameroff.
    Unfortunately, the turing test is not applicable to the case you propose because there is no way to evince meaning or experience through the outcome - a problem in part due to the human beings who will judge; they would have to agree among themselves upon the goal and outcome so that the problem would be narrowed down to something quantifiable and almost certainly trivial.
    A better test would be for a number of robot-photographers to work among themselves to determine goals and outcomes. That's where language works.
    FWIW, certain programs have passed the Turing test - and it is amusing. One was a program playing a Rogerian psychiatrist and the other program mimicked a paranoid-schitzophrenic. And the ultimate chuckle is to have them talk to each other. (It is easier to model dysfunction than proper function.)
     
  6. There is a "fun" variation of this that would be a good trick and very easy thing to code
    up.. since Philip Greenspun is a nerd at MIT perhaps he could get one of the AI lab kids to
    code it up. (1) The photo.net forum provides a great "rating" system of 7.0->1.0 images
    which could be used to build a neural net .. which basically is a "training" mechanism to
    make a program "distinguish" the highly rated photographs from low rated photographs ..
    once the program is trained (and tested (done exactly how you would suspect, trained on
    1/2 of the data, tested on the other 1/2) ) .. you can build a robot that goes around
    randomly taking photos (perhaps using a auto-low res recording mechanism aka robotic
    eyes (also trained) to predict whether a photo taken of the scene will be highly rated or
    not.. and after it takes the appropriate set of photos and rates them.. it will submit to
    photo.net (anonymously!!) and the users will decide appropriately..! If it does better than
    your average photo.net submitter the coders have done a reasonable job in the 'photo.net'
    turing test

    For extra credit, the same robot will do basic manipulation in photoshop (via scripts) to
    further enhance the rating... (as a start you can imagine contrast/saturation)

    I could code this up actually.. but this would be a good project for an undergrad/grad
    student somewhere..

    -avi
     
  7. Phylo, i believe photography or any image making IS a language, as it is a form of communication. But you are right, we have to make an assumption for this question. How clearly or unclearly communication is carried out in photography is up in the air...but that was all the topic of a post from a few days back...

    As for the machine/mind thing...I might take a look at that Penrose, tho still, i think a functioning-mind-machine-robotic-camera could get in touch with whatever quantum states he speaks of, especialy if our brains are capable of doing so. Quantum states are still within the realm of physical law, even if we can't wrap our mathematics fully around it yet, (else we would not be here now).

    Avi has a great proposal! That way there will be some kind of learned commonality of a photographic language. I heard of a similar thing being done not so long ago where a program learned succesful POP songs and was used to judge whether or not a newly recorded song would make it or not.
    Sounds like a thrilling project!....i swear i should've gone into engineering
     
  8. I could code this up actually..
    Oh could you?
    It is that simple, eh?
     
  9. Hi sorry i didn't mean that to sound as arrogant as perhaps it came off.. if i did it myself it
    wouldn't work at all obviously but it would be passable :)

    Just so you know I am a faculty member in a physics department (not at MIT!) (basically
    doing computational/numerical/theoretical biophysics) .. which is to say i dont code at all
    anymore but i do enslave graduate students to code for me :)

    While we don't use NN type tricks in our line of work, its very popular in a lot of 'heuristic'
    fields (and many that should not be that heuristic, like chemistry, economics) .. the code
    to do the pattern matching in this day and age is very easy and there are many examples
    of doing just that on pictures out there, so it would just be a case of replicated someone
    elses work and moreover there are a lot of NN libraries out there that you can use Gluing
    it to a robot/camera is somewhat more work but i'm sure there is someone in one of the
    Eng. departments who could sort it out..

    In all seriousness i think if you hijack an available NN/picture analysis code it would take
    a weekend, even w/out that NN code itself isn't really that complicated .. its just glorified
    curve fitting.. you know..

    Incidentally, I'm not a big fan of penrose but that is a discussion for another time probably
    :!

