Google Blogoscoped

Forum

Mutating Pictures  (View post)

Artem [PersonRank 4]

Monday, October 1, 2007
16 years ago38,262 views

It is somewhat difficult to rate until you see several creatures and get some kind of a scale. Would it make sense to:

1. List some general guidelines. E.g. 1-2 – have to be killed, not even close; 5-6 – might be a face drawn by idiot; 8-9 – I might have thought it's a face if I only had a 0.5 sec look at it

2. Or since it is about determining which creature is better, than another one, would it make sense to *compare* two creatures in the hot-or-not style?

Philipp Lenssen [PersonRank 10]

16 years ago #

I like the second idea Artem. It's a good clean concept for similar sites.

As for item 1, I should add that a picture will not get automatically killed even with 0 rating. It's more like an average of many user submissions. Because a picture will just "age" without creating offspring, and the older a picture the higher a chance it will get "accidentally killed" by another, higher-rated picture. BTW, just now I lowered the rating-to-offspring ratio a little, as too many "deaths" were taking place in the first minutes of the site...

mak [PersonRank 5]

16 years ago #

Amazing!
I'm seeing some organically generated creatures that resembles what we get in a genetics experiment.

I suggest also doing the same experiment with lines (not shapes) and try to get something that looks like an insect – a spider for example.

mak [PersonRank 5]

16 years ago #

[off topic]
This is the first time I see you adding a "digg" link! Is this a new trend ;)

Simon Hammond [PersonRank 0]

16 years ago #

If you are interested in evolving faces then you should definitely try http://picbreeder.org which has a good representation for this and also has advanced breeding features.

Philipp Lenssen [PersonRank 10]

16 years ago #

Mak, I did this button a couple of times before, but don't do it too often... in this case, the site needs a lot of helpers for ratings so that it will work...
By the way, good suggestios with lines & spiders. If there are enough visitors in the future to make "evolution" of some sort happen, I can start new picture pools, each with a distinct drawing algo (e.g. non-symmetric + lines for "animal").

Ludwik Trammer [PersonRank 10]

16 years ago #

This one is really good:

http://www.ludwik.trammer.pl/screens/mutating1.png

It even has a hat, a mustache and a tie.

Ludwik Trammer [PersonRank 10]

16 years ago #

and of course beard and vampire teeth...

Philipp Lenssen [PersonRank 10]

16 years ago #

Interesting, by now every single member of the original population already "died", replaced by new generations. The oldest living member right now is 58 minutes old (not counting the daily backups, which save every day's generation for later).

Joe Torben [PersonRank 0]

16 years ago #

I saw something similar with a world map. Does anyone have a link to that?

Jon Henshaw [PersonRank 4]

16 years ago #

Damn you and your clever distractions! ;-)

Mark M [PersonRank 1]

16 years ago #

I've seen two insanely face-like images in about 10 minutes. Not even, 'that could be a face' but 'it's impossible for this to not be a semi-pro artist's work'

Florian Mayer [PersonRank 0]

16 years ago #

will creationists sabotage?

Zim [PersonRank 10]

16 years ago #

Interesting research... There are some good images!

bitbutter [PersonRank 1]

16 years ago #

Excellent project! It's a long overdue web-variant of Richard Dawkins' Blind Watchmaker programme.

I'd also like to add my support to the idea of a 'pick the one that looks most like a face' style of rating system.

bitbutter [PersonRank 1]

16 years ago #

Also: What kind of backend are you using for this application? specifically the image generation part?

Tim Broder [PersonRank 1]

16 years ago #

I'll have to try this at home. Work is blocking it as "games"

Philipp Lenssen [PersonRank 10]

16 years ago #

> Also: What kind of backend are you using for
> this application? specifically the image generation part?

Bitbutter, if you arrived in this thread directly, check out the accompanying post – I'm using ExplorerCanvas/ Canvas to generate the images:
http://blogoscoped.com/archive/2007-10-01-n82.html

It would also be possible to generate them as PNGs using PHP GD on the server, but I figured that might bring down my server (too much CPU cycles and storage)...

bitbutter [PersonRank 1]

16 years ago #

Ah thanks: fascinating, i'll look into ExplorerCanvas/Canvas, haven't come across them before.

Mrrix32 [PersonRank 10]

16 years ago #

I had one that looked like a Devil if you cross your eyes, but guess that doesn't count :D

alek [PersonRank 10]

16 years ago #

Interesting stuff and classic Blogoscoped cleverness – DUGG for sure.

You mention some generational stats and other misc. above ... those are kiinda interesting – recommend you put 'em on the website. I.e. how many you started with, how many generations have gone, how many folks have visited, etc.

Philipp Lenssen [PersonRank 10]

16 years ago #

I dropped in a Google Analytics counter but due to not wanting to use a database, which always creates a bottleneck on my server setup, I don't track too many other stuff unfortunately... so right now all I do is some manual tracking... manual meaning I watch the existing 1000 population from time to time to see its oldest member, its new members this minute, and so on... I do make a daily backup of the population though and also a couple of snapshots throughout the day today...

Amagi Tremper [PersonRank 3]

16 years ago #

You seem to have a lot of statistics about the images itself, you'll make something more than that we have on the process page *pls* =)

alek [PersonRank 10]

16 years ago #

From the DIGG comments (it's on front page now), it sounds like they clobbered you at first ... but server seems to be hanging in there now – nice job Philipp.

Philipp Lenssen [PersonRank 10]

16 years ago #

Alek, I increased the delay in-between two rounds and also deactivated the progress/ history page (pssst, it's temporarily here now, shielded from view for most http://mutatingpictures.com/progr-internal-dont-link-here-please.html :)). Plus, I lowered the rating-to-mutations ratio again, it doesn't seem to matter tho since even with a lowered ratio right now every generation lasts only about 3 minutes (meaning every pictures gets randomly replaced once by a new-born at least every 3 minutes...)...

alek [PersonRank 10]

16 years ago #

So per the DIGG comments, have the "boobies" started showing up as people try to subvert the system?!? ;-)

Tony Ruscoe [PersonRank 10]

16 years ago #

This is fascinating, as always with these experiments Philipp dreams up! I'd like to see a "family tree" representation of the data, showing what spawned what, although I bet that would be pretty big!

Philipp Lenssen [PersonRank 10]

16 years ago #

Tony you are right, I wasn't able so far to figure out a resource-friendly family tree/ history feature, except for the snapshot of the population I'm making (at least once a day). But these snapshots don't show the mutation from one picture to the next, which would be a huge family line, and considering all the mutations and versions, and also dead ends I'd have to store somewhere, would become too big for the server....

Ozh [PersonRank 1]

16 years ago #

Philipp : yet another mind blowing stuff. You're definitely the most amazing blogger I've discovered this past 5 years.

Jason [PersonRank 0]

16 years ago #

The faces just look evil to me... how about some curves instead of just triangles. Nice concept though, althought I lost interest after two clicks just because of the monochromatic/artificial look of the faces.

Luka [PersonRank 10]

16 years ago #

1000th digger! Very good idea these mutating pictures!!

Matt [PersonRank 0]

16 years ago #

Really cool... so far I've seen several very face-like images... two like Jesus, two like demons, one like an ogre and another like a samurai. Interesting how very specific the "faces" end up looking.

Mr Funk [PersonRank 0]

16 years ago #

I'd so "which of these X looks more like a face"

I think it'd speed the process.

This is amazing [PersonRank 0]

16 years ago #

I'd love to learn where the concept for this came from and how it is coded.

I'm just amazed when distributed input can automatically create such beautiful results. (that drawing circle thing is another example)

André Sampaio [PersonRank 0]

16 years ago #

I first saw this page about 4 hours ago and I took a look right now on the progress page and was amused by how fast the faces are being evolving.

Jim Barnes [PersonRank 1]

16 years ago #

Why do they all look like evil warlords?

Karl [PersonRank 0]

16 years ago #

They're starting to look like villainous western cowboys, with full beards.

Jim Barnes [PersonRank 1]

16 years ago #

Maybe the full beard is an obscuring factor that's more likely to happen than the complex configuration that makes lips, chin, etc. So full beards are more likely to look human.

