You, too, can be an agent of selection


Here’s a website of mutating pictures, a collection of images made with a splatter of scattered triangles. Your job is to browse through them and score them for how much they resemble a face — which isn’t easy. If I stare at any random pile of symmetrical shadings, they all start to look like faces to me.

Anyway, pictures that get higher scores produce more progeny, with slight mutations, in the next round of picture generation. You can see where this is going…

Comments

  1. says

    I think the “progress” page is interesting to view– they’ve moved from blobs of polygon, to something like ugly transformers… it’s slow going, but I’ll definitely be going there each day to help out.

  2. bunbuns says

    How exactly do you rate it? Am I missing something? The page that I gets asks me to rate it, but there isn’t any way to do that, unless I’m supposed to send an email?

    Pretty cool website though

  3. says

    Some look a bit like Anime, others I don’t see a face at all. I stop rating them after about 10. Will it keep asking you until you give up?

    How exactly do you rate it? Am I missing something? The page that I gets asks me to rate it, but there isn’t any way to do that, unless I’m supposed to send an email?

    Pretty cool website though

    Click on the numbers at the bottom to rate it.

  4. HP says

    I found the context-free subjective rating a bit awkward — “Are those eyebrows or shoulders?” I’d like to see the same kind of mutations, but using the “Hot or Not?” methodology, showing the user two images and asking “Which one looks more like a face?” Then prioritize the next round of mutations based on total votes per image. I suspect you’d converge on face-like images a lot faster this way.

    It’s a bit teleological, though, isn’t it?

  5. zayzayem says

    Wow. This is pretty awesome project.

    It’s Rorschach, Darwinism and freaky Impressionistic art all rolled into one.

    The first one I had looked really face-like. But I looked at the random five original ones, and saw one that looked even more face-like.

    Everything looks like a face. I’m reminded of an incident in Carpe Juggulum. “Aaaargh. Everything is a religious symbol!!!!”

  6. says

    Excellent project anyway, a long overdue online answer to Dawkins’ blind watchmaker program. Making the principles of evolution interactive has to be a good thing.

    I agree with HP: it would be easier to give meaningful ratings if asked to compare two images and select the most face-like one, http://kittenwar.com/. I also think this approach would result in reaching ‘realistic’ faces faster.

    I found i was selecting for resemblance to transformer-ish robotic faces, because the geometric splinters lend themselves well to that kind of ‘look’. If the splinters were organic-looking blobs instead, i expect things would work out quite differently.

    Perhaps applying a filter, like photoshop’s ‘median’ over the generated splinters would be an economical way to experiment with blobs. Then you might end up with images more like this http://tinyurl.com/2fwk9j instead of this http://tinyurl.com/28mthj

  7. says

    Can we force extinction if we constantly rate them at 0?

    This is pretty interesting, especially since there IS no scale. It’s purely up to the (current environment)/ (individual rater) to select which ones are “fit”.

  8. Avenel says

    Reminds me of the crabs in Japan that have been selected to have faces on their carapaces. Seems a lot of samarai drowned in the area, and local fishermen got it into their heads that some of the crabs were reincarnated samarai. If the crabs shell looked like a face, they threw it back.

  9. Ragnor says

    What does it say about me that I think that the more “facelike” ones look like samurai?

  10. says

    This is a fabulous project. I am addicted. I wish projects like this were more visible to the general population because they demonstrate very clearly how natural selection works, specifically the tiny alterations that occur in each generation, adding up to major changes down the line. That was the bit about evolution that I only found out about a few years ago and it was so simple, so elegant and, once you knew it, so obvious that it changed my thinking about the world overnight (and made me an atheist!)

  11. says

    This is cool, and I’ll be watching with interest. But I echo HP @ 6: it’s not really natural selection, it’s kind of artificial in the same way as Dawkins’ WEASEL program. Still a fascinating evolutionary experiment; we’re telling it what looks ‘facey’ without specifying what exactly, and we’ll see an increase in ‘faciness’ from generation to generation. Can’t wait to see what faciness looks like!

