Since I posted my evolving Mona Lisa post I have received allot of questions and comments regarding the simulation.
One fairly common statement is: “This is goal oriented evolution, in real life there is no goal”
This is the first misconception, there is no “goal” in the application.
The “source image” is not the goal, it is the ENVIRONMENT in which the organisms live and try to survive.
It is that environment that determines if a specific individual is fit for reproduction or not.
The same way that all other evolution is also dependant on the environment, e.g. in real life the environment is much more complex and dynamic.
See it as a set of parameters that describes the world where the evolution takes place, the parameters just happen to be in the form of what we humans experience as an image.
It could just as well be a list of parameters such as temperature, gravity, resources, other organisms etc.
Of course _I_ had a goal for the application when I created it, but you have to differentiate between my goal and the goal of evolution.
My goal was to see if a specific problem could be solved, however, the application and the evolutionary process in it knows nothing about nor cares about that problem, that process has no goal, It simply lets the most fit individual reproduce and nothing more.
The second misconception is to see the “generated image” as how the evolved organism would look.
This is not the case, the “generated image” shows the LEVEL OF ADAPTATION of the organism, not it’s physical body.
A generated image that have a high resemblance to the source image simply means that the organism is highly adapted to the environment.
Not that we have an organism that looks like the goal of the simulation, that is an incorrect interpretation of the whole application.
The third misconception is that this can be used as some sort of bat against the ID / creationist movement, this is also wrong.
This is wrong simply because I can never ever write a simulation where I myself is not a parameter in the simulation, so _my_ applications proves nothing on that front.
The only thing we can confirm from this application is that evolutionary approaches can be used to solve tricky problems that we ourselves does not know the solution to if we provide rules and an enviroment in which the problem can be solved.
We can not use this application to confirm or dismiss the presence of God, Thor, Shiva or any other theological being.
well, isn’t the fitness function the definition of the goal? in real life, it’s “live until you reproduce”, in the mona lisa case it’s the difference between the generated and “source” image… in my C++/cairo-port of the application, I had a typo, so that each generated pixel at x,y would be compared to the source pixel x,x — and it resulted in a sunset-like horizontally striped image, because I set the “goal” wrong… (imagine my despair after seeing a very nicely adapted version of this — after some million generations the next morning)
so isn’t the environment designed to specifically fit that goal? (like when they gave the NEAT-robots wheels, finite batteries and food, they wanted their evolution to have the goal of seeking the food and refilling the batteries)
>>isn’t the fitness function the definition of the goal
The fitness function only measures the level of adaption of the individual, regardless of what the environment is.
>>so isn’t the environment designed to specifically fit that goal?
Yes the environment is designed, but the application contains no code what so ever that cares about this.
You have to differentiate what my intentions are and what the application itself knows and cares about.
Very well phrased!
It’s an amusing thing to think the purpose of life would have been encoded in our dna.
I don’t see the point. The goal of this is to render an image, not to teach us anything about biology and evolution, right ?
Another point I have seen throwing around is that in Mona Lisa their is no reproduction. It is true, from my understanding of the code, that their is no cross-over reproduction (2+ parents), but a single entity spawning a copy of itself, with possible mutations, is a form of reproduction. It is how the cells in our body reproduce.
Roger’s idea is also possible with cross-over reproduction, so I think all the arguments saying Roger’s algorithm and idea are not evolving need to brush up on their evolution.
The goal of life is not “live until you reproduce”, it is “live LONG ENOUGH to reproduce”. If you are not fit for the environment, you are not going to achieve that goal, and this is what Roger’s Mona Lisa shows. Candidates are judged by their fitness to the environment (the source picture), and if they are suited enough to the environment, they survive to reproduce.
I was thinking that a nice accompaniment to this post would be to do another competition:
Place some of the organisms that led to the Mona Lisa and have them compete in another, different environment. This would be an experiment in how these organisms adapt. My guess is that less specialized organisms would adapt more successfully to an extreme. Success could be measured in how many generations to a particular distance from the new image.
It would also be interesting to see if a particular pattern emerged in successful which had to survive in a cyclical environment.
@Lynch82: Live long enough to reproduce or do other things that help propagate your genes into the next generation (such as helping your relatives’ offspring live long enough to do same.)
I have to say that I disagree about the intelligent design comment. Indeed, I think this project is a very persuasive antidote against the (often willful) misconceptions of intelligent design proponents. It demonstrates rather elegantly that an undirected process* can lead to a gradual improvement of genomes in the presence of heritable variation conferring differing fitness to their bearers.
Of course, many IDers accept the tenets of natural selection in the context of population genetics. But for those who don’t, your GA is a nice concrete example of how it works and how natural selection is basically an algorithm, putting it more or less on equal footing with a theorem if not a law.
*In the sense that none of your images was designed with the Mona Lisa as a goal.
And if I say that the Environment is the goal of application ? It’s just a semantic confusion.