Augusto [PersonRank 0]

16 years ago #

I am taking right now an anthropology course....
Love the idea!

Ron B [PersonRank 1]

16 years ago #

I agree that it's really rather amazing how they all seem to be evolving towards the look of evil dudes with facial hair. Something to do with the large black areas?

And are we seeing any sort of evolution towards smaller shapes or even 'lines'?

Jim Barnes [PersonRank 1]

16 years ago #

They are getting little reflective slivers on their eyeballs. This is really crazy seeing what's happening, they are getting more realistic hour by hour. Wonder how far it will go.

Colleen Sullivan [PersonRank 1]

16 years ago #

You should add a feature that allows us to select areas of the picture that look the most human. It would greatly speed up the process if we could select the part that really looks like a nose or a hat, and that would be taken into account when the picture mutated.

John Clevenger [PersonRank 0]

16 years ago #

This is perhaps the coolest thing I've seen in awhile. Keep it coming.

Taylor [PersonRank 0]

16 years ago #

At this point most of the pictures look either like Japanese warloards or Jesus. I wonder if this has some odd underlying psychological explaination.
Hrm...

Ben [PersonRank 0]

16 years ago #

Lol i just saw Voltron

Colleen Sullivan [PersonRank 1]

16 years ago #

Just saw a very convincing monkey.

ciper [PersonRank 0]

16 years ago #

It would be far easier to compare two faces and choose which is better looking. Sometimes its hard to judge one by itself.

bob dole [PersonRank 0]

16 years ago #

the level of similarity seems to have increased insanely quickly, this morning they were blobs, and now i'm home from work they're mostly looking like faces.

How are the offspring generated? what level of varience is added each time?

MrWizard [PersonRank 0]

16 years ago #

It looks like the final result is going to look like a cross between a Transformer and Darth Vader... :-)

Slushpuppie [PersonRank 0]

16 years ago #

Yeah, clearly all japanese warlords, or possibly mogol hordes

cipher [PersonRank 1]

16 years ago #

I agree with ciper, this would work better if two pictures were shown side-by-side, and the user could pick the one that looks more like a face.

Abbie [PersonRank 0]

16 years ago #

I'm amazed at the progress made in such little time. Already almost all the images look somewhat like faces. I'm excited to see how they get refined.

Philipp Lenssen [PersonRank 10]

16 years ago #

> How are the offspring generated? what level of
> varience is added each time?

I started out with rating = number of offspring (e.g. a 6 would generate 6 "kids"). Each mutation in the beginning was 15% of the previous polygons replaced by new polygons with random positions. But then more and more people played the game and I lowered both the rating-to-offspring ratio as well as the mutation threshold, this also allowed the server to survive through this.

I have some ideas how to automate this mutation threshold and offspring ratio adjustment, e.g. the script can look at the average number of ratings and then adjust accordingly, or the higher the rating the lower the mutation amount per offspring, or a 1-by-1 comparison for each picture as many of you suggested. However... the only thing I need to figure out is how to implement this so it doesn't bring down the server, ideally without any database connection etc.

Here are some of the neat pictures from the current pool – indeed, a huge tendency towards Samurai warriors right now! I wonder if the system reached a balance on this picture now or if it's just one of many "face" phases.

http://blogoscoped.com/files/mutatingpictures/1.png
http://blogoscoped.com/files/mutatingpictures/2.png
http://blogoscoped.com/files/mutatingpictures/3.png
http://blogoscoped.com/files/mutatingpictures/4.png
http://blogoscoped.com/files/mutatingpictures/5.png
http://blogoscoped.com/files/mutatingpictures/6.png
http://blogoscoped.com/files/mutatingpictures/7.png
http://blogoscoped.com/files/mutatingpictures/8.png
http://blogoscoped.com/files/mutatingpictures/9.png
http://blogoscoped.com/files/mutatingpictures/10.png
http://blogoscoped.com/files/mutatingpictures/11.png
http://blogoscoped.com/files/mutatingpictures/12.png
http://blogoscoped.com/files/mutatingpictures/13.png
http://blogoscoped.com/files/mutatingpictures/14.png

By the way, as far as statistics go, 71,407 absolute unique visitors yesterday (53,831 visits from Digg.com)! The bulk of the visitors was from US, Canada, UK, and Australia, then European countries following, not so much from Asian countries yet.

DaveX [PersonRank 0]

16 years ago #

Everything I've seen so far has definitely been in the Japanese warlord "look"... but the progress from my first visits until now is amazing. I'm going to make this a regular part of my day, it's very interesting to follow along with.

Alex p [PersonRank 1]

16 years ago #

yup, japanese warlords, jesus and the devil

Anand [PersonRank 1]

16 years ago #

really interesting piece of experiment is going on here and the progress is very visible, congratz.i would keep contributing here to see whats the improvement.

cheers

   <(o)> | | <(o)>
/ /
   (_ _)
   /
   | ________ |
   --------
  
   -^-

Alex p [PersonRank 1]

16 years ago #

I belive that the factor for them lokking japanese is the wide hair ate the bottom and the eyes. This is becuase there triangular in shape and it will be a LONG time before circles stard to come thru, seeing a couple of rectangles tho!

bob smith [PersonRank 1]

16 years ago #

Where's the explanation of the mutating algorithm?

Chris [PersonRank 0]

16 years ago #

I made a list of my favourites so far. Some have round eyes. Definetly not limitied to east-asian looking faces.

http://thisidea.amuses.me.uk/faces/Blank.html

Onym [PersonRank 1]

16 years ago #

Wow, beginning to see a lot more fine detail / thin lines.

Philipp Lenssen [PersonRank 10]

16 years ago #

"Ape Planet Ambassador" is cool :)

June Bug [PersonRank 1]

16 years ago #

They almost all look like faces now. Perhaps we should start selecting for realistic vs. cartoon traits or something along those lines?

Chris [PersonRank 0]

16 years ago #

I was trying to select for round eyes. And then for ears. But it didn't seem to have much effect.

Onym [PersonRank 1]

16 years ago #

There are 4 or 5 very distinct faces emerging.

Onym [PersonRank 1]

16 years ago #

These pictures aren't being mated with one another? They're just having mutations imposed on them?

Ron B [PersonRank 1]

16 years ago #

So in the 'favorites' images you posted above, the primary triangles that make up the eyes are all in exactly the same place. Are these triangles hard-wired into the algorithm? Or do they all share a common ancestor? (ghengis khan :-). _Can_ you trace the ancestry of any image, incidentally?

Mango [PersonRank 1]

16 years ago #

Ancestry tracing would be a really good feature.

I think the samurai look is because of the use of triangles. Asian eyes are more easily approximated by triangles, and also the helmet. It would take many more generations to get a round head of hair.

What interests me most is that most of the images have those giant lambchop sideburns (which contribute again to the samurai appearance). Since they don't add anything to the face-like appearance, they must be more like a genetic artifact. Did they happen to appear in the same image as the semi-realistic eyes, and came along for the ride?

Kenny [PersonRank 0]

16 years ago #

I've started voting "samurai" pics as 0-4, according to how good the resolve as faces, and "non-samurai" pics as 5-10. I hate that one black triangle!

Mango [PersonRank 1]

16 years ago #

Onym wrote: "These pictures aren't being mated with one another? They're just having mutations imposed on them?"

Wouldn't it be great to have two image populations using different mutation algorithms? Sexual vs asexual reproduction. Don't let the participants know which is which. See how fast each population converges toward realistic-looking faces.

bitbutter [PersonRank 1]

16 years ago #

Fascinating to watch how this unfolds. The difference between today and yesterday is amazing.

[put at-character here]Ron B i can't speak for phillip but i doubt very much that any traits are hard wired into the algo. The eyes stay in roughly the same place relative to the rest of the face because we know that that's where eyes belong, and we keep on increasing the number of 'eyes-in-the-right-place' children in the database.

[put at-character here]Onym: sexual reproduction would be very difficult to implement in a sensible way (i think you'd have to work out some kind of 'face embrology' in order for the combinations to make sense). my guess is that asexual reproduction is happening here.

