Bitbutter

Very nice… I think I'm already hooked, I clicked at least 50 times…
Very curious how it evolves.

i also digg it! i've already seen some amazing things pass by, like a pig dressed in spiky armor! ;)

digg it:
http://digg.com/tech_news/Mutating_Pictures_site

Addicting!
It combines the fun of Kittenwars with actual genuine human curiosity (what will derive from the experiment?). The face seems to be going towards a chinese opera-mask. I wonder if the face will become prettier…

Maybe you should split the experiment up in a male and female version. or you could teach the computer different (subtle) human expressions.

I miss a counter though (how many clicks). And I wonder if there is a limited amount of generations?

Anyway…very very nice. This seems to be the kind of thing that puts the potentials of internet into a new perspective.

xxx Michiel

P.S. Why did you put the pictures of screaming girls with chainsaws in the background?

Got load that where variations of the one 'cat' (the first one below) and then got these:

http://img.techpowerup.org/071002/Screenshot-1.png

Both look very cool. Good job Bitbutter

(Cross posted on Blogoscoped)

hello there. For me it is very difficult to choose because in each picture there is at least 10 faces. I'm having problems understanding where the face should be or more how big the face should be. It's not a fair comparison trying to choose between the small face with the moustache or the really big eyes with the small lips.
What is interesting is that it does make me wonder about what items make up a face? is it the mouth that is most important or is it the nose? Or most likely are you looking for some eyes?

Thanks all for checking it out and for the comments.

@micheil: there was some talk on Phillip's blog about why there are no female faces that show up so far. It might have to do with the type of drawing elements, at the moment they all have hard edges and high contrast. Perhaps more feminine features, would emerge if soft gradients were used (perhaps pale shapes against a dark background?).

About learning: The system knows just as little about faces as it did when it started!, but its great as shuffling shapes around and waiting for us to say whether a combination looks good or not ;)

The generations go on forever. I might add a click counter, but so far I've left it out to keep the reload speed between 'battles' as short as possible.

@alex: hi! yes, i know that feeling of it being hard to choose. If both faces look equally face-like I choose the one that appeals to me the most. I find it interesting to leave the question of exactly what to do in these situations open to the user. I trust that things will 'level themselves out' somehow.

What I do is generally select for faces that appear fairly centrally, and that are sized roughly like a head on a passport photo. If there are more faces that i can see, I select according to how life-like the most 'obvious' one is (that is, the one that's most obvious to me).

Very interesting project. Perhaps more fine and lighter (grey instead of black) lines and more delicate, rounder shapes would result in females faces. Also the males faces seem to be around 30ish years of age!!!

As both an artist and a person with a medical background, I choose both on aesthetics (the mix of dark/light shapes, facial proportions and features) and functionality (like could a person see through those eyes or breathe through that nose or talk through that mouth)

Are you suggesting that most of the voters are males who are 30-ish? ;)

There are 2 'strains' in the mix at the moment that don't look like older males. One looks like a cherub, and the other looks like a geisha.

I'd like to see a bit more detail about how long ago the faces I'm looking at started to diverge, if possible. Right now all of them have "447″ as the original ancestor, which is no longer informative.

Hi Matthias, yes i think all faces in the pool are descendants of just one of the 1000 original images now, 447.

I've been thinking about implementing an option to 'view common ancestor' for any pair of faces, but it would mean a fundamental change to how the database is used, and a fresh start with new images.

At the moment faces are deleted from the database when they get 'defeated', but to implement the common ancestor function I'd have to keep many thousands more records in the database at all times, which might slow things down.

Looks like a female alien face is coming through… the cat disappeared! fun.

Is the loser deleted immediately or does it get a second chance?

Interesting, a lot of pictures have two pairs of eyes (on top of each other).

Looks like a new breed of cat is emerging… A Kilrathi, to be exact. I posted a rendering on my site, not sure if I can post images here?

http://yfinder.de/random/kilrathi.jpg would be it.

@Markus, at the moment the loser isn't immediately deleted. Battles result in points being awarded to the winners and deducted from the losers. When a face reaches X number of points it reproduces. The lowest scoring images are deleted to make room for the new ones.

@Philipp: yes that's quite odd!

@wayfinder: Interesting, I tend to see the big eyes at the top as part of an elaborate head-dress on a human head.

