Here’s an interesting example of genetic programming: use a program that slightly alters colored polygons, compares the results to a target, and selects variants that most resemble the Mona Lisa. After less than a million generations, a black square turned into this:
Not bad. The description of the algorithm is a bit thin, but he promises to release the source code soon. It sounds like a million generations is an overestimate, since his population size in each generation was 1, and it also sounds like his selection was far more stringent than you’d find in nature, but it’s an interesting if oversimplified example of the power of chance and selection.
Forgive me if my non-scientist self has misunderstood, but doesn’t this imply evolutionary intention? Dawkins seems to make a rigorous case against intention or “completeness” with regard to evolution, which has but one aim: reproduction.
Reminds me of the web dawkins generated…
Well it shows that the author had intention. Evolution, not so much.
That is really so fascinating to me from an art perspective (I’m an artist, not a scientist). In this case would looking like the Mona Lisa be compared to some kind of stress on the form, such as surviving in heavy snow?
Isherwood: Don’t confuse evolutionary algorithms with biological evolution. Evolutionary computation is a class of search and optimization techniques, and as such, we’re more concerned with performance than with rigorous adherence to biological principles. In fact, the algorithms are such gross simplifications of biology that they really share nothing more than just the basic concepts of selection with variation.
Nature has a built in fitness function in the probability of reproduction, but that’s an awfully indirect way of measuring how “good” something is. Clearly, in the case of biology, it works, but we could have almost certainly evolved human level intelligence in fewer than 4 billion years had someone stepped in and selected specifically for traits resembling human intelligence. That’s what the designer of an evolutionary algorithm is doing — selecting for exactly what is desired.
Coo!
Methinks it is like a weasel
The painting was intelligently designed! It’s still just a bunch of polygons!!1!one!
Actually, I prefer The Blind Watchmaker Applet. It’s more interactive. You start with one square and then have 11 “offspring” with mutations. After only a few dozen “generations” you begin to get interesting images. I wonder if you can get a Mona Lisa from it…..now I’ll never finish that paper.
Wow a painting that didnt have a painter. Havent Kent Hovind and Kirk Cameron been looking for just that very thing? :)
@ deong: Hey thanks! You answered my question too :)
What would be fascinating is to apply this technique, but use facial recognition software as a selector instead of an existing picture of a face.
But the program is intelligently designed, and you know, even one intended to simulate biological evolution is, so, you know….
It’s sort of the IDiot argument against doing any kind of intelligent research into evolution, their ultimate goal.
Glen D
http://tinyurl.com/6mb592
“doesn’t this imply evolutionary intention”
Well, in this case, it’s just that the most “fit” images are the ones that happen to resemble the Mona Lisa. The “environment” is more hospitable to images that most resemble the Mona Lisa. In that sense it’s no more intentional than furrier wolves being selected for by a cold climate. The environment could be thought of an “intending” that the wolves have thicker fur, but it’s really just a result of a selection bias.
Saw this on Slashdot, too. I’ve written similar algorithms to evolve digital logic functions. It’s really not a big deal. It’s just continually adjusting some polygons until it fits the original image. It’s comparing to and aiming for the real Mona Lisa. In other words, it winds up with an answer it already had.
It’s just curve fitting. I did the same thing with stock market data. Wound up with some neural nets that perfectly “predicted” past performance of individual stocks. When you start trying to predict the future, it diverges wildly after just one day.
The real power of these things is to develop something that accomplishes a known task, but in a unique way that you might not have arrived at via tradition design techniques. A good example is this antenna design by NASA.
http://ti.arc.nasa.gov/projects/esg/research/antenna.htm
What would be interesting would be this:
Take *two* different images, and use try to maximize the fitness function for both of them at the same time. Or fit to a picture and also fit the polygon parameters to some other data set entirely.
Hmm, I think I have a new project…
Wow a painting that didnt have a painter.
Are you suggesting the Mona Lisa is a natural formation? ;-)
I never found this sort of demonstration very convincing. Nobody doubts that randomly manipulating data and then selecting the data that looks most like some target will generate progress towards the target. The real point of evolution by natural selection is that merely randomly manipulating data and then selecting the data which happens to be most useful at present (rather than the data which most resembles a desired future state) can actually produce well-engineered complex systems over millions of generations.