Mango [PersonRank 1]

16 years ago #

Kenny I don't think that's necessarily a good strategy. A lot of those samurai pictures are really face-like, aside from the big triangle on the side and the funny hat. You might be throwing out the baby along with the bathwater.

Try giving the undesirable triangle a -2 penalty from how you would rate the picture without it. If it provides a reproductive penalty, the algorithm should eventually get rid of it.

bitbutter [PersonRank 1]

16 years ago #

"What interests me most is that most of the images have those giant lambchop sideburns (which contribute again to the samurai appearance). Since they don't add anything to the face-like appearance, they must be more like a genetic artifact."
[put at-character here]mango The big triangles that make up the 'helmet' do an important job of framing the face area, so i guess there was pressure for them to remain once they got established.

Mango [PersonRank 1]

16 years ago #

bitbutter: I don't think sexual reproduction needs to be so complicated. You take two 'successful' faces from a given generation, take 45% of the triangles from one, 45% from the other, then add a 10% mutation.

Kenny [PersonRank 0]

16 years ago #

Mango: Sterility is the ultimate reproductive penalty.

John [PersonRank 0]

16 years ago #

Why do so many faces have those large triangles by the cheeks?!?!?
What are you people thinking? LOL

I am trying to breed those out of the pool by not giving any triangle-cheeked face more than a five. Any help is appreciated

Philipp Lenssen [PersonRank 10]

16 years ago #

> Are these triangles hard-wired into the algorithm?

None of the shapes is hardwired, they can (theoretically) all freely move. Practically, due to user selections I saw certain "fashions". Yesterday evening there was an incredible amount of very weird devil horns, and "side burns" but much much higher on the canvas than the ones we're seeing today. And now, this "fashion" disappeared since yesterday.

I don't know what causes these, but as has been mentioned before, I think some of these polygons do a job of e.g. removing ambuigity (e.g. the ambuigity of some people looking for a face in the center, and others in the whole canvas – I think that was a big "issue" for picture survival strategies yesterday), so they become "survival" genes. And then perhaps some of these "surviving" strategies that remove ambuigity then happen to look like, I don't know, a Samurai helmet, and once they do there may be a survival "strategy" that favors other parts of the face to look like a Samurai!

And yes, they very often do share the same ancestor – if you overlay two or three pics and a triangle in both is in the exact same position, there is a good chance it's the same genes somewhere.

> These pictures aren't being mated with one another? They're just
> having mutations imposed on them?

Yup, no cross-mutation. Basically, a single picture mutates completely randomly by some percentage (like say 10%).

One thing I might try to do: add another face pool, like a second island, and then watch if this second island also evolves towards Samurais, or if it's completely different. I'm currently working on bringing a new drawing algo for a new "animal" pool onto the site.

Philipp Lenssen [PersonRank 10]

16 years ago #

> I am trying to breed those out of the pool by
> not giving any triangle-cheeked face more than a five. Any
> help is appreciated

This behavior is very interesting – maybe it explains why we saw a fashion come and go yesterday! Perhaps even the best survival strategy for a certain time will cease to be "fit" soon because the rating community gets annoyed by repetitive pictures. This is an interesting spin to it – I wonder if this means there will be constantly new fashions emerging?

I will have to add more options to the progress viewer so you can see these fashions, if they persist...

Dan Tobias [PersonRank 6]

16 years ago #

They kind of all look like cartoon villains to me.

Kenny [PersonRank 0]

16 years ago #

It would be interesting to see an animation following the mutation of a successful face.

bitbutter [PersonRank 1]

16 years ago #

[put at-character here]mango:

"I don't think sexual reproduction needs to be so complicated. You take two 'successful' faces from a given generation, take 45% of the triangles from one, 45% from the other, then add a 10% mutation."

Perhaps some interesting things could be thrown up by sexual selection, but i think that in general the system would have more trouble converging on realistic faces than it currently does.

An easy way to picture the problem is to image you have two very life-like faces. They have children via sexual reproduction like you suggested. ll the children end up less lifelike because they each have something weird happening: Many of the children end up with four eyes or two mouths, or two noses etc--The system doesn't 'know' which groups of triangles represent eyes, mouths, or an noses etc, so it doesn't now how to sensibly combine two face images.

The Fishmonger [PersonRank 0]

16 years ago #

What strikes me as odd here is that I haven't seen a single picture that looks like a woman. To what would you all attribute that effect?

bitbutter [PersonRank 1]

16 years ago #

Angular lines lend themselves better to drawing masculine looking features, and there was a very samurai-ish (male) looking image right at the outset (i just saw him on the 'before and after' page), I'm betting that he's the great granddad of the samurai army!

Philipp Lenssen [PersonRank 10]

16 years ago #

> To what would you all attribute that effect

One possible reason: the algorithm to draw the pictures uses triangles, so it's rather edgy, and perhaps that looks more male. I think it's also possible that there are more male and female raters, though I'm not sure if male raters are more likely to see male faces or if that is unconnected...

But now there's a second pool live on the site: animals, and I used a lot more round shapes for these ones (it's made up of triangles, but also circles and lines). I'm curious what will happen with the animal pool, as "animal" is a very ambigous term...

There's also a second face pool live now. It uses the same origin images as the first one.

Tony Ruscoe [PersonRank 10]

16 years ago #

I think a key factor of something looking like a face is some kind of symmetry and the random images with circles don't currently seem to have this trait.

Edit: Damn – I had no ideal I was looking for an animal with those circles. I completely missed the instructions since I'd been to the site before. Perhaps it would be an idea to display the subject in large letters, overlaying the image for a couple of seconds with a "continue" button just to make sure...?

Jacob Raves [PersonRank 0]

16 years ago #

Yeah, I'm interested in voting in the other ones (second face and animal). When will they be up?

Also, the samurai phenomenon is quite interesting. I want to see if thats where it will always go if triangles are used.

Philipp Lenssen [PersonRank 10]

16 years ago #

Tony, yeah, I was kinda afraid that would happen. I will have to think of a way to make that more clear, other than just using colors.

Anand [PersonRank 1]

16 years ago #

best ones i got :-)

http://img529.imageshack.us/my.php?image=facemutationct7.jpg

and the last one ,animal one, that was a shocker why not we just stick to faces ? :-P

Greg [PersonRank 0]

16 years ago #

They are up. Go to the main page.

Graham Snyder [PersonRank 1]

16 years ago #

I was going to suggest something that's been mentioned a couple of times before – that it might be more effective to compare pictures and select the one that looks more like a face – though perhaps rather than just 2 you could have 4 or 5 examples to compare.

Philipp Lenssen [PersonRank 10]

16 years ago #

Tony, pls refresh, I added a red "New!" sign when you see an animal, wonder if that'll do the job?

Tony Ruscoe [PersonRank 10]

16 years ago #

That got my attention a little bit... but I kinda new I should have been looking down there. Would it be better for people to switch between faces and animals manually themselves?

e.g. Faces | Animals

Anand [PersonRank 1]

16 years ago #

yes i like face and animal tabs

and btw animal thing is a bit hard to choose cause animals can be of really different shapes and sizes but lets see

Philipp Lenssen [PersonRank 10]

16 years ago #

> That got my attention a little bit... but I kinda new I should
> have been looking down there. Would it be better for
> people to switch between faces and animals
> manually themselves?

Hmm, I need to think about this. Last weekend during implementation I kinda realized this problem would appear, I pondered starting with at least two categories so people wouldn't be trained to ignore that sentece, which is what happened now. But I didn't see any cool second category that would be fun & somewhat recognizalble right from the start (except male/ female body, which lends itself a lot to symmetry, but I figured it might make the site non-family friendly after a couple of "evolutions"!). The current animal pool is really more of a "beginning" pool, you're barely able to spot anything, so I think the existing face pool kinda needs to "carry" it through this beginning phase until it will (hopefully) look more interesting. I hope people spot this New button now, the only risk is that people by and large ignore this, if only some people top-rate faces in that one these faces should automatically go "extinct" soon in the animal pool :)

June Bug [PersonRank 1]

16 years ago #

If only pac-man was an animal... Now I'm going to have to spend three more days on this site instead of doing work, just like the past two days!