There's a curious lack of noses in the current face pool - many have a small black animal-like blob that is joined to the mouth but very little resembling a human nose. It's an interesting trait because the noses on mutating pictures were quite prominent, and it's starting to bug me that a real nose hasn't been produced by face maker. Presumably that's down largely to chance, and as soon as one does come up it'll spread through the population. Or maybe it's a lower incidence of line-shapes in your algorithm that's doing it?

There are some more promising noses apearing, like this one: http://img105.imageshack.us/img105/5816/picture4iz7.png

I think you're right that good noses will quickly spread once they emerge, but it might be that other improvements take precedence (like eyebrows, irises etc).

As for the drawing shapes, there is only one elementary shape being used. It's a four cornered shape whose edges are curved. The lines we see are very flattened versions of this shape, it's true that these flattened versions are less likely to be created by chance.

Your current model uses a single test for faces, eliminates faces that do not evolve as quickly as the leaders, and seems destined to produce a handful of similar results. In nature and business, diversity appears when there are many niches to occupy. Now that you have rudimentary faces, have you considered setting up a second test page to encourage other traits? For example, a second page could ask "which face is most feminine?"

Beau: I have thought about splitting the pool and introducing new targets for the different sub groups, but I prefer the simplicity of the single instruction at the moment. Also I think there is still a long way for the face to go, I'm interested in seeing its continued development.

Yeah, I've been thinking about it some more. Your results are already showing that there is enough variation given the diversity in your voters. Nice work!

This is so cool… I wonder how far it will go towards a realistic face/head, or wether it will continue with the strange head dress that's currently apparent… sort of looks like samauri get up…

Some also look to have beards, while others the beard looks more like a shadow on the neck defining the jawline… very cool… wonder how far it will go by next month…

Since Saturday: One of two eye structural motifs became fixed, and the other extinct, or adapted to become eyebrows. A diversity explosion in noses occurred, out of which some very elegant solutions have developed. It seems the females have become extinct, or at least very very endangered. I wonder, once the assorted facial structures reach some level of sophistication, will we see a re-emergence of the various feminine facial markers, such as finer nose, and larger, more wide-set eyes? This is the question which will keep me returning to this web-experiment for some time to come.

There have been a couple of times when i thought it was very unlikely that the population would change sex and I was wrong both times. The lesson was that only a few subtle changes can be enough to switch from male to female or vice-versa so i wouldn't be surprised to see female faces returning at some point.

Noses along with the details of nostrils and lip to nose lines have started appearing since I started judging- very impressive. Female faces and male face are both present. The variations I see so far are: old thin male- sad, old thin male(white hair appearing)- content, young male samurai- nonplussed and happy, middle aged male samurai- nonplussed and angry, middle aged female geisha- nonplussed, middle aged female geisha- happy, young adult female geisha- nonplussed.

Out of curiosity: do generations 'breed' (ie: share/combine traits among high ranking images ala alphas of a given biological species having combined genetic sets create mutations more likely to be positive) or is it single parent mutations only (more like an amoeba or bacteria develope?)

-revscrj

revscrj: The reproduction is asexual here. It would be interesting to experiment with sexual reproduction too but for now i haven't been able to think of a good way to implement it.

An approach question:

Do you think that its better to choose a child with closer-to-face like features that has degenerative aspects to it, or a less-than-human-face with traits which seem to advance a feature? Mainly I am thinking now about the nose- having seen many instances of much better noses on less wholeisticly developed faces.

@ bitbutter
"…reproduction is asexual here… [sexual reproduction]… i haven't been able to think of a good way to implement it.

"sexual/asexual"- yes, those were the words I was fumbling for- lol. I have been giving the question some thought and the only possibilities I see are implimenting a pattern-analysis formula that would find similarities/differences between the 2 images in order to generate a sexually produced offspring… unfortunately I figure that would slow the generation of children down *quite* a lot.

@revscrj "Do you think that its better to choose a child with closer-to-face like features that has degenerative aspects to it, or a less-than-human-face with traits which seem to advance a feature?"

It's an interesting question. Assuming the aim is to get faces that are as realistic as possible as fast as possible I think that it's best to always choose for the face that looks the most realistic in it's totality (ie. skip the face with a perfect nose and mouth but no eyes). Part of the reason that I think this strategy makes sense is that most voters should also be following it, so the chance is smaller that you'll be voting against them.