    -best,

    -avi
     
  10. btw an aside (i just went back and read the your post a bit closer)

    my research is actually on understanding (classically) how complex protein networks work
    (i.e. receptors, proteins interacting) .. the generalization to neurons is straight forward and
    is a current project (though we are about 10 years away from having the computational
    power to do this properly). the generalization to multiple neurons is also straight
    forward..(though we do not nearly have the computational power, particularly on the scale
    of the mind). QM does not come into play at all, nor does it need to as far as I can tell.
    Perhaps, in the end this line of thinking is all for naught and penrose and his possee are
    right: we need "QM nanotubes" to vaguely, and non-specifically magically endow
    "consciousness" to such a simulation.. but its an interesting game btwn penrose and
    everyone else :)

    Btw not to be obvious, the only thing QM gives you is nondeterminism. Why
    "consciousness" needs non-determinism to be conscious is not particularly clear in any of
    penroses work, and regardless, even if "consciousness" was dependent on non-
    determinism/random behavior, it would be far simpler for biology to use classical thermal
    noise than having to get fancy and use "QM nanotubes" as its source .. particularly as there
    is good evidence that thermal noise/stochasticisty plays some role in simple biological
    processes already.
     
  11. Btw not to be obvious, the only thing QM gives you is nondeterminism. Why "consciousness" needs non-determinism to be conscious is not particularly clear in any of penroses work, and regardless, even if "consciousness" was dependent on non- determinism/random behavior, it would be far simpler for biology to use classical thermal noise than having to get fancy and use "QM nanotubes" as its source .. particularly as there is good evidence that thermal noise/stochasticisty plays some role in simple biological processes already.
    Why one or the other? Why not both qualities?
    But we can save this subject for another time. To give an inkling of what I'm dieing to ask you: what puzzles me is Penrose's description of the single-celled ameboa and how it "behaves" (avoids pain/precursors to certain damage, and feeds) when it has no nervous system, but it does have microtubules (admitedly functional for reproduction) - so he wonders if the little critter is functioning in accord with a quantum space. That's his Platonic take on it.
    But I'm mired at the moment in work, programming and can't give it more time. Perhaps another time? Please?
     
  12. You didn't sound at all arrogant, Avijit. As someone who spends a lot of my time coding, I can see where you're coming from. Once you know where the pieces are, gluing them together isn't that hard, unless you're aiming for bullet proof code, of course.
     
  13. Would you believe that several such systems have been created over the last few years. The most famous is probably "Lewis" from Washington University in St. Louis.

    http://www.cs.wustl.edu/MediaAndMachines/Lewis/

    Truth, as they say, is stranger than fiction.
     
  14. It can be both, and quite honestly.. *everything* is really quantum in reality.. by
    everything i mean the idea of two chemicals coming together and making a new chemical
    (i.e. any basic chemistry which of course includes biological chemistry) is really a quantum
    process and can't really be treated classically at all. However penrose doesn't mean
    "quantum" in that sense, penrose means quantum in the "collapse of the wave function/
    non-determinism" sense ..

    in the same way really the person from I think arizona state who is currently making waves
    among philosophy circles (whose name totally escapes me at the moment but he is
    australian) who apparantly has resurrected dualism by some magic quantum process.. This
    line of thought seems to link (in a way i do not really understand, which is perhaps why i
    am so skeptical) that collapse to "human consciousness" .. Microtubules are used for
    transport and cytoskeleton rearrangement.. i may be wrong but i've never seen really
    non-classical behavior from things that big or tubules (other than chemistry) .. there is not
    exactly big delocalized electrons hanging all over microtubules i dont think

    Interestingly, chemotaxis ..(as per your amoeba question) aka avoiding and being attracted
    to chemical phenomena and/or danger etc can easily be explained by "simple" chemistry
    and transport .. that is its just receptors, and protein-protein interactions.. these are
    models of "chemotaxis" by thinking of what the biology is doing .. and no need for
    neurons ... purely classical in fact

    Neurons are *a* way to gain memory .. and well more versatile .. but not the only way. As
    an aside actually it was Turing who actually first realized the complexity of "simple"
    chemistry and transport phenomena leading to "surprising" behavior aka "turing patterns"
    (A famous example here is very simple chemistry and transport leads to "leopard spot
    patterns" ) .. everything else is really just an extension of his work actually..