  12. Christian Burnham says

    Anyone remember the genetic poetry page? You got to vote on the best of two poems each time. They started off as random linkages of words but evolved into quite credible poems.

    Unfortunately- the page seems to have been lost to internet erosion.

    It seems more accurate to me to vote for the best of several faces than to rate each on a scale of 1 to 10- but I’m no jenetix expert.

  13. Sven DiMilo says

    I think it’s not necessarily teleological, in the sense of having a defined goal. For one thing, thre is not a particular face everybody’s trying for (like Dawkins’s example had a particular phrase); it’s more of a vaguish pattern. Also, the Japanese crabs referenced above provide a very relevant example–crabbers are selection pressures that favor a phenotype that looks face-like (in that particular environment), even though nobody is trying to evolve a more face-like carapacial pattern in crabs. Or, maybe think of it as intersexual selection, with a pre-existing bias.

  14. says

    True.

    Half millionth comment coming up soon! I think the contest must training us, but that’s behaviouristic and not evolutionary.

  15. says

    Here’s a more well-developed project along the same lines that’s been discussed on one of the discussion groups I moderate:

    http://www.picbreeder.com

    There are a few sites like this out there…pretty interesting, and possibly a nice demonstration of artificial selection for a biology class.

  16. ctenotrish, FCD says

    Fun! Mostly the face-looking ones look like anime cartoons, though. Maybe they’ll start looking more ‘real’ with continued selection!

  17. says

    > Can we force extinction if we constantly
    > rate them at 0?

    Heh, no. The way it works is that members of the population simply age, increasing their chance to “die” (by being replaced by new offspring, as the overall population size is constant at 1000). So you can prevent a picture to have offspring, basically “killing” its genes — if no one else generates offspring for it — but you can’t directly kill an image. Though at this point, the average time for a full generation to be replaced is just around 25 minutes, so images not creating offspring are aging quite quickly :)

    > Mine froze, however, after 20 faces.

    Hmmm… can you email me any additional details about the freezing you might have? [email protected]

  18. True Bob says

    Ramblin Dude @ #11.

    I for one will not welcome those overlords. Not only are there NO TENTACLES, but it shows the lack of initiative on the part of Murka’s mad scientists.

  19. techskeptic says

    uhhh… I hate to say this…. but isn’t this intelligent design?

    Aren’t we designers pushing evolution in one way or another?

  20. says

    Aren’t we designers pushing evolution in one way or another?

    No, we’re a selection process. You have no input into the actual configuration of each image, just in the likelihood that whatever algorithms used to create it are used in the construction of new ones (or whatever the rating does.)

    Think of a warming climate. It doesn’t design forms to deal with the heat, but instead selects for those organisms which are better equipped.

  21. Brendan S says

    #24:

    I’m sure that’s what ID advocates will say about it. It’s pretty close to Theistic evolution (God designed a system, then it ran). Plus the selection process involves ‘intelligence’. Although I think I’m giving too much credit to random web denizens there.

    I do like the idea, however. It’s just that there are numerous genetic / evolutionary algorithms in computer science, and thus far none of them have had the ‘AH HA!’ effect that people would like on the evolution deniers.

    It is really cool, though.

  22. says

    Look, I don’t see how intelligence plays into it at all. We may be, but all we’re doing is voting based on how much the organisms meet some condition.

    The ID advocates who claim this is an intelligent design process would be stupid to claim this is anything like ID, because as a process it’s indistinguishable from any non-intelligent selection process. They’d be claiming, in effect, that non-design evolutionary process can look like design ones, and therefore invalidate their entire bogus theory.

    Plus, I’ll laugh at them and call them silly names. Nobody wants that.

  23. says

    Ooo, I got a good one.

    She’s been staring at you all evening. There’s a lot of debris around her, and her makeup is weird, but you can’t mistake that hungry look in her eyes and that mocking smile.

    I give her a 9.