Pattern analysis would slow down the generation of faces, though this wouldn't necessarily impact on the speed of loading faces (the face generation could be offloaded to a 'background' processes). At the moment pattern analysis is is something I know too little about, perhaps it's something for the future though.

A thought for a future version:

It seems that in nature there can be sudden leaps in evolution where a species rapidly transforms- far faster than the biological model usually allows for. To simulate this one could 'award' points to a winner based on the points of the loser so that if a combination of traits appear that are superior to others that trait-set would quickly rise to the top and generate more children. Also, a high ranked trait-set would gain few points (perhaps fractions of) by winning over newly generated children. This would reward compitition among higher ranked trait-sets.

Just a thought.

@bitbutter "I think that it's best to always choose for the face that looks the most realistic in it's totality (ie. skip the face with a perfect nose and mouth but no eyes)."

I agree. In nature, the -effective- deletion of an entire gene usually carries a greater Fitness penalty than a change in the gene coding region such that the gene product functions differently or is made in lesser or greater amounts.

Also, Natural Selection has no memory, and Evolution has no (specific) goal, but Artificial Selection does. At the risk of igniting an ID vs. Darwin debate, calculating Fitness based on over-all realism is closer to Natural Selection as it actually occurs, whereas selecting for the enhancement of one trait without considering others is closer to Artificial Selection, or Breeding (think dogs and goldfish).

It could be said that part of this web-experiment is the possibility that one or the other selection strategy may be more fit (0ooo! That's meta!).

@raccoon tail: I guess it's debatable whether artificial selection has a goal or not. I suppose what I have in mind is that artificial selection can share the same 'short sightedness' as natural selection (like the many examples of 'unintentional artificial selection'). Artificial selection need not have a long term goal.

I'm drawn to lack of memory that you mentioned. I find it fascinating that the repetition of simple, short-sighted, decisions can lead to very refined results. I designed Face Maker not to have an explicit goal, the only 'commandment' is to select the image that looks most like a face. A newcomer can start voting with only this single instruction.

Of course there is an expectation among the people taking part that the image will come to look more and more like a face (that's part of the promise in the 'about' text too!), but all this is secondary.

@ Bitbutter
". I designed Face Maker not to have an explicit goal, the only 'commandment' is to select the image that looks most like a face. A newcomer can start voting with only this single instruction"

In a sense, by making this commandment, you create the same structure as natural selection. Nature's single commandment is "Survive" and the implied purpose is to spread ones genes. Just as in nature, what 'survival' entails depends on the enviornment so here does "looking like a face" depend on what the person thinks of a face as looking like (as demonstrated by "which set of eyes do you see?" blog entry.)

@revscrj: yes, in effect I'm asking agents of artificial selection to act in a way that mimics natural selection :)

bitbutter: Check out my Evolve 4.0 software for an example of another alife project that mimics natural selection, as its only selection criteria is survival.

Facemaker is awsome. I vote every day, and now I am seeing the hair is finally fillinf out. I am very curious about the genetic code that draws the faces and the mutation algorithm.

Also, you might want to introduce sexual merging of top rated faces and see if that helps. The dual images is easiest for the user (unlike that other project that asks you to select a rating from 1 to 10). You might even be able to show as many as three to four images to select from, without unduly burdening the user. (but two has the advantage is utmost selection simplicity for the user).

It's truely amazing to see how the face emerges! The only limit is probably going to be the underlying drawing algorithm and the set of images contained in the biomorph space.

This project continues to fascinate. I suspect it may fascinate even more if one has been voting since the very beginning, to see and remember the evolution. Do you know if there has been any attempt to include this kind of project to educate people in general about evolution? It is one thing to abstractedly hold the notion of small changes being introduced through long periods of time, and it is another to actually vote for changes and see the effects. Now, not only animal/plant breeders and research scientists can get a 'realer' sense of evolution. Ah, I love the smell of disruptive education in the morn.

As for the evolving faces, the nose is now a functional and aesthetic pleasing one (thank goodness that painfully pinched, hardly functional one a la Michael Jackson is gone), a head of hair is appearing (albeit with a rakish hat), and intriguingly at times a court jester's hat replete with bells. Eyebrows (which had developed from a second pair of eyes) seemed to have evolved out of existence. A chin is becoming refined a bit more.