    I guess w/ regards to it being "both" at the end of the day its not really clear to me what
    one is gaining by having this aspect of 'quantum phenomenon' necessitate consciousness
    other than the fact that thinking of ourselves (and our thoughts) as 'clocks' in a practically
    deterministic world seems to make people a bit squeemish .. but if it seems to describe
    all the behavior i would bring up occams razor and say there is no need to introduce any
    more complications until its absolutely evident that one has to, and i'm not really sure we
    are at that stage yet.
     
  15. "... pass those photographs off as being created by a human conciousness thereby passing a photographic Turing Test?"

    The point of the Turing test is not whether the machine can pass off its response as being convincingly human, as much as whether or not the human can catch the machine out: i.e. use the interaction to elicit from the machine a response which would not convincingly have been given by a human.

    Programs that have been claimed to "pass" the Turing test have invariably been given a restricted version of the test.

    Turing's test ("The Imitation Game"), published in 1950, appears in The Mind's I, an interesting anthology by Hofstadter and Dennett which appeared in 1981. The test is quite subtle, or at least more so than is commonly credited.
     
  16. H. P. [...] As someone who spends a lot of my time coding, I can see where you're coming from.
    COBOL? What do you code that is original? Tell us.
    --
    Pico - who did once write some COBOL but I swear I never compiled it.
     
  17. avijit ghosh. It can be both, and quite honestly.. *everything* is really quantum in reality.. by everything i mean the idea of two chemicals coming together and making a new chemical (i.e. any basic chemistry which of course includes biological chemistry) is really a quantum process and can't really be treated classically at all. However penrose doesn't mean "quantum" in that sense, penrose means quantum in the "collapse of the wave function/ non-determinism" sense ..
    Penrose has agreed to that. He has written so. Simply saying that all chemistry is (somehow, not explained by yourself) related to QM is not making a case at all because your statement is dismissive without a rationale to consider the specifics.
     
  18. "Penrose has agreed to that. He has written so. Simply saying that all chemistry is
    (somehow, not explained by yourself) related to QM is not making a case at all because
    your statement is dismissive without a rationale to consider the specifics."

    I am not really sure what you are saying. Penrose has agreed to what? Anything having to
    do w/ electron flow is inherently QM in nature, the common examples include bonding
    processes. This is not the kind of QM Penrose is talking about when he invokes "nano-
    tubules/microtubules" The "nanotubules" for Penrose are supposed to be something akin
    to quantum dots i.e. delocalized electron densities whose "collapse" confers some sort of
    behavior with regards to consciousness.
     
  19. Actually let me follow up. I do not think you need to invoke the rather non-intuitive nature
    of QM to get consciousness. Penrose believes you do. Thats why he sees "QM collapse/
    non-determinism" as the underlying agent of bacteria chemotaxis, neuronal behavior and
    so forth. Why you do, as far as I can tell, has never been explained by Penrose to at least
    the satisfaction of me. You might argue that I am saying the equivalent but opposite
    statement and therefore making an assertion on that is equivalent to penrose, but I am
    not. I am saying you do not "need" QM to invoke conscious, but that does not imply "QM"
    can not have anything to do w/ consciousness. Penrose is making the positive assertion
    somehow wave collapse has something to do w/ consciousness so the onus is on him not
    me. Plus he is more famous so its up to him to carry a good argument :)
     
  20. Probably the SNOBOL and SPITBOL programers didnt have time to worry about such issues.
     
  21. Kelly Flanigan Probably the SNOBOL and SPITBOL programers didnt have time to worry about such issues.
    I do go back far enough to know SNOBOL. In fact, I have the first edition of the manual. It was typewriter done and then photographed. The Big Orange Book. I'll be putting it on that auction site after January 1.
     
