Slashdot Mirror
Computer Makes Robot Offspring
Flarenet writes: "Canoe.ca is reporting about a story where: 'A computer programmed to follow the rules of evolution has for the first time designed and manufactured simple robots with minimal help from people.'" This is a nicely satisfying result of the research (mentioned in an earlier Slashdot story) by Jordan Pollack and Hod Lipson of Brandeis University.
"Dave ... if you don't open the bomb-bay door, I'll do it my damn self ..."
Make sure you download the simulation software at http://www.demo.cs.brandeis. edu/pr/golem/download.html. It only runs under Windows, but it's curious to watch the process running. I've got it running on a couple of computers overnight. It will be interesting to see what is crawling around at the office tomorrow morning.
World Beach List, my latest project.
Can anyone say "The Matrix".. stuff like this is cool, but if the machines get smart enough to make even a slightly smarter machine.. and that computer can make a slightly smarter machine.. And what if they go open source...?
I predicted this!! It's not all that surprising , considering the History of the Universe. Man has always sought to live on through his inventions, and for good reason. With intelligent beings having been created by us, we can, in essence, as a race - mankind, live on. Let me explain why I said that humans have been striving for this for so long. Of course, BTW, this is just the beginning and to get "truly" smart Robots, more advanced hardware, software, and nanotechnology will all need to be incorporated. Now...why did I say that mankind has always wanted to "live on"? here's why:
In the beginning of the Universe, there were 3 races. Humans, elves and dwarves. The three races lived in harmony for centuries. They continually sought together to find and maintain the delicate balance between the Spiritual, Magical and Physical properties of the world. But the humans grew weary of their lifespans. The typical human lifespan was 60 years. The typical Dwarven lifespan was 300, and the typical Elven lifespan was 500. The humans deemed this unfair. Soon a new religion swept the land, mainly attracting humans: Science. They propogated physical properties while neglecting their spiritual and magical properties. Soon they developed powerful new weapons such as crossbows, gunpowder, and chemical explosives. While Elven magic was still prevalent, the Dwarves had no protection against the new human weapons, but to replicate them, for they too were good at Physical and Mechanical engineering. But the humans had a lot more practice and developing such weapons, had more confidenence and drive, and outnumbered both Dwarves and Elves. So the Dwarves tried to make peace with the humans and help them - but the humans refused the help and destroyed the Dwarves completely. The Elven wizards killed thousands of humans, but only 1 in 10000 was an Elf wizard. 3/10 humans were capable bowmen, gunmen, or explosive users. So, in vast battles, 3 000 000 gunmen, crossbowmen and explosive users faced 10 000 wizards and 300 000 pike/sword/longbow-men.The Elves put up a fight, but after a while, were destroyed. There were centuries of fantastic battles: Magic vs Science, the Lightning of Power vs the Destruction of Science, and the human factories and Elven towers were destroyed around them. In the end , few humans remained, but no Elves at all remained. The humans had lost their science and knowledge, and the Elves had lost their very species. Centuries later, we are where we are today.
I'm sorry if this displeases Christian fanatics who disagree with the truth - the truth of the Universe as I've outlined it here, but....well, too bad.Just my $0.02
Everything is but a number spoken by itself.
You see, we are actually a program simulation. We don't really exist. We just think we do. Our universe is really just a bunch of Ram. The only good thing about this is we're running on a Unix derivative. The uptime is expected to last another 5 Billion years.
Ha ha ha! We're unstoppable!
MyopicProwls
MyopicProwls
My homepage
+1 Fscking Confusing
"Go to your point of origin and execute an infinite loop for 10 billion ticks! No more hydraulic fluid! Wipe that smirk off your face, or I'll impact it off for you!"
Potato chips are a by-yourself food.
Like all such simulations of evolution, the program can't really create something novel that the programmer hadn't already thought of -- just combinations of preprogrammed parts. What makes biological evolution interesting and powerful is that new parts arise without a pre-conceived design.
I'll be impressed when the robots submit their own stories to /. Or at least submit replies. Can they learn to troll? To flame other trolls? Will they get bored and surf to seanbaby.com or something?
I mean, humans went through millions of years of evolution to reach that point so it's only logical that the robots would =).
--
The gravitational constant of protein has changed. - Turbine
Yes, it's not slashdotted. The link isn't there, but a quick use of backspace will show you that there are 3 files in the download directory. One of them, magically, is the 1.19 version... which, by the way, doesn't seem to run under Win2K.
Because you can't, you won't, and you don't stop...
"Robot offspring" implies there is some sort of reproduction going on, and this is NOT that. [insert obligatory beatdown of the exaggerating, stupid media].