This sort of simulation can reinforce misconceptions about evolution as much as it can inform and educate.
It’s an interesting idea, but I’d expect the results to be not very shocking, depending on how well the facial recogniton algorithms themselves are understood. Hazarding a guess, I’d say you’ll probably get things that look a lot like Eigenfaces (see http://en.wikipedia.org/wiki/Eigenface), or images with whatever features your algorithm picks up as being a “face” exaggerated, noise or blank everywhere else.
Isn’t it obvious it was the hand of God giving us a sign as to his power and knowledge of art history?
This is relevent, a Genetic algorithm that evolves a better car:
http://www.boingboing.net/2008/12/08/genetic-algorithm-ev.html
Well, I think the real trick is how rapidly it can change. It’s one thing to converge on a target, but evolution is also noting that targets “change” based on environmental changes. How many generations would a Mona Lisa face like this need to go through to turn into an Abraham Lincoln?
Chance and cumulative selection.
Let’s add insult to injury, and apply this procedure to a painting of a banana by Thomas Kincaide.
And anyhow, how can you be sure that God didn’t tweak the program without being directly detected?
That’s about what Behe’s “argument” comes down to in its most recent incarnation.
Glen D
http://tinyurl.com/2kxyc7
Well, yes, this is not very convincing evidence of the reality of natural selection; good thing biologists have almost never used genetic algorithms as evidence for biological realities. If anything, genetic algorithms have followed biology: as computers have increased in power and programmers in skill, genetic algorithms have begun to incorporate elements long know to be part of real natural selection.
Isherwood:
Not really. All that matters in a genetic algorithm is that an optimum (at least a local one) in the fitness function exist. Saying that in biological evolution nature provides the fitness function is too anthropomorphic; it’s not meant to imply anything about intention.
I just lost my appetite.
He set God’s permission to read only.
Possibly the way he set up the algorithm is that the population consisted of the individual polygons that make up the image that represents each generation.
This is not evolution, it is stochastic climbing, a type of Montecarlo numerical analysis.
If you like evolutionary simulations, the best one I’ve ever seen was called Primordial Life (http://www.io.com/~spofford/prim30.html). It simulates tiny organisms using a pretty simple set of arm units (predator arms, photosynthesizing arms, defense arms, movement arms) and totally undirected evolution. You can direct things a little bit by fiddling with ambient conditions to encourage plant growth or motile units. (Or if you like to play god, there’s a tool to directly smite evildoers.) Sadly, the project is dead, but I might have the code for the 4.0 beta lying around somewhere.
That was good thinking! Got to control those variables.
Glen D
http://tinyurl.com/6mb592
Bad example, actually, since fitness is defined as proximity to the Ideal Model, this is at best Lamarckian evolution… a better example would have been to code a simulator of painting popularity rating and use it on the successive generations, but that’s much harder ;-)
“it’s an interesting if oversimplified example of the power of chance and selection.”
Except that they started out with the goal of getting to the Mona Lisa.
Come on, PZ. We all know you’re a dishonest sleaze, but how stupid do you think we are?
We or just you?
@#6 It is back’d like a weasel.
“He set God’s permission to read only.”
Can you really do that? I thought God had automatically root privilege…?
su –
JesusMary0000
cd /local/humans/followers
rm -r rationalthoughtinstructions.txt
ObPitr: God, root, what is difference?
The good Rev BigDumbChimp (OM) has eaten another troll…
I think this will only encourage IDers, since the end result is so specified.. Nature doesn’t work that way (at least I don’t think so…is there is selective pressure for precisely a timber wolf?)
What might be better is to put as simple a constraint on the end product as possible, something like “be faster!”, and see what simulated evolution comes up with.
Just such a thing is shown in Jordan Pollack’s GOLEM project (Genetically Organized Lifelike Electro Mechanics). Here they use genetic algorithms to start with a population of randomly configured non-moving robots and then they evolve the ability to move. What’s most fun is at the end they print the robots out on a 3d printer and they move along on their office’s carpet.