Shawn P [PersonRank 0]

16 years ago #

This is great. I think for the animal specification it might be best to specify more types of animals. The face recognition seems to be good for specifically human faces, whereas the animals that show up seem to be all nonsensical probably because where I see a bird someone else might see something that is going towards a horse. If you split it up in to a few more specific categories it might start getting as nice as the faces are.

Graham Snyder [PersonRank 1]

16 years ago #

[put at-character here]Philipp: You mentioned in an earlier post that the mutation algorithm replaces a percentage of the polygons in the picture. It occurred to me that it might work better if it were possible for the algorithm to also "tweak" some of the existing polygons – slightly changing the dimensions, position and/or rotation each generation. For example, that way we might see things like the "sideburn" triangles shrinking into lines, rather than uprating them because they're otherwise realistic faces, or attempting to breed them out.

Graham Snyder [PersonRank 1]

16 years ago #

I just got a vaguely female-looking face! I think it's the pointier chin and the lips that do it. For want of a better system, here it is on imageshack:
http://blogoscoped.com/files/mutatingpictures/graham1.jpg
If only there were a "link to this face" button! :P

[moved image from img253.imageshack.us/my.php?image=faceft3.jpg to here for faster loading, if that's OK... -Philipp]

Philipp Lenssen [PersonRank 10]

16 years ago #

> It occurred to me that it might work better if it were
> possible for the algorithm to also "tweak" some
> of the existing polygons

Graham, I pondered this alternative I think it's really cool. Maybe I can do this for a future pool, right now I didn't quite know what to do with polygons that would travel outside the screen this way. I could just bump the polygons off at the edges if they intend to move outside, but that might start looking artificial...

Georgie R [PersonRank 1]

16 years ago #

I didn't think the animals was going to work because they are not symmetrical enough, but I think lots of different animals will come up and get refined as long as the rate of mutation is not too high. There seems to be lots of potential bird shapes!

James Xuan [PersonRank 10]

16 years ago #

I got some that looked like tigers.

Woo 100th post!

Graham Snyder [PersonRank 1]

16 years ago #

> right now I didn't quite know what to do with
> polygons that would travel outside the screen this
> way. I could just bump the polygons off at the edges
> if they intend to move outside

Why don't you just let them move outside? If the votes tend to make a polygon move towards the edge then it's probably better for the picture, right? Then once they're completely out of frame, you could either delete them or create a new random polygon somewhere in the image to replace them.

Sarah M [PersonRank 0]

16 years ago #

With faces, so far they all look like stylized samurai. Hard to get past that when large out of shape polygons fit so well with their helmets.

All I've been seeing in the animal section so far is bird shapes, but the ones that are starting to come through are starting to become fairly detailed in their profiles.

It will be interesting to see how it turns out in a few days, almost like a different form of Global Consciousness Experiment.

Patrick T [PersonRank 1]

16 years ago #

I think the reason why the big triangles have stayed is that:

1) They often delineate the face in a clear way so that everyone has the same reference as to where the eyes, the mouth and the nose should be. That will most likely be a problem for the animals, as we neither know which animal we are seeing nor its orientation. Faces are quite similar, and it was obvious that the face was going to be seen face on and not sideways. For animals, I suggest that you name the animal, and that we see only its face, face on.

2) They hide (partially or completely) many other triangles, thereby making the solution more likely to have a high fitness. Otherwise each hidden triangles has a potential of making the image less face-like. Here, I suppose there is no mecanism to avoid completely hidden triangles.

Colleen Sullivan [PersonRank 1]

16 years ago #

Bummer. It looks like face 1 is steady-stating; at least they look the same as when I went to bed. I was kind of afraid this would happen: you get to a certain point, and then mutations tend to move the image backwards, not forwards. May I suggest lowering the mutation rate as an image pool gets more realistic? Also I second the suggestion of allowing polygons to change shape, rather than just adding or removing them.

Patrick T [PersonRank 1]

16 years ago #

As to why the faces look masculine, I will speculate the following: feminine faces are more rounded, therefore producing curvier outlines, and slower transitions from dark to pale. I think there would be more feminine-looking faces if there were also gradients, arcs or circular segments http://mathworld.wolfram.com/CircularSegment.html

bitbutter [PersonRank 1]

16 years ago #

Hi, i've just deployed Face Maker, which is an _awful_ lot like Mutating Pictures (big thanks to Phillip for his permission to deploy it), but uses different scoring and drawing algo's.

No triangles this time, i wonder if the samurai's will still show up?
http://facemaker.redshiftmedia.com/

Anita [PersonRank 0]

16 years ago #

Wow, this is great!

One thought – why not reduce the # of polygons that change (or the % that they can move/alter) for pics that are rated higher? It seems strange to mutate a close match by the same % as a match that isn't at all close.

John [PersonRank 0]

16 years ago #

I noticed birdlike images coming up in the animal pool also, I thought it was just me.

Also as to what someone said about the faces starting to look the same, there is an evolutionary theory i remember learning that after a certain point in evolution, it is driven by "BIG" mutations that happen to stick and subsequently smaller mutations to perfect it. Maybe we are at that point now with the faces, where things will be relatively static until something dramatic happnes, that by chance is a big improvement on the face.

Philipp Lenssen [PersonRank 10]

16 years ago #

Wow, terrific job Bitbutter.
How big is the overall population?

On a meta level it's kind of funny how MutatingPictures.com itself created offspring now, with a mutated algorithm. If your faces will look much cooler than the ones at MutatingPictures.com over time, it will be additional inspiration for me (and others) to switch to a two-pics approach... which would mean that your mutated approach in turn created offspring.

By the way, sometimes the "evolving" is hanging (shows forever..).

dugg:
http://digg.com/tech_news/Face_Maker_help_pictures_evolve_with_a_spin

Philipp Lenssen [PersonRank 10]

16 years ago #

> One thought – why not reduce the # of polygons that
> change (or the % that they can move/alter) for pics that are rated higher?

Actually, that already happens since yesterday evening...

bitbutter [PersonRank 1]

16 years ago #

Hi phillip, thanks!

the population is about 500 (slightly more because of a bug early on ;)). I may generate some more though because there are lots of ppl voting.

I fixed the hanging bug (happened when trying to rate a pic that someone else had already deleted).

Philipp Lenssen [PersonRank 10]

16 years ago #

Seems your population is headed straight towards a cat right now bitbutter! :) Cool project.

Stephen Tordoff [PersonRank 10]

16 years ago #

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

Anita [PersonRank 0]

16 years ago #

Bitbutter – very cool, although it's looking like the "cat" is starting to pull away. A few hours ago I went through ~100 pics and only 10% or so seemed cat-like. Just now I did 50 and almost 40 of them were cats. Sometimes I feel like I need a "neither" option when voting – hard to choose which is better when faced with two similar cats. =)

Stephen Tordoff [PersonRank 10]

16 years ago #

Last screenshot (honest, I just think this is really cool / weird). Stylized face and skull type shape on head. Makes me think of bikers / rockers.

http://img.techpowerup.org/071002/Screenshot-2.png

MolokoV [PersonRank 1]

16 years ago #

This is quite interesting. Bilateral symmetry is more likely to activate human pattern recognition for faces. All faces basically have the same features in that symmetry, 2 eyes, nos, mouth, outline of face. It converges rapidly.

The animal gene pool is extremely interesting. It has no intrinsic symmetry and the category "animal" has many attractors. No viewpoint or orientation is specified, so it may take a while for some patterns to emerge for different kinds of animals at different angles. I notice some of them look like the faces of dogs or horses – large scale full-screen portraits. Others, only parts of the image look like an animal, a bird, a fish, a quatruped of some kind. Also a head with two ears and a body seems to be emerging in some images.

It's possible that being able to review the samples and read this forum might contaminate the schedule of re-enforcement, but I think as long as people can't direct specific components on the screen to do something specific (change size or position or orientation) there's no "intelligent design" going on, only selective pressure.

I think the animal picture is going to take a long time to converge.

Another problem with the animal picture is there are often several regions of the image that might look like an animal. People who recognize one but not the other may rate that image higher or lower depending on what they see or don't see in the image. Not really a problem, the animal images will converge to the most stereotyped animal shapes, I would think.