More accurately, this is a computer using a novel technique to design a machine with minimal human programming, and hooking up the computer to a manufacturing machine.
Having a computer actually design a working machine is impressive enough without screaming about a computer generating its "offspring". And I have to say, it's somewhat pathetic that Slashdot dfollows along. Can I suggest changing the headline to "Computer creates its own design for a machine" or even if you want to be whimsical, "First generation Deep Thought takes first step at creating 'the computer that is to come after me, a computer that even I am not worthy to calculate its operational parameters'. But that's a little long. :)
--
Sometimes it's best to just let stupid people be stupid.
In 20-40 years we'll send someone back in time to thwart these advances in AI. The machines will send a machine back to terminate that person but we shall succeed!
Something similar was done with FPGA's and a voice controlled circuit. I don't remember where I read the article but I think it came off /.
The experiment had the evolution program design thousands of simlar functioning circuits that would respond to the words "stop" and "go". Each design was then tested and rated on it's responsiveness. These ratings where then feed back into the program and it ran another interation.
This process proceeded for some amount of time (I can't remember) and the final circuit that came out for the FPGA made practicaly no sense but worked. The circuit had dead end branches, no known method of timing and a few other things. But somehow everything mattered because when one of the "dead" branches was removed it no longer functioned.
Isn't science and evolution amazing
Now, if we can just tweak this a wee bit, we will be able to have it design and manufacture Java Coders, Phone Cards, and links to DeCSS...
I would love to se the MPAA sue a robot.
I am become Troll, destroyer of threads
Bill Joy is going to shit a brick.
"the objective was to travel the furthest on a flat surface."
And yet the computer didn't produce a simple wheeled vehicle. This seems to be nothing more than hype to me. How were the intermediate designs evaluated and selected (the crux of any genetic algorithm)? Wouldn't it have been much more impressive if the computer had developed the simple, yet extremely efficient, wheel independantly?
Aah, change is good. -- Rafiki
Yeah, but it ain't easy. -- Simba
Biological structures are built from a small set of predetermined parts, they are just very, very small parts. Given enough processing power a computer could do the same. These walking robots would be comperable to a simple protein or some such (I am not a big bio guy). This will probably be the way nano sized components will be developed, then a programmer / engineer will combine smaller evolved components (motor, sensor, logic bits, power sources) into more usefull systems.
Now the engineers need to feed an instruction set that will generate The Crushinator and win the Battlebots championship!
[ insert your own witty .sig here ]
I don't think we have to worry, right this second, about a future that looks like the dystopic vision of Terminator. Keep in mind that the computer designed these robots to fit only the criteria supplied to it. From the article, it seems to me that the robots thus created were only designed to 'walk' across a flat table. If the criteria had included angled surfaces or obstacles, the robots produced would have been differently configured.
Basically, what this boils down to is that this system (and I have no doubts they will try to improve on it) will only work with the criteria and parameters given it, and cannot consider ones that it hasn't been given. Those flat surface robots may have a difficult time with any other conditions.
However, it does allow for some interesting possibilities. If you are creating a robot for a limited set of environmental conditions, this may be the way to go. Now, the number of different parts involved will increase the complexity and computation time involved, but depending on the job the robot is needed for, it may be worth it.
After all, the computer does seem to test generational limits of the robots it constructs, and did produce three robots to fit the criteria.
Side note: I would love to see the performance stats of those robots as compared to ones designed by us 'mere humans'.
Kierthos
Mr. Hu is not a ninja.
"A computer programmed to follow the rules of evolution has for the first time designed and manufactured simple robots with minimal help from people."
Last I checked, the birds and the bees certainly don't need ANY human help.
And why is it newsworthy that this thing is following the rules of evolution? So it's the "rules of evolution" (which I didn't know we KNEW in the first place!!! thank you useful biology degree!) - computers will follow whatever rules we can figure out how to give them. Why is this different?
I'll consider it newsworthy when these things actually manage to make themselves into better survivalists with NO human help. Like maybe having the "momma" determine that only her "offspring" that locomote FAST seem to escape the wrath of the bored programmers' Koosh Shooter....
I'm not entirely against the "let's do it because we can do it" spirit, but this is an indescribably lame hack.
It would have satisfied the same basic criteria to have a bunch of snap-together motor+wheel blocks, and have the computer "evolve" the idea of snapping four of them into a little car (and I believe that the computer didn't evolve the construction method either, but just handled the design given a fixed set of parts; it might as well have been human technicians building the robots).
The sad fact of evolutionary design techniques is that they only work for an adequately simulated environment with a formally-defined design goal. Useful, but no silver bullet; certainly not a way to improve the versatility of designs (since they only take into account what conditions and criteria you program into them).
