Slashdot Mirror
Digital Life and Evolution
mrivorey writes "Discover Magazine has a story about The Digital Evolution Lab at Michigan State University. Scientists there have created virus-like computer programs that replicate, mutate randomly, and compete with each other... in other words, they evolve. Among such feats as learning to add and compare numbers, these digital life forms also once avoided scientists attempts at "killing" them, by playing dead.
You can download the project yourself from SourceForge." We first mentioned this in early 2003, but it appears to have developed a good deal since then.
The only interesting part that caught my attention is:
"One of the biggest questions in evolution is, why aren't all organisms asexual?" says Adami. Given the obvious inefficiency of sex, evolutionary biologists suspect that it must confer some powerful advantage that makes it so common. But they have yet to come to a consensus about what that advantage is.
I think this built-in inefficiency is to control the population, no? So it's important to introduce the idea of "mating" to virus/robots to keep them under control.
500,000 slashdotters hitting refresh constant-simultaneously is probably still tolerable, how about 4,000,000?
Oh wait... I guess I'm confused between inefficiency and deficiency now.
Rock that crushes, Paper & Scissors that don't matter.
Dan Simmons included this idea in his Hyperion book series, where evolving digital life spead into the "infosphere" and became artifically intelligent. Later it tried to exploit the human race and wipe out large portions of it. People who download the project beware!
Philosophy.
I wonder how long until the first virus based on this code is released?
I'd like to see this run as a distributed computing project, as a sort of race to achieve measurable consciousnessness among the organisms.
Art Schools Dietzilla
The only sentience that humans have experience with is our own, and I think it is safe to conclude that most scientists working on AI projects would try to replicate human sentience either intentionally or unintentially. Human beings have a very, very robust survival instinct and are extremely destructive when threatened. Do we really want to take the risk that we will create an AI that has our suvival instinct as well as a human-style thought process?
I have caught flak for it in the past, but I have argued for a constitutional amendment banning the U.S. military from employing robotic combat units as anything more than a small minority of our combat forces. The last thing we need is either a weak AI or strong AI being used as the basis for taking over our military and then taking over our country. That's always seemed to be Hollywood's greatest feare. He who controls the AI controls the nation. From Terminator to the Matrix, the dark side of AI has been presented, but how many people don't take it seriously because it's "just a movie?"
I have no problem with limited AI research, but I'll be the first to admit that I am something of a technophobe when it comes to AI. It's simply because of the fact that what we are doing is a playing God with a type of intelligence that is quite suitable for quickly taking total control over our civilization. It makes as much sense to me as putting our worst enemy in charge of our national defense in exchange for a nice chunk of change every month.
This is the classical arrogance. We think that we can control another intelligent being. If we can't control a third world nation that can't possibly wage a real war against us without being obliterated from the face of God's creation within literally a few days if we tried hard, then how can we control a mechanical intelligence that can adapt and grow and potentially learn how to control everything from Wall Street to our strategic defense?
The reason that T3 was so scary to me was that it was the ultimate combination of a rogue AI and grid computing. The only way to stop that new version of skynet would be a scorched Earth policy on our entire electrical grid to power off every node.
And lastly, how on Earth do we expect to negotiate with a hostile AI? What could we possibly offer it except absolute fealty? It has no sensual desires, no use for wealth, only perhaps power over other intellects.
Click here or a puppy gets stomped!
I just started using this thing, and all of a sudden I heard a quiet "Move zig..." over the speakers....
While the "digital life" models may be helpful in visualising what's going on in real life, and in devising experiments to test real life with, the digital environment is about as artificial as it gets.
That said, what the models are showing is that sexual reproduction accumulates changes faster, but does not change the quality of what accumulates. The next step will be to tweak the models even further from reality in order to see them accumulate more advantages than handicaps. Otherwise the results are too depressing.
In analogue life (ironic that digital life should be an analogue of analogue life), genuinely advantageous mutations are collectors items - or would be.
Got time? Spend some of it coding or testing
I'm probably sounding a little too sci-fi here but how controlled is this "controlled" envivonment. Given enough time can't it evolve past it?
This reminds me of the "Progranisms" project I saw over on the Gentoo Linux forums:
i ght-progranism.html
r anism-neilh.c r anism-neilh-condensed.c
http://forums.gentoo.org/viewtopic-t-255505-highl
http://www.progranism.com/
Basically some guy put together an executable which makes a few (mutated) copies of itself when it runs, then executes those copies after a short delay. The idea is that executables might evolve which show interesting behaviors.
You can download his source code here:
http://www.progranism.com/junk/progranism-2.3.1.c
Because I like doing strange things, I made a variant of the program which mutates the source code and recompiles it (mutating until it gets something compilable), rather than mutating the executable directly:
http://www.its.caltech.edu/~neilh/progranism/prog
http://www.its.caltech.edu/~neilh/progranism/prog
http://www.its.caltech.edu/~neilh/progranism/ (some cleanup and maintenance scripts)
Unfortunately, it's stuck in a pretty steep local minima -- it makes some trivial mutations, but nothing major. One interesting possibility would be to have it search your hard drive for other executables and source files, and try to "mate" with those.
Another scary possibility would be to have viruses/worms with non-trivial evolution capabilities. That'd be a pretty nasty outbreak to try to control.