The first face genepool seems somewhat biased towards Samurai and Heavy Metal faces for some reason. It could just be that the angular/high contrast renderings resemble the graphics used in those genre, and japanese inkbrush paintings.

This may have value as some sort of projective psychological test, but it would take way too long to administer, you'd have to sit there and click buttons for days to get it to converge to some sort of personal Rorschach icon.

Scott [PersonRank 0]

16 years ago #

Am I the only one that thought of this clip from the Carl Sagan series Cosmos?
http://youtube.com/watch?v=RiNKt6gcEM8

bitbutter [PersonRank 1]

16 years ago #

[put at-character here]anita: "Bitbutter – very cool, although it's looking like the "cat" is starting to pull away."

Thanks. It seems the 'cat' has spawned a few strong children with a distinct character too. In one line the puffy white cheeks turned black, turing it from podgy to gaunt and its wide eyes shrunk to little black pebbles, giving a line of very solemn looking devils like this one:
http://img221.imageshack.us/img221/8638/picture24wn8.png

Georgie R [PersonRank 1]

16 years ago #

My bird theory was right although I thought other animals would come up too, I hope it will have evolved a pupil and legs by morning!
The faces don't seem to be getting any better perhaps the rate of mutation should be slowed down even further?
You should make a FAQ it's very interesting. When I first came to the site I misread and thought the project had been running for a year, it's amazing to see what can happen in just a few hours!

MolokoV [PersonRank 1]

16 years ago #

Same here.

"Heike crab" was the first thing I said after someone pasted the link to this site in an IRC channel.

http://farm1.static.flickr.com/157/373216489_e18a6473af.jpg

Jim Barnes [PersonRank 1]

16 years ago #

"Makes me think of bikers / rockers."

Yes, these faces are very metal.

Molokov: great thought re: the heike crab.

MolokoV [PersonRank 1]

16 years ago #

use of negative space ot imply form and this sort of 'gestalt pattern recognition' using fragmentary outlines seems to be happening in the animals gene pool.

these sorts of things:

http://mv.lycaeum.org/mutate/implied_negative_space.gif

I'll be stashing interesting things I run across at:
http://mv.lycaeum.org/mutate/mutatingpictures.html?z=071002

J. Mack [PersonRank 0]

16 years ago #

It seems to me all of the really good 'faces' look like samurai warrior types – what gives?

Onym [PersonRank 1]

16 years ago #

Regarding sexual reproduction. I think that a good strategy would be to execute some sort of correlation function between the two mated images. This would identify common features, after which a good strategy for closing in on a human face would be to give the common features less of a chance of being mutated (ie retain them) than features that are not shared by both images. I think that this would work quite well.

Onym [PersonRank 1]

16 years ago #

I don't think that your animal pictures are going to converge to anything – your question is far too general, so people are going to be seeing all sorts of species in there. Maybe pick a specific type of animal.

Onym [PersonRank 1]

16 years ago #

Also, I think that running the animal and face experiments at the same time is going to confuse new comers. They may not realize that two questions are being posed, and so screw up the results.

andy [PersonRank 0]

16 years ago #

A BIRD is very likely to evolve in the animal section as it only takes a circle and a triangle attached to it to suggest a head and a beak. This simple pattern, I guess, will then be the focus of attention and the rest will follow. It is hard to find such easy a pattern for other animals.

This actually is an interesting point when you think about the whole ID vs Darwinism debate where one argument for ID has always been "irreducible complexity", an argument which has pretty much been disproved. Here as well, there won't be any irreducible complexity, so the "evolution" will take one step at a time, and each step has to make some sense on its own.

Kenny [PersonRank 0]

16 years ago #

ugh yes let's turn this into a debate about creationism thanks

MolokoV [PersonRank 1]

16 years ago #

A much less interesting project than this would be to simply do a least-squares comparison between each mutated random frame and some preset image, then feedback the 0-10 measure of correlation.

Obviously the image would converge to the photograph over time to the extent that the variable image elements allowed.

There do seem to be some attractors shaping up in the animals genepool though. A horse head. There's a tendence to make blank areas surrounded by black flack outlines with a few features inside the blank area, and a tendency for large black objects to cluster together into body-like shapes.

Adding a "human figure" or "stick figure" genepool with no symmetry might converge a lot faster than the animals, but more slowly than the faces. The pose/posture of the figure would be ambiguous and variable so several different sub-images might emerge. Since the features of a human body are well-constrained different poses would probably hybridize reasonably.

MolokoV [PersonRank 1]

16 years ago #

hmn. Somewhat interesting would be the least-squares feedback scheme using many preset photos of several different target objects, – to see whether diverse targets feeding back anonymously into the same gene pool would make make multiple images arise, or whether they'd interfere with each other and cancel out/randomize the re-enforcement or otherwise prevent good convergence on any of the target images.

Another possible outcome could be convergence on the brightness average of all the target images, or the extraction of features common to all of the images.

MolokoV [PersonRank 1]

16 years ago #

I really like this "animal" thing. It has a human psychology component that can't be replicated (yet) using any other technique. Basically, we're being asked if anything in the image looks like any sort of animal at all... not whether it looks like some particular animal, but whether it evokes "animal recognition" or makes us project the image of an animal into the pattern presented.

It's kind of an exercize in evoked pareidolia.

A weirder, more general genepool would be "Does this look like anything to you?"

If there were a genepool fission algorithm that would detect clustering of responses a question like that could give rise to an unlimited number of very representational images. If the responses were uniformly distributed it would do nothing, if the responses clustered around two values with large gaps around the values, software would ask people what it was they thought it looked like, then split the gene pool, presenting the list of responses for each cluster as the question.

Still, I think the idea here is to see what can evolve without intervening at all in the 1-10 fitness value as the only source of feedback.

Philipp Lenssen [PersonRank 10]

16 years ago #

> ugh yes let's turn this into a debate
> about creationism thanks

As for the scientific debate, a lot has been posted here in the comments:
http://scienceblogs.com/pharyngula/2007/10/you_too_can_be_an_agent_of_sel.php

I also pondered the general "does this look like *something* (anything)?" question. Would be curious what it might evolve too on a non-symmetric picture. On a symmetric picture, I suppose it might evolve towards human bodies and faces, just as a guess (though mutatingpictures.com is now itself heavily biased to make people think of faces when you might be asked a general question like that while being shown a symmetric picture I guess...).

As for the current state of the bird, I'll give it a little more time before I start wondering if this specific pic creation algo is an evolutionary dead end :) Right now, the animals got maybe 1/20th or so of the ratings attention the face pool got on the first day, thanks to Digg.com, and it is admittedly harder to see something in it.

But here are two pictures I colored from the first face pool. It's starting to look like avatars for a fantasy role-playing game by now – great artwork (but lacking heroines right now :)).

http://blogoscoped.com/files/mutatingpictures/shadow-warrior.png
http://blogoscoped.com/files/mutatingpictures/devil-general.png

John [PersonRank 0]

16 years ago #

This could be sold as art you know...some of these faces are pretty cool. Art created by thousands of people, working "together".

Anand [PersonRank 1]

16 years ago #

some will look good as tattoo's ^_^

John [PersonRank 0]

16 years ago #

Yes, I know I'd get a couple of these faces as a shoulder tat.

Jedi [PersonRank 1]

16 years ago #

I swear I just saw the character "V" from V For Vendetta – amazing!!!

James Xuan [PersonRank 10]

16 years ago #

I love the ones you coloured!

John [PersonRank 0]

16 years ago #

I saw the V for Vendetta guy too. fascinating.

I think the main users of this fall into a similar demographic of young white american males, and similar faces look "cool" to them. It would be cool to see how this experiment would turn out with different demographics.

John [PersonRank 0]

16 years ago #

^^^just a guess about the demographics, btw

Anand [PersonRank 1]

16 years ago #

if i give 10 pts to some pic, would it stop evolving?

Philipp Lenssen [PersonRank 10]

16 years ago #

Anand, not at this time – right now, every offspring causes at least some degree of mutation.

bitbutter [PersonRank 1]

16 years ago #

"every offspring causes at least some degree of mutation."