I find it amazing how quickly certain detail aspects have become consistant after appearance, say for instance 'the shadowed beneath the nose' or the 'nose to lip lines' both of which only began appearing about 2 or 3 days, at most, ago but are now pretty solidly established in almost every child.

@ logicel: yes it has been amazing and entertaining to watch the transformations occur! When I first started voting there was more chaos than structure to the generations, now to see the inverse as true and have faces of varied emotions and character taking form is really bizarre and sometimes breath taking.

I wonder what the end resulting faces represent to the judging populace? Some Jungian Icon form? An expression of common athstetic regarding the pure idea of Human, as perceived by Human? Hmmm…

very neat. i was trying to 'breed' for a fat face or other weird traits but could not.

@ken Evolve 4.0 looks interesting. Unfortunately i don't have a windows machine to run it on.

@Logicel: I don't know of an example of this kind of demonstration being used to educate people about evolution in a more formal setting (school etc). This is something I've had in mind while developing Face Maker; i think that this kind of 'toy' is a great way of presenting the fundamentals of evolution in a direct and engaging way. If you have any ideas/suggestions in this direction please let me know.

@revsrj: "I wonder what the end resulting faces represent to the judging populace? Some Jungian Icon form? An expression of common athstetic regarding the pure idea of Human, as perceived by Human? Hmmm…" I've wondered about this too.

A related thought: It would be interesting to arrange a similar experiment with animals as 'voters'. For instance chicks would be presented with two images and their response to each would be logged somehow to determine which the 'winner' was (perhaps the length of time the image was looked at could be measured).

We already know a lot about the kinds of symbols that animals react to (eg. silhouettes of birds of prey) but I wonder if an 'evolving image' approach could succeed in 'revealing' the most important images that are encoded in the animal brains with more detail/clarity than traditional test card experiments have been able to.

@ Bitterbutter:

Thats an interesting concept… I could see a series created by testing various mentally ill people(chronic and acute) with 'guidepost questions' (questions that give the general lay-of-the-psychological-land of the person) first on a larger level- either regarding inner sweeping definitions "More like happiness/pain/shame/contentment/love?" or inner-view on external self "more like your future/childhood/daily routine/your effect on the world?"- then on a smaller "More like your friends/family/mood/society/body?" etc. Top scoring images would be kept to form a deck.
Provided that, for example, in test results boderline Schitzophrenics tend to not only actualize similar final decks but also are definably different than other test patient catagories final decks. If that ended up being the case (I suspect it would to a fuzzy-bordered degree) then they could be used to ID said illnesses by presenting them in sequences to a patient and asking 'which was more preferable?'
A potential snag is 'Does a borderline schitzophrenic end up choosing the same type of image that they would create?'

Bitbutter: I am in process of researching online education sites that would be interested–no dice yet. Unfortunately, one site that I thought would be interested has lapsed.

My husband (a Paleoinformacien, an occupational word coined by him to denote his expertise in old software) remembered the work of Baricelli who apparently was the first to toy with numerical evolution back in the Fifties. We were able to unearth a nice bunch of info about Baricelli's work on the net.

My husband's critique of your project's possible use as an educational tool was based on the high probability of ID proponents decrying it as not showing evolution, after all, the the voters are obviously intelligent designers. He then mused about creating a mandelbrot universe (resembling natural landscapes) in which evolution of creatures can occur without any intelligent design input. If you want to pursue this further with him, just contact me via my email.

I JUST SPIED THE TINY BEGINNINGS OF EARS–SYMMETRICAL ONES TOO. WOW!

Hi logicel,

Rob Brown wrote an interesting article here that reflects the way I was thinking about the potential of Face Maker-like toys as educational aids. http://karmatics.com/docs/evolution-and-wisdom-of-crowds.html

I think that any ID-er who might complain that such sites aren't a demonstration of evolution would be wrong. This is evolution driven by artificial (intelligent) selection, but evolution none the less. But I'm sure your husband is right that this kind of complaint is something we would expect to hear.