Here is a YouTube video about it, but skip the 39 second boring credits screens, and then after you tire of seeing the animations of the robots, the real ones are shown being 3d printed at 4:10 and then moving at 4:43:
#15: “I never found this sort of demonstration very convincing.”
These things aren’t demonstrations of evolution, they are demonstrations of genetic programming algorithms. Genetic programming has a lot of engineering applications.
“Come on, PZ. We all know you’re a dishonest sleaze, but how stupid do you think we are?”
I think your quite stupid actually. The rest of us not so much.
Isherwood: sure, the author intended the result to look like the Mona Lisa. Which is where a Creationist would like sticking in a wedge and stating, “See? Evolution only produces something like this if it has a designer, or something to guide the process.” The problem here is, of course, that we didn’t have to look how we do now. Creationists like to work backwards and say that we evolved towards where we are now. In actually, we evolved in a completely meandering fashion and ended up looking like this. If I wander aimlessly in a field and take my position after 45 minutes, the Creationist would state, “what are the chances that you ended up in that exact location? You must have been guided there by an intelligent source.” The Evolutionist would realize that the chances of ending up somewhere are 100%. This is the common argument for improbability, and why people invoke the 747-jet spontaneous assembly analogy (and fail to realize the assumption that nature has no intent).
Wow! I would like to see what I could do with that piece of code. Interesting experiment indeed.
In other words, evolution is why most women don’t look like the Mona Lisa, and to be fair, why most men don’t look like Michelangelo’s “David” (then again, he’s not well-endowed, so that’s not all bad).
And why, if we all died off, nothing that looks like humans in general would evolve.
The point I like to make with GAs is not that they are good simulations of evolution (I don’t like the teleological aspect at all, that being a mistake even many who accept evolution make), but that it gives the lie to the notion that evolution isn’t useful in a way that people can understand (“guiding principle of biology,” while important and true, means little to most people).
GAs definitely owe their origin to evolution, and are useful in much the same way as evolution is expected to occur, by making gradual changes. Leaps in thought require intelligence, with GAs and with evolution. Designers can use both intelligence and GAs, while evolution can’t–and life is devoid of the leaps of creativity of which human brains produce.
Glen D
http://tinyurl.com/6mb592
One might be interested in this as well:
http://www.wreck.devisland.net/ga
Flash program which features an ‘evolving’ car.
misc that’s pretty cool!
given these steps:
why would anyone expect different final results?
This is not evolution.
@NewEnglandBob
Ever heard of the “Ecological niche” concept?
Hint: There’s an article on Wikipedia.
Back in the ’90s there was a Genetic Art project by John Mount that was a better analogy for biological evolution, since there was no predetermined target image or built-in fitness criterion. Instead, fitness was determined by the number of eyeballs visiting each image in the online gallery. Images that attracted the most hits were bred to produce new hybrid images. The Intenet user community formed the environment in which images competed for attention and evolved toward greater success in that competition.
is she smiling?
can’t quite make it out.
NewEnglandBob:
Of course it is not evolution. It is just the Montecarlo numerical algorithm applied to an image:
http://en.wikipedia.org/wiki/Monte_Carlo_method
Except that they started out with the goal of getting to the Mona Lisa.
No, not really. What they did was make the Mona Lisa the least “appetizing”. Each generation you make a number of random variations to the previous generation. All but the one that is most like the Mona Lisa get “eaten”. Repeat for many generations.
The difference is that each generation is a random variation of the previous. If it was directed you would be doing the comparison to the “target” before generating the next generation, not eliminating the “bad” ones afterwards. That is the difference between Lamarkian evolution, where the current generation directs what the next generation looks like based on current need. Darwinian evolution simply eliminates the least fit of the random variations.
Of course it is not evolution. It is just the Montecarlo numerical algorithm applied to an image:
But isn’t evolution just a Monte Carlo algorithm applied to life?
@moother
Isn’t it plain obvious? She’s 83 percent happy, 9 percent disgusted, 6 percent fearful and 2 percent angry.