Finally, a rather neat-looking project is AI.Planet, which is trying to create an 3D evolving ecosystem/world of intelligent "organisms." Framsticks, a 3D life simulation project, is also pretty cool.
Simple definition usually is that if two organisms can't produce viable, fertile offspring then they're of different species.
Even in sexual organisms, this definition of species has some problems. You can easily have organism A that can breed with B, and B can breed with C, but A and C are incompatable. How do your divide up the species lines in this case?
In general, when a new species forms, each organism has to have others it can mate with, or else they would just die out without any offspring. The speciation process is a gradual one, and so, theoretically, there is probably a path you could follow between any two sexual organisms where any pair on the path could theoretically mate.
In Avida, for simplicity, we determine species by testing each orgasnism against the species of its parent. If it can cross-over at most points with the prototype of that species, it is marked as being part of that same species. If it cannot, we create a new species for it where it is the prototype. Not an ideal method, but it works in most cases (and we rarely need to resort to the species concept).
What's fun, is that this even works for asexual organisms. We can force all possible crossovers (in isolation of course -- this never feeds back into the system) to see if they would have any ability to mate if they has been sexual.
Dr. Charles Ofria
Director, MSU Digital Evolution Lab
ie. The Avida organisms would evolve not as i386 organisms, but as Avida organisms that are rewarded for producing i386 code that gains them more CPU/Memory time/space to reproduce.
I guess one nice thing about open source software is that even those who disagree with you can help you. :)
From the article:
When the Avida team published their first results on the evolution of complexity in 2003, they were inundated with e-mails from creationists. Their work hit a nerve in the antievolution movement and hit it hard. A popular claim of creationists is that life shows signs of intelligent design, especially in its complexity. They argue that complex things could never have evolved, because they don't work unless all their parts are in place. But as Adami points out, if creationists were right, then Avida wouldn't be able to produce complex digital organisms. A digital organism may use 19 or more simple routines in order to carry out the equals operation. If you delete any of the routines, it can't do the job. "What we show is that there are irreducibly complex things and they can evolve," says Adami.
The Avida team makes their software freely available on the Internet, and creationists have downloaded it over and over again in hopes of finding a fatal flaw. While they've uncovered a few minor glitches, Ofria says they have yet to find anything serious. "We literally have an army of thousands of unpaid bug testers," he says. "What more could you want?"
> On the face of it, it would seem to provide some evidence for the Intelligent Design crowd.
No, it doesn't even provide evidence that biological evolution doesn't work, because it's just a simulator. A "failed" simulation hardle proves that the real thing doesn't work. Especially when the simulator doesn't even try to be a detailed model of the real thing.
And even if it did provide evidence that biological evolution doesn't work, that would not be evidence of an intelligent designer.
Sheesh, evil *and* a jerk. -- Jade
"these digital life forms also once avoided scientists attempts at "killing" them"
Imagine the video games that could come out of this?
Sorry, but geologists rationally took apart creationism 200 years ago.
I've said it before and I'll say it again - the computer you're using, the chair in which you sit, the glass from which you drink all had an intelligent designer. What makes the planet and the universe different? To be quite frank, I think the chances of so many different species of life forming on one planet from some primordial soup is pretty far out there. I think it takes more faith to believe in the (ever changing) beliefs of science than to believe in the Bible.
When millions disappear from earth, it's not aliens, it's the rapture.
Doesn't this prove that evolution exists? I'm not talking about the theory that WE evolved, but the fact that things evolve?
Most people don't distingush the 2 concepts, and they just say that they don't believe in evolution, but they don't really know what evolution is, and that there are multiple types.
I like to show that our evolution is very possible based on the fact that things DO evolve, and it would be unlikely that we are an exception to that process.
This is no news in the AI community, algorithms such as GA's long ago learned how to add numbers, etc. I won't even go into detail here as I assume most readers are aware of this
This story is merely a case of someone who is excited about their work explaining it to an author who doesn't know as much about the subject matter. The author then turns around and writes a story for the lay-person who is not versed in the field. These people in turn jump to humorous conclusions.
This is a common occurance in magazines such as Discover and Popular Science, as much as I enjoy them. A good example is stories on robots, such as Honda's ASIMO. People see ASIMO do amazing things and assume that in 10-15 years we will have these robots in our homes. What the articles often fail to mention is that while ASIMO can do complex tasks, it has very limited ability to recognize a situation, such as a staircase in front of it, and decide on a course of action to take, such as executing its stair climbing procedure.
The true point of the article is that AI algorithms can teach us things about evolution. To make grand jumps and assume that these programs are even in the same playing field as SkyNet or the Matrix is to miss the main point.
As I said above, this is merely the case of a complex subject being explained in a way that is easy to digest for the masses. Even someone who had only taken a few graduate AI courses would find that many misguided statements are made in the article.
AFAICT, at this point the system is treating food (numbers that can be added) and code (the instructions the organisms are made of) as distinct kinds of matter. How about instead of just feeding numbers into the system, postulate that code, food and maybe processor time (energy?) can be traded/transformed into each other, and are conserved at some level - e.g. a "dead" organism can serve as a food source for another.
We could see the emergence of new behavioral patterns - predators, carrion eaters, parasites, and God knows what else.
We fetch your mail, we route your packets, we guard you while you surf. Don't fuck with us.