  24. True Bob says

    I give up. I do not have the patience for this project. Imagine, some people think a magic skypappy did this for m/billions of years, to every living thing there is. Makes ID look attractive – over and done with.

  25. says

    I’d disagree that this process is indistinguishable from “non-intelligent” design processes. It’s the different between natural selection and artificial selection. It is doubtful the earth would ever have seen something that looked like a poodle if humans weren’t around to selectively breed them.

    It seems to me there’s a spectrum of agent intervention in a design process, from pure, agent-planned design, like an engineer planning out every parameter beforehand, to something like artificial selection, where the engineer sets up constraints but allows parameters to fluctuate based on random mutation, to natural selection, where there is no intrusion on the direction of design by an agent.

  26. Benjamin says

    Great post! I just watched COSMOS recently with Carl Sagan, and this is just like the segment he presented on the samurai crabs in Japan.

  27. says

    It is doubtful the earth would ever have seen something that looked like a poodle if humans weren’t around to selectively breed them.

    Are you kidding? Just look at star-nosed moles, deep-sea anglerfish, humans (as compared to other mammals), or pretty well any collection of benthic weirdos. Hell, it turns out the designers of the alien from the eponymous scifi/horror movie, using all of their nightmarish creativity only managed to recreate something the natural selection already produced.

    Without prior knowledge of an agent at work, would we really be able to differentiate between artificial and natural selection?

  28. says

    Yeah, we should definitely be able to distinguish between artificial and natural selection, especially if we look at other factors such as the time span in which change occurred and attributes aside from anatomy, such as behavior. You think an alien biologist who landed on earth would be able to explain the current state of diversification of dogs in the timespan in which it occurred by any means other than artificial selection?

  29. CJO says

    It is doubtful the earth would ever have seen something that looked like a poodle if humans weren’t around to selectively breed them.

    No more so than it is doubtful that the earth would have ever seen something like an orchid wasp, if there were no orchids, or orchids if there were no wasps, or…

    Co-evolution. That’s all. I son’t see how the relative intelligence of a given species involved in a symbiotic evolutionary relationship is relevant. Artificial selection is a subset of natural selection. Clearly there are going to be some cases, like poodles perhaps, where you feel like there’s a clear distinction to be made. But humanity is a fertile niche for countless species. Some we call domesticated, some we call pests. Some we hardly notice. I maintain that there are no clear lines that can be drawn to separate out all of these cases.

  30. Mike Fox says

    Hmm … interesting, but quite limited. I tried selecting for eyes on stalks, but the frequency stayed about the same.

  31. Rey Fox says

    “Without prior knowledge of an agent at work, would we really be able to differentiate between artificial and natural selection?”

    Just run everything through Dembski’s Amazing Design Detector, remembering to calibrate by Behe’s Fudge Factor.

  32. says

    CJO, you think there is no conceptual difference between an orchid and a wasp coevolving and humans selectively breeding dogs or livestock? Seems to me that there’s a qualitative distinction between two species interacting to further their own reproductive goals and influencing the gene frequencies of each others’ species, as opposed to one species controlling nearly every aspect of the other’s reproduction, exerting a much more one-sided selective force.

  33. says

    Or to use another example, is there no qualitative difference between observing the evolution of fruit flies in the wild, versus selectively breeding a controlled population of flies in the lab for a particular trait modification (e.g. larger wings or eyes)?

  34. says

    Yeah, we should definitely be able to distinguish between artificial and natural selection, especially if we look at other factors such as the time span in which change occurred and attributes aside from anatomy, such as behavior.

    Well, timespan may be a good indicator, but are you claiming in the above that behaviour is the product of artificial and not natural selection? (Please say no, please say no; because I don’t want to have to get all up in yo’ grill.)

  35. says

    Further Derek, in talking about the image selection website here, if timespan is the only thing that can be used to distiguish natural from artificial selection then my point that the IDists would be stupid to claim this as an example of ID stands.