Its the same in Face Maker. I have it set up so that if a face gets x number of points, it reproduces 5 offspring and retires. The offspring have mutation factors of 1,2,3,4,5. Every time a clutch is born though there is also a 1-in-50 chance that the offspring have mutation factors of 10,20,30,40,40 instead. This ensures that 'big' mutations that might give rise to new traits do appear in the 'gene pool' from time to time.

Bill Mac [PersonRank 9]

16 years ago #

I got this one early yesterday. I added dots for the eyes, but I think it makes it look really cool.

http://thebillmac3.googlepages.com/face.png

Georgie R [PersonRank 1]

16 years ago #

I thought the bird would develop faster than it is, I think the problem is there are too many extra shapes which often just get in the way rather than add features. Does the algorhythm allow reduction of number of shapes?

Also I don't understand what method is used to keep the genepool at a steady rate? If someone gives a 10 to a picture does it delete 10 random faces in order to accomodate the 10 newly created offspring?

Philipp Lenssen [PersonRank 10]

16 years ago #

> Does the algorhythm allow reduction
> of number of shapes?

Right now, only in so far as it allows both black and white color areas, and a white put in the background would effectively be invisible (until it mutates, though...). Note that the speed at which something mutates is only barely related to "human" time – the only real thing that matters is how many people use the site. The Digg crowd from day 1 might account for, say, a week of "normal" traffic. Though, the animal pool is getting a dozen or couple of dozen new members every minute since its inception...

And yes, the higher your rating the more mutated offspring a picture produces (not necessarily 10 for 10 though), and for each offspring a random member of the current population is removed & replaced.

bitbutter [PersonRank 1]

16 years ago #

philipp, i adjusted face maker lately so that images accumulate points based on the outcome of the battles, when new faces are generated (it happens when a face reaches x points), ithe faces with the lowest point scores that are removed to make room. I think this helps avoid destroying 'fit' faces. I thought your system used a similar approach, but you say that faces are removed at random?

(btw: There's a distinctly feminine looking face that's emerged in the facemaker mix. I thought it was interesting to note that its component parts are all pretty angular still:) http://img115.imageshack.us/img115/3568/picture6jy7.png

John [PersonRank 0]

16 years ago #

Finally those horrible Huge-Triangle-Cheeked faces are disappearing. My campaign is working.

Colleen Sullivan [PersonRank 1]

16 years ago #

Seems official. The first faces are devolving. I think they looked better after about 15 hours than they do now.

Any chance of smartening the algorithm to "learn" what kinds of features result in a higher score, and implement those more?

Jim Barnes [PersonRank 1]

16 years ago #

There is this new regime of extremely realistic chiaroscuro faces (I suppose they are all chiaroscuro.) I was floored by a couple, how realistic they looked.

The faces get symmetry, one of the most important aspects of face-likeness, for free. The animals don't. After a couple of days, the only thing that has emerged from the fog is that bird, which hasn't really changed except for the addition of an eye. Maybe we can't agree on what a bird should look like, and therefore we aren't able to "see" the birdlikeness that other people notice, things making it more birdlike.

I'd like to see one of these done with 3D polygons. You'd have to have a rotating head for it to work.

Another thing would be a symmetrical female form in a method similar to the faces.

Philipp Lenssen [PersonRank 10]

16 years ago #

There is now a new pool on the site: "a body" (with only circles, black and white, but with symmetry).

Jedi [PersonRank 1]

16 years ago #

The animal is definitely starting to evolve now... I have seen both a bird and a four legged animal of some kind – perhaps a dog/horse?? A few very nice faces popping up too..

Jedi [PersonRank 1]

16 years ago #

Well, in the animals I've now seen a bucking bull and a goose wearing a hat. With the faces, I'm starting to notice a few Eskimos and Indians – :)

MolokoV [PersonRank 1]

16 years ago #

part of the male face bias is that random flak inside the boundaries of the face get interpreted as facial hair, and angular forms outside the face get interpreted as battle armor or headdresses.

bitbutter's example of a female-like face is rather good.

Here's the one I ran across a day or two ago. It leans rather heavily on far-fetched media stereotypes to be interpreted as 'female'.

http://mv.lycaeum.org/mutate/face_f_071002x.jpg

MolokoV [PersonRank 1]

16 years ago #

the animals thing...

http://mv.lycaeum.org/mutate/haring_07.jpg

http://mv.lycaeum.org/mutate/guernica.jpg

One of your tinted avatars made me think:

"ALL YOUR FACE R BELONG 2 US"

Jedi [PersonRank 1]

16 years ago #

prediction: The face is going to get more abstract. The body is going to end up looking like the Michelin Man or the Staypuff Boy, and the animal is going to be a dog/duck hybrid... LOL

Anand [PersonRank 1]

16 years ago #

i think the the kind of shapes we have for "body" would be more useful for "face".

ibanex [PersonRank 0]

16 years ago #

The animal is starting to look like this...

http://mathworld.wolfram.com/images/gifs/rabbduck.gif

MolokoV [PersonRank 1]

16 years ago #

part of the problem with the animal is that the fitness landscape isn't monotonic... in order to get from one good shape to another good shape, or sometimes to a better shape, the image has to pass through a much worse shape.

Also, there's no 'speciation' so hybridizing a dog in the upper right corner with a bird in the middle wrecks the image entirely.

One of the definition of species boundary that has been used in biology is that the two genepools are capable of producing fertile offspring. So, as gene pools drift, at some point they become incapable of hybridization.

Allowing images to hybridize only with images that are less than some threshold least-squares distance away from each other would create de-facto speciation and would model well what happens in natural evolution.

Another option is to bias the probability that two images would hybridize inversely to their least-squares distance.

Another thing that happens in natural genetics is that a single gene will spit into two copies of itself which then mutate in different directions and come to perform different functions in the organism. This increase in complexity over time could be modelled by starting with very few polygons, then occasionally duplicating one of them. The result of this for the animal would most likely be starting with rudimentary shapes (head and body circles, for example) then adding limbs and other detail features once the main shape is established.

The tilted circle segments for the "body" are kind of dissonant. If a new shape-set is needed to make it obvious which genepool is being worked on while using the scoring selection page, making a GA that uses oriented, resized ellipses might work well. The gene parameters for each ellipse would be radius, aspect ratio, x position, y position, rotation and color.

Some genes in biological cells seem to be protected or 'highly conserved' there may be more than one process by which this happens. Obviously if all mutations of a critical gene are lethal, it will be highly conserved, but some genes are actively protected from mutation or copying errors by processes that evolve within the cell.

Adding a fitness score to each element inside the genome would somewhat model that, but corrupt the simplicity of the process.

It would depend on the intention of the algorithm. Are you trying to make something that converges well toward some ideal state, or are you exploring the limits of the simplest possible genetic algorithms?

  

MolokoV [PersonRank 1]

16 years ago #

This item about "human computation" is rather interesting.

Makes me wonder what kind of meta-consensus-finding applications genetic algorithm games like mutatingpictures.com might find.

http://www.youtube.com/watch?v=dtFroEJN1nI

Philipp Lenssen [PersonRank 10]

16 years ago #

For MutatingPictures.com ideally I want to create some brute force approach right now, mainly because there is limited "rater attention energy" to help power the site, and wasting that helping energy from all of us by fine-tuning complex algorithms (which are harder to invent & calibrate) doesn't seem to be an option at this time. However, if the site attracts a growing audience, I can launch more and more experiments, with more complicated algorithms... each distinct pool could have its own drawing styles and algorithms...

But even when you use brute force, like for instance with the animal pool, certain drawing genes can be more successful and show through. The animal pool has three shapes (triangle, circle, line) and two colors (black and white). However, only the overall amount of shapes is fixed, but if say the triangle shape would be fitter it could over time "eat up" the other shapes, filling the whole available "shape space" with its own form (similar with the colors black vs white).

It's quite interesting that from a programming point of view, you start to see this "crowd intelligence" almost like some living being, or like an energy cloud.

(On a side-note, CHI:
http://blogoscoped.com/archive/2005-03-25-n43.html
Amazon later implemented the Mechanical Turk which is the closest real-world implementation of this I've seen, but it's not really global – e.g. you need a US bank account. It may also be nice to see a non-profit do this, with helpers simply chipping in to advance science/ problem-solving/ culture.)