  22. avijit ghosh:

    Good points, and I am not in a position to represent Sir Roger Penrose. The statement that I recall regarding Roger, QM and another science was pointed to a biology professor who claimed that her discipline was not a part of physics. (It must suck to be her.)<p>
    Regarding the 'collapse', please consider that organic forms sustain the outcomes for a long time and therefore create a kind of contiguity for the quantum outcome... for better or worse! <p>
    On a completely unrelated vector - I must admit I am a Penrose fan and possibly the worst reasons. I was a "lecture thief" at Oxford. I'd sit outside classrooms and listen in, avoiding expense and exams. :)
     
  23. Its okay both my GF who is a borderline "philosophy of science" type and my cousin
    (whose advisor was Searle) and I have a habit of getting into many terrible drunken
    debates that are somewhat akin to this :)

    If you like Penrose, you would like this modern dualism debate. It is really interesting
    actually .. it has a lot of similar parallels.. i think i was struck by it as I didn't realize there
    was any such thing as a dualism debate in the modern age.. ! (such is the arrogance of
    materialist physicists :))

    Oh now i remember the guys name is *David Chalmers*. Really fascinating reading.. very
    much in the penrose camp .. weirdly (or not) he actually was a math. /cs undergrad and
    got his ph.d. w/ Hofstadter over at Indiana. He is now at arizona state directing the
    "consciousness center" or something.. He has apparantly made serious waves to the
    extent that there have been nytimes articles written on him.

    p.s. f77 all the way :)
     
  24. He is now at arizona state directing the "consciousness center" or something..
    Ah jeeze! You aren't refering to Lyn Nadel are you? He's the guy who gave the nod to Gary Schwartz to study afterlife phenonema. No? I don't mean that in a bad way.... although Gary... he believes in mind reading so easily that it seems to be his fundamental given belief, and THEN he becomes a skeptic!
    What an interesting place this has become.
    Okay, I'm outta here to work in the shop. Building a camera. Someting tangible.
     
  25. Oh. Maybe Nadel was earlier. I know of Chalmers' writing. Never seen him IRL.
     
  26. Well, we won't know whether or not there really is a rover on Mars taking photos until we spend a trillion dollars and thirty years to put a manned mission on Mars...


    Also, let's goggle for turing machine...
     
  27. I think, Avijit, that you are falling into the Pico trap. It's pretty obvious that, whoever he is, Pico is a fantasist of the first order. He claims to have done everything and known everyone but I suspect he's some sad, fat old bloke who's never done much with his life and now uses the anonymity of these forums to act out his fantasies. Frankly, I find him funny these days, a sort of human reverse Turing machine. He gathers his material from all sorts of places and is programmed to react viciously to any response to his nasty comments.

    As you're obviously someone who knows what you're talking about, Avajit, how would you define the border between reaction and consciousness? Do you think there's a sudden change from the automatic reactions of a single cell to the self awareness of the human brain or is there a gradual move as complexity increases, with no discernible point at which instinct becomes consciousness?
     
  28. H. P.: I think, Avijit, that you are falling into the Pico trap. It's pretty obvious that, whoever he is, Pico is a fantasist of the first order. He claims to have done everything and known everyone but I suspect he's some sad, fat old bloke who's never done much with his life
    Strike two.
    Avajit, how would you define the border between reaction and consciousness?
    In your case, at this moment, I'll guess the reaction borderline is the second pint, when your consciousness winked out.
     
  29. Billy, to return to your original post, it occurs to me that there is an interesting question at the end where you ask: "Or would a robot be merely an over the top overly sophisticated shutter button, initiated by the human who turned it on?"

    I don't know about you but I have this science fiction idea of a robot as an autonomous machine, functioning totally without human intervention, which Isaac Asimov, among others, popularised in the 'forties and 'fifties. So it seems to me that the Turing Test for Cameras might be whether the viewer could detect that a human had not initiated the picture taking. Conversely, one might ask, as you do, whether the viewer would look at the image and say to themselves that it was a completely mechanistic representation and therefor the work of a machine.

    I think, as Phylo says, that there isn't really any syntax to image making and you could certainly fool a viewer into thinking that a picture was initiated by a machine. Going the other way, I suspect that even a short 'conversation' would reveal the mechanistic origins of the photographer, unless the controlling database of rules was very large, so that the style of photography changed subtly from image to image.
     