I'll look up Baricelli. Earlier I was looking into ways that I might implement sexual selection, I'm sure that would be much more interesting (it's sad to see promising new features emerge and then disappear again because they didn't happen to be part of the 'winning lineage'). But so far the material on 'least squares analysis' which I think I'd need to use, is a bit heavy going :)

I'd love to see an animation of the evolution thus far!

It looks like the traits/elements originally selected FOR are now selected AGAINST . . . a trait originally useful as for one reason now useful for another, while some other trait has taken its place. This correlates in interesting ways with our understanding of evolution, and is an argument against intentionally selecting for certain traits.

I WAS wondering: "What's with the head-dress?" Until I saw earlier generations in which the 'face' was actually up in there!

I think this whole idea is FASCINATING, and it gives me a much more 'visceral' feel for the evolutionary process!!

Isn't there a specific part of our brain reserved for facial recognition? We must be somehow forming an aggregate approximation, like a map(?), of that potion of the human mind!

I'd bet that there are also applications here for camouflage-mask making?

Anyway, this is fascinating on several levels! Thanks!

There's something I don't quite understand . . . it probably has to do with the mathematics of the algorithm you're using - which would be over my head. Virtually all of the current images have certain features in common . . . which don't seem in any way useful! One is in both upper corners: a set of curved, dark, large shapes with a kind of lightening-bolt white streak through them (I'm looking at gen 176-ish). Another is a pair of inverted triangles which to me seem to delineate the sternal border and the collar-bone. No matter how I try to make these part of a face, I can't seem to. They seem completely irrelevant to any face I can make out. So why are they so persistent and universal? Any idea?

Hi PirxPilot, I'm glad you're enjoying it!

I think that those black 'antlers' with the lightening bolt through them do look more like a real helmet/head dress than the 'stuff' around the heads of previous generations.

In general the genes responsible for expressing the head dresses are free to vary quite a lot without affecting the 'face-like' ness of the images. It could be that the basic shape of the antlers you see happened to coincide with an especially realistic combination of facial features in a previous generation, and quickly became widespread as a result of the success of the facial parts of that line of images.

I notice that just in the course of today, the "antlers" have evolved into more of a headdress. This morning, the images all looked like the demon from Legend, now they look more like African or Mongoloid warriors. The rate of change is just staggering. Ears were barely starting this morning as nubs, and now I'm seeing several images with full ear lobes!

It's interesting to me that the nose has been slow to narrow, even those it's cartoonishly wide, still.

Man I wish we could see the evolution in time lapse! Even just having the "top faces" up and not touching it for a few hours shows big differences.

Bitbutter, how many battles have there been so far in total?

Unfortunately the system is designed in such a way that there's no way for me to know how many battles have taken place! I can make an estimate though:

Since the site was launched on October 2nd 2007, it has had 26,398 visitors. The average time spent on the site is 01:56. Most of the visitors use a fast network connection so a new pair of faces will take about two seconds to load after voting. If you leave time for reading the text on the site and pausing to decide which images look more face like i think the average user might vote twenty times per visit.

20*26,398=527,960 votes. My guess is somewhere in that region.

There's no way to tell from the number of generations? If there's a stack of 1000 images, and they're on generation 199, wouldn't that necessitate somewhere around 200,000 losers, and therefore 200K battles?

An image only reproduces when it gets to X number of points (currently set to three), but while a win adds a point, a lose takes a point away so there's no way to know how many battles each face had before it reproduced.

Also, looking at the current pool of images and their generation numbers doesn't take into account all the battles amongst the now 'extinct' branches that are no longer represented in the population.

So whatever the real number of votes/battles is, we know that it's much higher than 200k.

Having played with this ALL DAY, I have to confess, it's hard to get out of the habit of defining a face by the nose! :)

wow! that's a remarkably detailed face that has come out of this. do you think more development from the current face is possible? by now i can hardly say one image looks more like a face than the other.

can we expect new starting pictures?

it would be nice to see more of the evolution of the face. are all the images preserved or is the material online all there is?

interesting project!

Hmmm, facial development seems to have stagnated for the last couple of days.

Are you familiar already with the Morphs project? If not, here it is:

http://www.cs.bgu.ac.il/~sipper/biomorphs/evolution.html

Seems that it is sexual reproduction.

@Logicel: Development has slowed because the number of visitors dropped dramatically over the weekend and has been low since.