What I think is fascinating is that with this helper pool, a solution can be found to problems not a single individual participant from the pool may achieve on their own. I don't have proof but I *think* you don't need to have a single artist among the MutatingPictures.com raters to create incredible art work & beauty – because it's an easier job for us to tell if something looks "right" then it is for us to draw it right, for some reason. I would love to know what other kinds of problems can be found where the CHI/ crowd intelligence solution is better than the sum of its parts.
For instance, how do you tackle a complicated math question with crowd intelligence, if no math expert is among the crowd? Is there any way to break up the problems into parts that can be solved by the crowd – assuming even the programmer of the "crowd intelligence" program is *not* a math expert? (You can't just show the math question, because it would be too hard for every single member of the crowd... 50,000 people where no single one is able to solve the question do not speed up the solution!) If that's not possible, then what kind of problems lend themselves to the crowd intelligence approaches (and what generic approaches can be found to break up problems in crowd intelligence solvable sub parts)?

bitbutter [PersonRank 1]

16 years ago #

[put at-character here]philipp

"It's quite interesting that from a programming point of view, you start to see this "crowd intelligence" almost like some living being, or like an energy cloud."

"For instance, how do you tackle a complicated math question with crowd intelligence, if no math expert is among the crowd? Is there any way to break up the problems into parts that can be solved by the crowd"

There's a book i think you would find very interesting called 'the wisdom of crowds' it talks specifically about exactly these things.

bitbutter [PersonRank 1]

16 years ago #

I think that 'the wisdom of crowds' is an important advantage that these collaborative online experiments have over the original version of Richard Dawkin's Blind Watchmaker programme.

Philipp Lenssen [PersonRank 10]

16 years ago #

(Read it before, love the book...)

MolokoV [PersonRank 1]

16 years ago #

This kind of thing might be useful for solving non-trival problems dealing with joint probability distributions, or with answering questions about equlibrium states of economic models or game theory.

There may also be applications in establishing the weightings or topologies of bayesian networks, or in finding "hidden markov processes" in sequential phenomena for which there are no current models.

A somewhat trivial test would be to find the center of gravity of an irregularly-shaped 3D object. The object would be presented in random orientations, and people would be asked to click on where they think the center of gravity is. Each click would represent a line in 3-space along which the center of gravity might lie. The centroid of the region of greatest line density would presumably be the consensus estimate for the center of gravity.

"the wisdom of crowds" may not apply well at all, or be worse than analytical methods for highly counterintutive problems or highly nonlinear or non-contiguous ones. This may be one of the problems with democracy.

Crowd consensus does best when the underlying problem is one of central tendency or average values for some easily observed phenomenon.

Other interesting problems might be: predict the curve of the Euro-Dollar exchange rate for tomorrow given today's chart, estimate the path of an earth-crossing asteroid given a view of the local solar system with the recent paths each body represented as a velocity-direction vector, etc.

It's hard to find unambiguous mathematical problems where numerical methods don't give more accurate results than human opinions when all the variables are known.

I wonder if this approach would work with something like deciphering Easter Island rongo-rongo script, which is an arbitrary human phenomenon that's analytically intractible.

Another domain for this might be finding archaeological sites. "Click on locations where would you build a village in this aerial photograph of tropical jungle."

Philipp Lenssen [PersonRank 10]

16 years ago #

> A somewhat trivial test would be to find the center of
> gravity of an irregularly-shaped 3D object. The
> object would be presented in random orientations, and
> people would be asked to click on where
> they think the center of gravity is.

Is this problem solvable with current "AI" (math algorithms)?

If so, I'm curious if you can find some problems that aren't solved by today's state of AI.

> Another domain for this might be finding
> archaeological sites. "Click on locations where would
> you build a village in this aerial photograph
> of tropical jungle."

That is interesting.

Wonder if you can design a website this way: present two random designs, ask people how beautiful it is, or, how fitting to "brand X". (On a sidenote, I get a feeling in Facemaker's app, people – or at least I! – naturally tend towards the more beautiful face, even when the question is just "which looks more like a face?")

Wonder what would happen with crowd intelligence playing chess? Present an interactive chess board on a website, a master player is on one side, on the other side 50,000 people have to each decide on a move, the most popular move will be chosen (I don't think it matters if all of the 50,000 actually know chess rules? As long as the majority does!)
Would their chess move be the "dumbest" possible move? Or a terrific move? And would the crowd intelligence average towards some long-term strategy at all?

bitbutter [PersonRank 1]

16 years ago #

phillip: "(On a sidenote, I get a feeling in Facemaker's app, people – or at least I! – naturally tend towards the more beautiful face, even when the question is just "which looks more like a face?")"

haha yes. It was interesting to see that today Face Maker's population has swung quite strongly from female towards masculine faces, which are also slightly more lifelike and detailed than the female ones they replaced. I'm also guessing that the users of the site are predominantly heterosexual males (though there's no way of knowing)--if that's true it looks like people aren't just selecting on attractiveness, though I'm sure that still plays a big part (for me also).

What i think might be an important factor is a tendency for raters to value novelty. ie. in an all-female population of faces, a lone 'male' face that emerges will get more attention and be chosen more often as a result.

takitus [PersonRank 0]

16 years ago #

I find that in doing this there are many non-preferable pieces being used and I generally give the pictures including them a lower score in order to keep the faces from including those pieces. Before the samurai look there was one that looked like pigtails hanging down the sides of the face that I constantly voted under 5 for in order for it to phase out. This gets very frustrating however because of the fact that a face might almost be perfect except for those single pieces being where they are and usually oversized as well.

What would be great would be a palette of maybe 9 images to choose from in order to guide the direction of flow.

Also one thing I would like to recommend is seperating the animal, person, and what other ones you might create into seperate pages so those people who are trying to optimize one particular image dont have to deal with the others.

bitbutter [PersonRank 1]

16 years ago #

takitus: "I find that in doing this there are many non-preferable pieces being used and I generally give the pictures including them a lower score in order to keep the faces from including those pieces."

Did the pigtails make the image look less like a face? If they weren't affecting the face-ness of the image then selecting against pigtails might be counter productive in the long-run (slowing down the evolution by throwing out otherwise promising faces).

If a good face arises that has pigtails, i think it's likely that the pigtails would eventually disappear or mutate into something else anyway.

What I'm seeing happen in Face Maker is that as certain key features are 'solidified' in the population, selection moves on to improving less crucial things. First eyes and mouth are established, then nose, later facial details, and 'framing' of the face are refined. Once the face is looking pretty face-ish, attention naturally turns more towards improving the 'embellishments' (pigtails etc).

"What would be great would be a palette of maybe 9 images to choose from in order to guide the direction of flow. "

I much prefer it that the the 'goal' isn't defined (beyond the deliberately short-sighted instruction to rate each pic according to how face-like it is).

"Also one thing I would like to recommend is seperating the animal, person, and what other ones you might create into seperate pages"

I'd like this too. I have a great deal of trouble rating the animal pics, so i tend to skip those.

Kenny [PersonRank 0]

16 years ago #

I've not seen the progress in two days, and it appears to me that a) the progress in the second face strain is stagnating, b) the animal is not likely to EVER appear, c) the body hasn't come far either. That's rather disappointing.

Philipp Lenssen [PersonRank 10]

16 years ago #

Kenny, the progress the site makes is not related to "world time" but only to how many people visit to play. Right now, according to the site stats, these pools are in about "hour 4" or so compared to the the first pool on the first day! Why? Because the whole week *taken together* after launch day didn't have half as many hits as launch day yet – and now there are 3 pools instead of 1 so attentions is furthermore divided. And if I look back to day 1, in hour 4 or so, there wasn't much of a face visible either. So what I'm saying is not that these new pools must necessarily develop great over time, just that it's hard to judge the results right now because there would need to be more visits to make any good assumptions in regards to "pool fitness". In a way, if you expect the kind of results of launch day to repeat themselves, it's only going to happen when the site attracts 70,000+ visitors a day again, and that's a really, really rare thing for a site to happen I think (of course it'd be neat :)).