  30. So it seems to me that the Turing Test for Cameras might be whether the viewer could detect that a human had not initiated the picture taking. Conversely, one might ask, as you do, whether the viewer would look at the image and say to themselves that it was a completely mechanistic representation and therefor the work of a machine...
    You don't understand what the Turing Test is.
    By your misinterpretation of it, most machines could pass it.
    Turn that around - what can a human make that looks exactly what a sophisticated machine can make. When you come up with a 100% handmade Leica, let me know. Best get to work digging for ore. You've a long haul.
     
  31. HP : I dont know to be honest.. if i were guessing there is probably a gradual transition,
    which is most likely dependent on the "vagueness" of the term "conscious" .. actually to
    clarify i think there is a gradual transition from 'unconscious beaker chemistry' to
    consciousness

    though would also hasten to add i dont think "consciousness" is outside of "chemical
    reactions" (including quantum nanotube postulations depending on your fashion sense ) :)

    btw as Joseph pointed out the robot is already here .. its just a question of gluing a
    photo.net aesthetic detector to it .. does greenspun actually ever read these posts? would
    this be something he or photo.net would be interested in? I happen to (vaguely) know the
    kismet/kosmo robot people (Sherry Turkle is my gf's advisor and i'd guess the STS people
    would like this sort of stuff ) ..

    i will write up a paragraph to forward to her as a project (i do not want to be involved but
    i think its a cool idea) if someone from photo.net would be willing to let the project use
    the photo.net database as a training set .. Turkle is from MIT if that helps..

    i was thinking this would be a great thing to try and get someone like canon to sponsor,
    but then you might imagine "Canon : we are replacing our customers with robots" might
    not be a good advertising slogan :)

    -avi
     
  32. I take Emergent Behavior to be a term one uses when it is not necessary to describe the
    specific means by which certain behavior occurs which is different than it's parts' behavior.
    Whether the term is being used to avoid a necessary understanding or not depends upon
    the specific discourse. A person who reflexively explains behavior as "emergent" should
    be prepared to explain exactly what Emergent means in the case at hand.

    As to consciousness arising from chemistry, well a chemist could say that and a physicist
    might say it arises from physics. Isn't all chemistry physics at the root? :)
     
  33. Getting back to the original question and ignoring the very interesting and informative variations on the theme:

    "Could a robot making photographs pass a photographic turing test?"

    Yes. We are talking about a form of art, which has no objective criteria other than what it chooses to accept. There is no technical impediment to the robot's picture taking. The photographic process could be randomized to prevent duplication. Therefor, the robot would produce images. These images would then be viewed by humans who would respond to them by liking them, hating them, or some other feeling somewhere in between.

    Some, if they felt it improved their social standing as in The Emperor's New Clothes, would pretend to like the robot-produced images.

    I forget who first observed that if an infinite amount of monkeys were given typewriters, one would write King Lear.

    The photo-robots could certainly keep my ego in check!
     
  34. I think back to High School introductory photography class, where a quick overview of rule of thirds, exposure, etc. were given and then students were turned loose on the world. Most of what came out of the darkroom a few weeks later was what I call "faux artsy"... ie they used techniques that can make really great pictures (the camera was rotated a little bit, or the main subject was out of focus, an extreme vantage point was used, very over exposed or under exposed etc...) However, these techniques were used haphazardly and with out any real vision, with the expectation that any image that used any of these techniques was "artsy". One could easily build a robot that would then RANDOMLY choose to use some of these effects to various degrees, and produce what would easily get into a High School art show (@ the non-art school I went to, but still).
     
  35. "Beauty is in the eye of the beholder" so in a sense it's not the mechanical photographer that should be undertaking the Turing test, but the viewer. Seeing a set of 100 mechanically taken photos we could well pick out 10 "artistic" ones but it is in this selection process which defines the art. Effectively they are objets trouves.
    On the broader subject, I think Turing proposed his test to define intelligence but accepted that it does not necessarily imply consciousness.
     
  36. Here are a bunch of links to robot cameras in places like airports all over the earth. Are any of these images great art? Perhaps not, but an interesting way to kill time.
    http://www.webcamplaza.net/master_frame_fix.html?http://www.webcamplaza.net/cams/airport.html
     

Share This Page