Incidentally, if anyone reading this wants to help get development moving more quickly again it helps enormously when people 'bookmark', or 'vote up' the site in social bookmarking website like stumbleupon (signing up takes a couple of minutes). About 90% of the site's visitors so far have arrived there from using the stumbleupon toolbar.

Thanks for the morphs link. It looks like a variant of Richard Dawkins' biomorphs program. I think the task of implementing sexual mixing in a system like morphs is quite straight forward because the system already represents alleles. In Face Maker there are no chromosomal slots, instead the face genome is just one big unordered string of numbers, each of which is free to mutate. The problem that I get stuck with is that there's no (simple) way of taking a shape (or group of shapes) from one face and determining which shape (or group of shapes) is its equivalent in another face.

@joor: hi! I'm not sure about adding new starting pictures, i think i probably won't in the near future. On the one hand it would be interesting to see how a second evolution develops (and to compare it to the first). But on the other hand I'd like to devote all the 'clicks' that the site attracts to developing the first population, to see how far the current population of faces progresses.

And yes I do think further development is possible. I think that progress can continue (provided the site keeps being visited) until the best images are indistinguishable from a good drawing of a face. At this point the perfect symmetry of the images will stand out as the most 'unrealistic' aspect of the face.

Unfortunately it would have cost a large amount of data storage to keep all the the ancestor faces so only the original images and the current population are stored at any time, this meas its not possible to view the progression unless you've been taking screenshots. I'm thinking about ways i could change this for if I make a future version.

bitbutter, thanks for the feedback, I suspected that site traffic was dwindling.

Hey Bitbutter,

Love the site, have spent waaay too much time on it over the last week. I do hope the traffic picks up — maybe a link on the Wikipedia page for artificial selection would help? I think that having a second face going simulataneously might actually increase the rate of development of the first — it's quite interesting to see actively evolving faces at widely different stages of development and that might get visitors clicking on both. Plus that way the site remains valuable even if traffic does stay low.

As far as sexual reproduction goes, I don't know what algorithm you're using now, but if each face is defined by a long string of numbers, could you just randomly take half the numbers from each parent face and let the selection sort it out from there? Seems like that way you'd preserve half of each face's patterns (plus any mutations) without the need to map one nose to another or anything… May not work, but figured it was worth a shot!

alright, i understand your arguments about not starting a new series, but agree more with fang2415 that it c'd also generate more interest and even help to develop the current evolution, while giving new insight to what kind of evolutions are possible within the program.

regardless; it might be interesting to do a little item about the project at ketel tv. i can't find your contact info right now, so it might help if you e-mail me when interested.

@fang & joor: maybe you're right about adding a new population, I might do that soon.

@fang: great idea about adding a link to the wikipedia article, I've done that.

"could you just randomly take half the numbers from each parent face and let the selection sort it out from there?"

Perhaps it's worth a try. If i get chance I might run some local tests to see how this might work.

You'd probably get greater evolution if you had your artificial selectors target some specific goals. For example, narrowing the nose, getting a bridge to form, reducing the "collar" effect, eliminating some of the external noise elements from around the sides.

I'd been trying to do all these with my selections over the past few days, but clearly my winners are being overridden by people driving other criteria.

@Nathan: If specific directives were to be mentioned ("narrow the nose", etc) in response to the current population of images then the project would have moved more towards being an exercise in trying to get people to create the face I'm thinking of, which is less interesting.

Part of its attraction for me is that these faces currently develop through a very democratic process. In this process people's votes will sometimes work against each other, but that's also an important strength of the the system works.

Differences in people's voting criteria will work against each other and get canceled out while the agreements in voting will drive the direction of the face development. So there is currently an aggregation of individual voter 'wisdom' about the question 'what makes an image look like a face'. I have a strong idea that this wisdom would be diluted if more specific directives were given.

bitbutter, while I'm a fan of the spontaneity aspect, I'm just thinking of how this relates to artificial selection in the real world. Crop-breeders, for example, have a fairly narrow set of goals in mind when they start their breeding cycles.

At this point, from what I can see, everything looks like a face. So "which looks more like a face" is not a sufficient guiding criteria for selection in my opinion. Even something subtle like "which ENTIRE IMAGE looks more like a face" would start pushing to eliminate artifacts.