This doesn’t show evolution; it shows convergence, driven by random change with modest levels of selection. One interesting question is how modest does the selection have to be to produce a convergence. Another, as Joe Shelby pointed out above, is that the the environment that defines success of fit (for the purposes of evolution) continually changes, so that the target of convergence is moving: how adept is the program at shifting its aim and, possibly, its selection to begin homing in on another good fit?
This is not really a genetic algorithm. It’s just iterative hill climbing (http://en.wikipedia.org/wiki/Hill_climbing). He’s got mutation and a fitness function, but that’s it. A real genetic algorithm would have a population of chromosomes being tested for fitness, probability of selection based on fitness, mutation and probably crossover (mating). It would not only be a more coherent model of real evolution, but would probably be a faster search in this case.
@skepsci
> I never found this sort of demonstration very convincing.
> Nobody doubts that randomly manipulating data and then
> selecting the data that looks most like some target will
> generate progress towards the target.
Although in this case there was clearly a very specific, desired outcome, other genetic algorithms do perform a better job of demonstrating some of the key principles of evolution in the real world (emphasis on ‘some’).
For example, take water network design, in this case the target design is clearly not known in advance (I’m talking pipe diameters here, not pipe layout, where the options are fairly limited), and is an extremely complex problem. The selective pressures are flow and pressure at various points on the network (of course, the size of pipes will be constrained by industry standards), and cost.
The genetic algorithm will attempt to meet the flow and pressure requirements within the network, by ‘mutating’ the pipe diameters, whilst keeping the cost associated with the solution to a minimum. The algorithm has no idea what the optimal solution looks like (it may find an optimal solution, or stop at a near optimal solution – in the worst case scenario it may fail altogether), all it can do is attempt to meet the demands of its environment.
“you’re”
This doesn’t show evolution; it shows convergence, driven by random change with modest levels of selection.
But that is evolution. How is this any different than exposing bacteria to antibiotics and evolving a resistant strain? There you are giving the bacteria a “goal”; to resist that chemical. Or a species moving into a new environment, like an island that doesn’t have the food it used to eat. Now there is a “goal”, to be able to live on the food available. Random mutation and selection and the the species will converge on a capability of eating the new food (or go extinct).
@Quiet_Desperation (#13):
Another example similar to the NASA antenna was an expeiment done using a FPGA to evolve a circuit that could differentiate between 1KHz and 10KHz input signals (more information here: http://www.setiai.com/archives/000031.html ). There are better articles than that one but my google-fu is weak at the moment and I couldn’t find the one I was after :(
I took around 5000 generations, but they had a solution which worked. However they noticed some strange things:
* The FPGA consisted of an array of 64×64 cells – but the end design only used one 10×10 section.
* The final design had circuits which didn’t connect in any way to the input or output. Yet when they removed them the whole thing stopped working.
* When they copied the final design to another, identical array (identical in model type, etc) the performance was degraded. The final design actually utilised some of the physical properties of the chips in a way that was not expected.
* When they copied the design in a simulated environment, or using an alternate technology (using CMOS chips instead of a FPGA) they didn’t work.
@15: Look into Tierra for one of the first examples of a system with truly open-ended evolution. You quickly get an ‘ecology’ of hosts, parasites, immune hosts, cooperation, etc. (I did a version myself and came up with a few interesting results, too.)
Andy C:
I dig some genetic algorithms, so I would have been very confused if this didn’t produce something similar to the Mona Lisa. That would have been a very poorly designed fitness function. The most interesting thing about this is the relatively small number of generations used to produce it.
Genetic algorithms and art have a bit of a history, and it’s interesting to see how different artists develop an algorithm.
As an aside, check out Saint Ambrose by Rodney Waschka II (I took his class on computer music at NC State U).
http://www.capstonerecords.org/CPS-8708.html touches on the approach to composition and the inclusion of algorithms in the composition (in this case, not too heavily – he started with ‘musical’ bits instead of random sound or notes). There are also some audio samples.