Luke [PersonRank 0]

16 years ago #

[put at-character here]philipp:

recently read this on techcrunch:
just what you proposed
http://www.crowdchess.com/

Philipp Lenssen [PersonRank 10]

16 years ago #

Luke, fascinating. These idea are in the air, it will be interesting to watch this evolve.

Jeff [PersonRank 0]

16 years ago #

I think an interesting direction to take this wouldn't be the change the initial shape pool, but instead start with the same sets of pictures, asking different (but related) words. Does this look like a... 'mans face/womans face' 'eagle/parrot' 'horse/cattle', and see how the picture pools diverge.

I'd imagine there would be some starting pictures that would be more face-like, and get high number votes for, say, both a man's face and a woman's face, but over time, the traits would start to look more and more like one or the other. Seeing a side by side animation of one picture changing into a face, then into 2 different faces (male and female) would be very interesting.

Simon [PersonRank 0]

16 years ago #

Would it be useful to start a gene pool with variable shapes within as well as a upper limit.
This would lower the workload on the server and allow for more simplification of overly complex images. The problem with some of the images seems to be that there is no place to put the excess shapes except around the image.

Philipp Lenssen [PersonRank 10]

16 years ago #

> The problem with some of the images seems to be
> that there is no place to put the excess shapes except around the image.

Right now, the excess shapes can overlay each other to turn "invisible", or also turn white (on white). Not sure if that's the best solution to the problem but it's one "survival strategy" to reduce elements.

Right now, the biggest detriment to speedy evolvement is the lack of raters unfortunately, though I'm really grateful to everyone who continues rating. In comparison there were around 70,000+ visits the first day, and now there's around 2,000+ visits/ day.

spot [PersonRank 0]

16 years ago #

nice app, see also http://electricsheep.org including full graphical display of family trees: http://sheepserver.net/v2d6/cgi/node.cgi?id=99829&detail=lineage

Kate C [PersonRank 0]

16 years ago #

Lovely site, I do love genetic algorithms. Is there any chance that you would make a t-shirt with one of the composite samurais on it?

Philipp Lenssen [PersonRank 10]

16 years ago #

Kate, great idea. I'll give it a try and post an update here.

Tikitu de Jager [PersonRank 0]

16 years ago #

I'd love to see an animation of the population average through time, if you've got enough data in your snapshots to make it fairly smooth-looking.

Also: nice work! (And yep, I'd buy a t-shirt.)

David Mulder [PersonRank 10]

16 years ago #

Is it just me, or have the results become worse?

Philipp Lenssen [PersonRank 10]

16 years ago #

I think there are some really varied, interesting faces still evolving. But it's not going towards more precision at this time, but towards more variety, so some of the new faces also look less face-like. Overall though, traffic further dropped, so evolution pace is dropping along with it. I'd estimate the current evolution rate per pool, simply going by visits and number of pools, is 1/192th of what it was during launch day.

Max Williams [PersonRank 1]

16 years ago #

Hi Philipp

Do you wonder at all about the size of the evolutionary step? In real evolution changes occur at a tiny rate, so in order to make the site more interesting, i guess that your evolutionary steps (ie the amount of change between a parent and child) are relatively large. As a result, it may be possible that a parent could be quite a good face while the children all diverge in different directions and are bad faces.

I don't know how this would be adjusted – of course, some 'intelligent designer' could come in and say "that looks very face like, lets decrease step size for that 'family' ". But that seems like cheating. Alternatively, step size could be set to gradually decrease as time goes on, but the problem with that is that evolution could slow down before anything interesting has happened.

I don't have any solutions myself, just curious as to your thoughts.

Philipp Lenssen [PersonRank 10]

16 years ago #

Max, yes I've been pondering the mutation amount too. I don't have any specific "best" mutation amount per offspring that I know of. However, since when you give a rating of say 10, the parent will create multiple offsprings. Furthermore, the same parent can be presented to others (until it dies) to create even more offspring.
So within all these offsprings, which all have the same mutation chance (but not necessarily all the same mutation amount as end result, it's only a random average chance), some of the pictures may actually create rather small or fine-tuned changes, while others indeed end up being "destroyed" versions of the parent. The results after 24 hours kind of seem to indicate this.

So in the end, with enough people rating, these "noisy" signals *may* die off over time, and the mutation is still optimized.

*However* fine-tuning these rules may speed up evolution, or differently put, the project may make less or more rational use of the amount of raters available to it, depending on the algos.

If, for instance, I would have managed to find an algorithm (if there is one!) that would have created terrific pictures at a visitor number of say 2,000 per day, then maybe these 2,000 people would have come back every day, told more of their friends, etc., so that the project would now have a higher or steady rate of raters – causing an even higher rate of good mutations/ results. Instead, some people became disappointed at the lack of evolution in some of the pools and perhaps deserted the site (the site is now at around 500+ visits a day). So, I think in that sense it *does* matter how well exactly you pick the specific algos for these kinds of experiments, because it will determine the "survival" factor of the site itself.

I am pretty certain (not 100% sure, but quite optimistic) that if the site still had 70,000+ visits a day, it could produce some more fascinating pictures, even for harder pools. Also, I as programmer of the rules would get very quick feedback in regards to which fine-tuning or which pool rule set works and which doesn't, allowing me to further optimize the sites algos. But right now that's not an option. Sill, it was a lot of fun for the first results this produced, and maybe every now and then the site will be linked from some bigger sites, allowing it to get a bunch of new visitors.

Max Williams [PersonRank 1]

16 years ago #

Ah, i never thought about randomising the amount of mutation – kind of obvious when you mention it.

It's interesting what you say about the 'meta' aspects of the rules that encourage people to use the site by encouraging quick evolution. A possibly interesting parallel between this and natural evolution is that generally an animal species wants to *not* get noticed by other species, while many vegetable species (flowering plants, fruiting trees, etc) *do* want to get noticed by other species. I wonder if there are any lessons to be learnt from the vegetable kingdom :)

Stephen Tordoff [PersonRank 10]

16 years ago #

What happened to the body pool?

bitbutter [PersonRank 1]

16 years ago #

"i guess that your evolutionary steps (ie the amount of change between a parent and child) are relatively large. As a result, it may be possible that a parent could be quite a good face while the children all diverge in different directions and are bad faces."

The way I've approached this with Face Maker is to assign each child of a new 'litter' a maximum mutation value. The actual mutation of the child's parts is determined by a random number between 0 and the maximum mutation value for that child.

I weighted the maximum mutation values for the chidren so that most of the children in a litter have a low maximum mutation value, but that there is always a more highly mutated child in the group.

I agree that manually tuning the mutation rate in response to vote frequency seems like cheating :)

An interesting, but more involved alternative would be to encode the child mutation weightings into the genome, and allow that to mutate, and to be inherited also. This way we might expect the mutation rate to regulate itself.

Philipp Lenssen [PersonRank 10]

16 years ago #

Stephen, it's still live, or what do you mean?

> I agree that manually tuning the mutation rate in
> response to vote frequency seems like cheating

What I did was to settle on a fixed mutation rate which is automatically dependent on the rating (a higher rating will receive more offspring, but less mutated amount per picture). But it took me a day or so to figure out a good mutaton rate, I finally settled on something around "16% – rating" or something!

But you could also (automatically) adjust this value by looking at the number of overall votes so far, or, perhaps more dependable, in the case of MutatingPictures.com you can also check how many high ratings there are on average recently. A pool that is receiving a lot of high ratings may simply get more fine-tuned mutation amounts. But for MutatingPictures.com, I didn't want to use any database to do any of this for reasons of performance. Luckily that meant the site was alive throughout most of the first 24 hours...

Stephen Tordoff [PersonRank 10]

16 years ago #

Edit: Ignore the following, just checked again and the body pool is back to how it was. Must have been some kind of glitch / bug.

Sorry, didn't make myself clear.

A few days (weeks?) ago, there was a number of image that were looking close to being bodies. However, now there are just semi-circles.

Forum home

Advertisement

 
Blog  |  Forum     more >> Archive | Feed | Google's blogs | About
Advertisement

 

This site unofficially covers Google™ and more with some rights reserved. Join our forum!