Evolving the selection criteria seems every bit as much part of the artificial selection process. To me, at least.

Hi Nathan,

"At this point, from what I can see, everything looks like a face. So "which looks more like a face" is not a sufficient guiding criteria for selection in my opinion."

I agree. Yesterday I adjusted the question so that it now reads: "Which one looks more like a (realistic, human) face". What I hope that this conveys is that the first priority should be to select the most face-like image, and if both images are equally face-like, select the one that looks most like a human face.

In the event that both images look equally face-like and equally human and realistic, it doesn't matter which image is chosen. I suspect that at this point people will tend to choose towards characteristics that they prefer to see in a face.

how big was the starting pool? it seems to have been rather small.
five to ten images maybe?

"I suspect that at this point people will tend to choose towards characteristics that they prefer to see in a face."

true, i'm currently trying to 'create' a female face without a mustache…

@someone: the starting pool was 1038 images.

meh, it ate the rest of the post…
the 'new' faces on the progress page and all the faces i have seen so far seem to come from one seed, so i thought there were few seeds to start with.
if really only one strain survived, i'd like to see a new pool where one strain cant eliminate all the others.

sorry for triple posting… can't edit.
seed #447 has eliminated all others.

Because reproduction happens asexually it's inevitable that after a while all faces in the pool will have the same ancestor (in this case image 447), because one 'branch' will emerge as the most face like one.

@bitbutter, someone

I think there are more factors involved than just the effects of asexual reproduction. Since I have been judging from so early into the experiment I have seen many trait-sets vanish AFTER the only survivors were 447 descendants. It seems that not only has one common ancestor beat out all others but one 'localized strain' has won out over the others. In human terms: we all descend from the same hominid who was competing w/ several in its day, and due to isolated group evolution grew into the various races of today but the current model is as if eventually everyone exhibited the Lithuanian physical traits worldwide.

Any thoughts on why?
(yes I know the 'human terms' is a flawed parallel due to sexual reproduction, but it was a simplification for demonstrative purposes)

-revscrj

@revscrj: "It seems that not only has one common ancestor beat out all others but one 'localized strain' has won out over the others."

That's exactly right. The reason we don't see independent strains emerging and persisting is that groups of faces are never isolated from each other (the way that geographic isolation of a group of animals can give rise to a new species). Any face always has a chance of 'battling' with any other in the population, all faces are in direct competition with one another. So when a strong trend emerges it will spread at the expense of other less powerful trends.

An animated version showing one branch from beginning to end would be amazing.

@bitbutter: regarding kc's idea, if you had a set of images that fit the bill I would be happy to make the animation- call it a 'thanks' for putting up such an interesting experiment. You have my email addy if you are interested.

@kc revscrj: Thanks for your offer rev. I agree that something like this would certainly be interesting to see. Unfortunately only the original images and the current population are stored in the database to avoid running into storage space issues.

If I make a future version of Face Maker i'll look again at possible ways of incorporating a feature like this.

Looking good! I can't wait for it to look like me.

Anyone else noticing there are a few tuxedos evolving into the mix. Your face subject is getting a little more sophisticated, hehe…

Two new face populations have been added. You can expect to see pairs of abstract explosions again amongst the more developed faces.

Won't the new ones just die out immediately since there are more adapted species already in existence?

Actually cancel that. It's obvious that the populations are living in different places!

Yes exactly. Members from different populations aren't in competition with each other.

Can you make an option to only select faces from the new population? It's a bit of a waste of time selecting from the old population. They're barely changing.

The old faces are changing at the same rate as the new ones. There is more variety in the new populations because no face has yet had chance to dominate them. If and when traffic to the site picks up again in the future I might add an option to select which population you want to vote on.

They're still barely changing. If you can select from only the new population they'll evolve twice as quickly.

"They're still barely changing. If you can select from only the new population they'll evolve twice as quickly."

One aim of the experiment was to see how closely the images will come to resemble faces, so the older populations will stay, alongside the newer ones.

The problem is that no "new" face is going to win a contest against any of the "old" faces for a very long time.

Why not pit *only* the new faces against each other until they look as realistic as the old ones, and then enter them in direct competition with the old ones and see how they interact?

In the meantime, you're wasting three quarters of everyone's clicks. At least give us the option to choose to exclude the old ones! Please.