The earliest (in my recollection) occurrence of genetic musical composition was a composer/professor who set up a station in the hallway. Students could sit down and act as a fitness function by choosing “like” or “dislike” for each iteration of the software. This led to a lot of realizations about how we perceive music in the greater landscape of “sound” and led to genetic compositions that had better starting material.
Anyhow, enough rambling.
Ray @ #61: I’ve always felt that Tierra has been overrated as a model of evolution. The ancestor organism was hand-coded, and the descendant organisms were all just degenerate versions of it, using various shortcuts (such as parasitism) to streamline reproduction. No novel coding idioms evolved in the course of the simulation; just recombinations of the ancestor’s intelligently designed code. The most interesting result was “unrolling the loop”, and even that just involved multiple copies of a code fragment present in the ancestor.
What this says to me is that the assembly-code model used by Tierra replicators is not a very evolvable system. The fraction of possible code sequences that are successful replicators is vanishing small, much smaller than, say, the proportion of RNA molecules that can autocatalyze their own replication. A more robust LISP-like system in which code “atoms” self-assemble into executable “molecules” might provide more latitude for spontaneous innovation.
There is a similarity to natural selection in this,if you consider mimicry.Bark,leaves,twigs and whatnot are the target and birds are the fitness function.Or even emperor crabs in Japan,where centuries of fishermen threw back crabs that had markings that looked like the emperor.They eventually came up with crabs that had a reasonable face like marking.
Hah!!!
This is proof of teleology. Evilution is reverse engineering software.
Dammm, everytime I lurn something new it’s wrong.
If it works, it isn’t AI.
Epikt:
I’m not quite sure if you understood the point of my post (I do accept evolutionary theory in case there is any confusion).
skepsci had originally commented on ‘these types of demonstrations’ reinforcing misconceptions about evolution, because of there being a predetermined target. My intention in bringing up the genetic algorithms for water network design was to point out that useful genetic algorithms exist that ‘evolve’ towards a state that is not predetermined, in the same sense that evolution does not know where it is going in advance, thus making it a more representative example.
IAmMarauder @60:
Uh oh, pretty soon, engineers will be out of jobs. Self-writing programs and genetic algorithms will make human thinking obsolete. How sad.
“Another startling example of the sheer strangness of successful genetic programming is computer code that was evolved to help a patient control a prosthetic hand on the basis of erratic nerve signals picked up by electrodes taped to the patients wrists. Reported in Scientific American:
‘The evolved code [was] as messy and inscrutable as a squashed bug. [The] gesture-predicting program consists of a single line so long that it fills and contains hundreds of nested parenthetical expressions. It reveals nothing about why the thumb moves a certain way-only that it does.'”
–The Intelligent Universe, quoting from Gibbs’ “Programming from primordial ooze”
and earlier…
“NASA used genetic programming to come up with an optimal design for a girder to be used on the ISS. As reported in U.S News and World Report, the result was straight out of a science fiction novel:
‘There emerged, from 15 generations and 45,000 different designs, a truss no human engineer would design. The lumpy, knob-ended assembly [resembled]…a leg bone, irregular and somehow organic. Tests on the models confirm it’s superiority to human-designed ones as a stable support. No intelligence made the designs. They just evolved.'”– this time, Petit, “Touched by Nature: Putting evolution to work on the assembly line.”
Awesome. I’m a Computer Science graduate student so I understand very well how he did this. The only thing missing from his explanation is what algorithm he used to compare the “evolving” painting to the original. It’s frankly amazing that 50 semitransparent polygons can render the Mona Lisa that accurately.
It suggests a potential algorithm for video game designers where an extremely complex model (far to complex to render in real-time on the target platform) is reduced to far fewer polygons via a genetic algorithm.
I’ve been wanting to play with genetic algorithms for a while but haven’t gotten around to it.
That’s cute.
Now what I really want to see is an animation. Maybe one second of each generation up to the 10th, then of each 10th generation up to the 100th, then of each 100th generation up to the 1000th, etc. It’d be short, but potentially very informative. If it’s less than 1 million generations the animation would take just around a minute.