  36. says

    Any attribute is subject to selection…are you saying particular behaviors can’t be selected for artificially? What about the tameness of domesticated animals? Or the responsiveness and attentiveness of dogs to their human owners? I’m not saying these behavioral traits are not complex, but definitely behavior can be modified, sometimes drastically, by artificial selection.

  37. says

    I could see some semi-clever IDer asserting that god’s role in the evolution of life on earth is analogous to artificial selection. I don’t think I’ve seen that particular argument, but I don’t see why it couldn’t be made. In fact, I’m not sure exactly to what extent people like Behe think that god intervenes in evolution…whether they think he comes down and rearranges genes or modulates the environment to tweak selection in particular direction. Either way it’s silly.

  38. says

    Any attribute is subject to selection…are you saying particular behaviors can’t be selected for artificially?

    No, I’m not saying that; you’re completely correct in saying they are. I’m just saying that they are also selected for naturally, and so the existence of behaviours modified over time are not a necessary indicator of artificial selection.

  39. CJO says

    CJO, you think there is no conceptual difference between an orchid and a wasp coevolving and humans selectively breeding dogs or livestock?

    “No conceptual difference” is too strong. Otherwise, why even recognize a subset? Obviously, an ecology dominated by a widespread, technological species with strange tastes in canine beauty is going to turn out some oddities, and rapidly, as against geological timescales.

    My point is really just that no clear distinction can be drawn that will clearly delineate cases of “artificial” selection from ordinary “natural” co-evolution. Squirrels are quite a bit more intelligent than oak trees, for instance. But we don’t say squirrels selectively breed trees for bigger, tastier acorns.

    Re: fruit flies, From the point of view of drosophila genes, aren’t genetics labs just a niche? With sugar water everywhere, free for the taking, and a distinct lack of swallows and dragonflies?

  40. says

    My point is really just that no clear distinction can be drawn that will clearly delineate cases of “artificial” selection from ordinary “natural” co-evolution.

    Well, that’s true of all language and all scientific terminology. Virtually any term that attempts to capture real-world complexities will deal with ambiguity and fuzzy boundaries, but we are usually able to agree on the majority of cases that don’t fall along the fuzzy border, and continue to hash out the ones that do.

  41. Jess says

    All I saw was Praying Mantis after Praying Mantis. And a Transformer or two thrown in for good measure.

  42. CJO says

    Derek,
    Fair enough. The area of artificial versus natural selection is one in which I feel the “fuzzy boundary” encompasses more cases than is commonly recognized.

  43. windy says

    You think an alien biologist who landed on earth would be able to explain the current state of diversification of dogs in the timespan in which it occurred by any means other than artificial selection?

    What would the alien biologist say about the timespan of speciation in Lake Victoria cichlids?

  44. bunbuns says

    To #4 – I would click on the numbers, but I don’t see any when I click on the link. Perhaps it’s because I’m using Internet Exploder… Will try with Firefox once I get home.

  45. Hank says

    Whatever validity this has on anything else, it’s a pretty sweet experiment on inducing clicking behaviour in homo sapiens…

  46. noncarborundum says

    #44:

    I’m not sure exactly to what extent people like Behe think that god intervenes in evolution…whether they think he comes down and rearranges genes or modulates the environment to tweak selection in particular direction.

    In Behe’s case, I’m pretty sure it’s the former. As I understand it, his contention is that the mutations necessary to form an “irreducibly complex” feature must happen in a coordinated way that is vanishingly unlikely given known mutation rates. Whether, once those features come into existence, the selection pressures applied to them are natural or supernatural doesn’t really make a difference to his argument.

  47. says

    It took too long to do at work, but one comment I have is that my scores changed over time. I became more critical later on, and tended to score lower. A “practice-run” might have prevented that.

  48. Isaah Vincent says

    i try to be an agent of change everyday. so i drive on the sidewalk on the way to work

  49. isaah Vincent says

    sooooooo close. hope this works. would love to take that trip. can i take PZ with me. or like a PZ-doll?

  50. Mindbleach says

    Everyone already is an agent of selection. Every time you kill a fly without a struggle, you help the fly population endure.