Andy C:
My point–and there was one buried in there–was that I agree with you. I simply wanted to reinforce the part of your comment that seemed to be saying that GAs could solve cool problems without having the solutions hard-wired. And I suppose it didn’t help when I threw in a mild disagreement with Glenn. So, no, I didn’t mean to suggest I was taking issue with what you said. Apologies for the lack of clarity.
As some have pointed out, this example is nothing more than a convergence algorithm and doesn’t say a damn thing about whether evolution could have happened. If you think otherwise you are self-deceived (don’t be offended — I am practicing for a Molly award). But what about other genetic algorithms where the right answer isn’t already known and programmed into the solution? Well those are just search-optimization algorithms and don’t prove a damn thing about evolution either.
Maybe someone would care to propose requirements for simulation which might crudely represent evolution. For an example of what not to try, consider the convergence algorithm proposed by Kel and see if you can satisfy some of the concerns I raised.
*eyeroll*
Randy, what are you blabbering on about?
Seriously, “could have happened”? What kind of horseshit is that? Biological evolution is an empirically observed phenomenon, beyond any shadow of a doubt: it happened. To simply deny the empirical fact that evolution occurred is just willful ignorance and stupidity, and no amount of feeble-minded intellectual masturbation over what “could” and “could not” have happened is going to change what did happen or make the collosal mountains of evidence vanish in a puff of wishful thinking.
Now whether natural selection, as a theory, adequately models all the mountains of empirical evidence completely is open to discussion, but the facts and the evidence are not. Nobody claims that Darwin’s original conception of natural selection is a complete and exhaustive explanation today, which is why, after 150 years of research and refinement, we have a greatly more complex and complete theory.
What part of this don’t you Creationists get? It’s fucking frustrating.
My algorithms didn’t have to do with evolution, they were there to show simply that you can get something complex through a process rather than by chance. Again, please read The Blind Watchmaker. Dawkins points out when he wrote similar algorithms that they aren’t simulations of evolution, rather than are an analogy of how processes can generate complexity.
The Mona Lisa in this case is the environment, the random changes are the mutations and it’s closeness to how it looks like the Mona Lisa is the selector. You can see over time that selection shapes the design of the painting, much like in reality that selection shapes the organism by selecting positive changes. Most mutations in nature and in this instance would be neutral, but the few positive changes cumulate over successive generations.
I disagree with those who say this has nothing to do with evolution. First of all it IS evolution. It is not biological evolution by natural selection, but the image does evolve, obviously. The word had meaning prior to its biological use, and that meaning applies to both.
More importantly, this directly takes on the argument that biological evolution is “random” and therefore shouldn’t be able to make things of sophistication. It is true that every “improvement” in biology resulted from something random, a non directed mutation. Likewise, every improvement in the image was random: a non-directed modification.
In both cases, the other half of the process, selection, is what gives it direction. But selection didn’t actually affect the modifications themselves, it came afterwards and discarded large numbers of them.
So there is a huge parallel. Maybe as someone who already gets evolution, this is not so useful. Maybe if you are arguing with a hard-core creationist, this won’t convince them. But to your average person who hasn’t thought about it that much, but is puzzled by the idea that all sophistication of life came from random mutations….well, same here, and it obviously *does* work. If it doesn’t “prove” evolution, it does get people more comfortable with some of the main ideas behind it that seem counterintuitive until you see them working on something far simpler.
BAHAHAHAHAHAHA!!!!
Laptop computer: $2000
Internet connection: $60
Rent for mom’s basement: $0
Misspelling words while trying to criticize other people’s intelligence: Priceless
randy, WHERE’S MY DAMN ANSWER!!! Put up a fucking response on your site you twit, otherwise you have no right to be wrong here!
AND YOU STILL NEVER RESPONDED TO THE BLIND WATCHMAKER VIDEO BY cdk007!!! One issue at a time, God Dammit!
Kel, there was another of cdk007’s videos that talked about precisely this. But as i refuted him with an earlier video, and he ignored the importance of that (MOTHER FUCKER SHIT DAMN FUCK DAMMIT BASTARD!! *pants*), i doubt any words will work. I am sick of randy the IDiot with selective reading, and at this point would consider him worthy of the Axe from PZ just for being so damned boring
All that said randy, you are a nice guy and if i met you in real life we might be able to become friends, if you never (ever ever ever ever…) brought this subject up.
I’m sick of Randy misusing my post, but what can you do about it? My post was not about evolution at all, yet he keeps on about it being a bad example. That’s because it was not one!!! On my tags I didn’t even label it evolution, the tags were “mathematics, programming, science, statistics”
He’s just trying to infer something from it that wasn’t there, and then use his inference to make a point. As I said on there before he even turned up:
Yes, selection doesn’t work like this. But this wasn’t about evolution, it was about information theory and how we can determine what is and what isn’t a product of an intelligent agent – something important to distinguish when tackling the subject of DNA. There are some who say “DNA is a code, and all codes are a product of intelligence”, and this along with a future post are to address that issue.
As for how to simulate natural selection and evolution, I’m still trying to figure out how to adequately represent the process in a computational sense. But I’ll save that for another time.
This example, Kel’s algorithm, and the blind watchmaker algorithm are all convergence algorithms. The algorithms don’t represent random mutation because they have the right answer built in.
The evolved antenna was the result of a search-optimization algorithm. I’ve written similar programs. There algorithms start with something that works already — an approximation of the optimal answer and modifications are tested from there. It’s not like you are generating something from nothing using random mutation.
The video that RickrOll pointed me to is more interesting. It’s not like I can’t refute it. I just need time to think about how to say it so that his pea brain can understand it.
You mean they don’t represent natural selection? The randomness is built in behaves (practically) like randomness in nature. It’s the selection criteria that’s different, we are playing the roll of selector by having one point on which to obtain, where in natural selection it’s simply a matter of survival and whatever satisfies that survival criteria.
Like Dawkins stated 21 years ago, these serve as nothing more than analogies. They should the power of cumuulative selection on random heredity, but are not representative of how evolution works beyond the analogous.
I fail to see why you are trying to pretend otherwise, we know they aren’t representative of evolution, just useful to explain how the process works.
Kel said:
Let’s not pretend that you know something about information theory or that this blog post of yours says something about it.
Randy, getting snippy because you didn’t like what i said? Then talk to Me! Don’t get your panties all in a knot and argue with someone else about something You know Nothing about simply out of misplaces feelings.
I’m going to vote “not evolution” on this one too.
memcpy could have done the same thing, far faster, far simpler.
The objective function needs to be something that implicitly specifies the goal in a complex space. There are much better examples of evolutionary computation out there.
RickrOll,
I wasn’t getting snippy. I am just trying to get nominated for a Molly award. Calling you a pea brain hardly compares with
I think you are projecting your feelings onto me. You also have a much better chance at a Molly award.
You’ve been pretending to know about evolution for years without actually doing so, why should knowledge actually get in the way?
I’ll state quite clearly: The algorithms do not simulate evolution, they are merely analogous. Dawkins even admits as such when he did his version of the algorithm. Why are you persisting complaining about it not being evolution when no-one is claiming it to be?
Moi, cervelle d’oiseau? *forgets to laugh*
I meant: ([aside] MOTHER FUCKER SHIT DAMN FUCK DAMMIT BASTARD!! *pants*); which, by the way, was just regular cursing and swearing, not directed at you per se. It is infuriating to watch you continually avoid answering questions, Like Just Now, for instance.
Mollies can wait. I think i’m in the running for May of ’09 ;)
Kel,
The title of your blog entry is “There’s a monkey sitting at a typewriter” implies that you are trying to make a point about evolution.
Then you ask
The answer is that it would take so much time that from a practical sense the answer is “no”. But then to paraphrase you ask if a convergence alogorithm can generate a single sentence? The answer is “yes”. But who cares. You told it what sentence to generate.
Randy,
The only people who use the monkeys at a typewriter for evolution are creationists.
Yes, but so what? It shows that being selective is the opposite of chance.
Randy, don’t leave now, you haven’t made a point yet! Or ever, if truth be told.
I am a biology grad student and fully adhere to Darwinian evolutionary theory but I must say I have never been impressed with these computer programs. Evolution is not goal-directed, and my experience with my undergrad students is that computer simulations often introduce misconceptions of their own.
I think a better idea would be to have the program evolve similarly to sexual selection: maybe have it select for pictures that have characteristics like symmetry instead of a specific goal of the Mona Lisa. That would demonstrate that subtle selective pressure can lead up to the illusion of design, instead of evolution having design in mind in the first place as the program used implies.
I am appalled by this; was the aim to convert Kazimir Malevich’s “Black Square, 1913” into Leonardo Da Vinci’s “the Mona Lisa”? I have seen the original of both, and I must say that I prefer the “Black Square”.
Why would you need a computer program to show that?
For Randy:
Because a lot of people make the mistake of the dichotomy between chance and design. How many times have you heard a theist say “I don’t see how this could all come about by chance” when either talking about life, the universe, or anything? It seems that’s the theist argument for God and it’s a bullshit one because it negates the natural processes and laws of reality that can build such complex structures free of intervention.
So why would we need a program to show that? Because simply stating that selection isn’t chance doesn’t work for most people.
http://www.homowebensis.com/bichos4.html
Randy Stimpson aka Intransigent Dissembler:
False dichotomy. The pure-random scheme on Kel’s website also “knows the answer in advance.” It has to, in order to know when to stop. The fact that you may be using wetware to do that doesn’t mean that it isn’t part of the algorithm.
Stimpy:
I see. So on the Mona Lisa site, this “1) Copy the current DNA sequence and mutate it slightly” actually means: “1) Copy the current DNA sequence but do not even think of mutating it slightly”
Silly me. I have no idea why I thought they included mutation.
Seriously, have you actually written a genetic algorithm code? If you don’t put in mutation, your system will almost certainly get stuck in a local optimum, and you won’t find the global one unless your choice of initial population was fortuitous.
@64: Gregory, as a full model of biological evolution, Tierra-style models definitely fall short. However, I think you understate their results somewhat. Depending on how you classify what counts as a ‘coding idiom’, I’d certainly say that some have developed. In my own version, the ‘organisms’ ‘discovered’ aspects of the opcode set that I myself had not anticipated, and took me some investigation to figure out how they worked.
Later elaborations like Avida show that entirely novel algorithms can be developed in this manner, though there’s more of an explicit ‘fitness function’ in that case.
You’d probably like the Darwin at Home site, which gives you software to construct an object made of triangles. It can move and bend, a little. It is selected for movement. The program makes a bunch of them, then selects the one that moves farthest as the parent of the next generation. The site has some videos of what evolved from that simple selection pressure–all different, but all moving.
I just had the most amazing idea. Evolutionary art gallery!
So here’s what I propose. You have an art gallery that starts with a random assortment of abstract images. The images vary in some way (this is the hard part that I don’t know enough about art or computers to be useful for) and visitors to the gallery get to decide whether they like or dislike an image. Every so often, the images that were most like get to breed and restock the gallery. Presto!
Hmm, interesting idea chris.
PZ Myers: “(The Mona Lisa algorithm is) an interesting if oversimplified example of the power of chance and selection.”
The algorithm demonstrates the obvious: If chance and selection are asked to work in concert to reach a targeted outcome, and if the selection function has the target in mind, then the target will inevitably be attained even if each step towards the target is randomly induced. The only question is how many generations are required to reach the target. Since natural selection has no targets in mind, the Mona Lisa algorithm has no relevance to the Darwinian mechanism of random genetic mutations and natural selection (as the creator of the algorithm has more or less admitted).
The Mona Lisa algorithm – like all such algorithms – smuggles intelligence into a process that is supposed to be acting without intelligence. If these algorithms argue for anything, they argue for design, not Darwinism.
Re: #102 and other “evolve artwork with aesthetic appeal as fitness function”:
http://www.karlsims.com/genetic-images.html
Also, I agree with everyone who said the Mona Lisa thing is just Monte Carlo with a questionable analogy to evolution implied by the terminology used.