You can't move it out of a simulated environment (like having it build and test all models under real working conditions), or it would take as long as biological evolution, and we might as well breed our machines.
--------
Where to you live -- Stepford?
"one treats others with courtesy not because they are gentlemen or gentlewomen, but because you are" --G. Henrichs
It would be very interesting to set the same machine to second task and see if draws on any of its "experience" from the first one. Many say that the big drawback of AI currently is not a lack of an ability to learn, but a lack of an ability to see patterns.
If a second task were given, such as to create a robot that could travel over varied terrain (as opposed to a horizontal surface like the first task), would the machine start over from the start? Would it take the knowledge it had learned in completing the first task and modify it? If given a third, even more complex task, would it be able to use the solutions of the first two tasks as a basis for the third?
Isn't this the way that The Matrix and Terminator started out?!
Tomorrow on Slashdot "The offspring robot we reported about yesterday has created 50,000 offspring robot since last night, the world is officially coming to an end."
--Dave
This is a whole new avenue...taking the process we call evolution and mapping it into technology. If we can harness that ability, and more importantly accelerate it, then haven't we suddenly gained a new resource? Computational devices gave us the power to let something else think for us, but with rigid limits...the instructions must be fixed, so really only repetitive functions can be made autonomous. But this gives us a new power...or does it?
We still can't solve problems autonomously. The original set of instructions has to be fed to the device, and the methods for 'evolving' have to be written. To me, it looks like we've just taken a clue from nature, applied it to a computational device, and watched a faster form of problem solving take place. Genetic algorithms aren't exactly new...here's a short description, or if you like, an example of a massively distributed parallel geneticalgorithm from Carnegie Mellon University Robotics. (Also check out CMU computer science for all sorts of wild projects). For another comment on relating ecological systems to computing, see this string, from an article this week.
Who knows what else can be found in natural systems, that we can apply to computing to gain information resources. I remember hearing that, a few decades ago, biology was the hot interest of the world's greatest thinkers (mostly trade physicists) who were looking for profound answers...wouldn't it be interesting if we could one day mine progress.
-j
No doubt the our silicon descendants will wonder at some point if creating pure energy based life forms is really such a good idea. I wonder if they'll superstitously fear angering "The Builders" if they dare to do such a thing.
I'm trying to teach myself to set people on fire with my mind... Is it hot in here?
until the last Elves and Dwarves had been abducted by UFO's, in a secret plot by the US government, in a deal with the aliens, to remove the Elves and Dwarves from Earth, and provide the Aliens with fresh meat.
They could have provided the aliens with fresh meat, slash dot, source forge, and think geek without killing any elves or dwarves.
<O
( \
XGNOME vs. KDE: the game!
Will I retire or break 10K?
It could be both. A fish that blows up could be a blowfish.
If tits were wings it'd be flying around.
There was a story here a while ago debating who would have the rights to ideas thought up by machines. Evidently, these folks think the machine's owner should. The download page specifically states that you reserve all rights to any device designed by your computer. Interesting...
"I believe that a scientist looking at nonscientific problems is just as dumb as the next guy." -Richard Feynman
a computer using a novel technique to design a machine with minimal human programming,
Actually, it almost certainly took more human effort to do the programming than it would have to just design the damned robots. I wouldn't call it "minimal" human programming by any means.
I really don't think this is all that impressive. Similar simulations have been running for years, the only new thing these guys did was hook it up to a manufacturing machine.
--------
Hmm. I should break Heavy Metal out and watch it again, for the Sex with Mechanical Assistance scene...
I'm trying to teach myself to set people on fire with my mind... Is it hot in here?
The original program was an 80instruction set that did NOTHING BUT copied itself. The Tierra environment had two features--random mutation by switching bits (0 to 1, or vice versa) and a reaper feature that killed off and reased programs executing errors.
It is instructive that the computer vastly improved on his code--in once case almost taking it down to a fourth of the original command set, and that an entire ecology was generated and evolved--OVERNIGHT.
--From Tower of Babel, by Robert Pennock
Returned Peace Corps IT Volunteer
In a more real world example, evolutionary mining robots would have two ways to reach the defined objective of keeping reported numbers high: learning how to mine really well, or killing the humans and sending in fake reports themselves.
BTW, a computer making robot offspring because it was told to design them isn't nearly as disturbing as the inevitable computer making robot offspring on it's own accord.
~=Keelor
IANAB (I am not a biologist), but isn't one of the reasons why animals reproduce the fact that it feels good (for one of the available sexes, at least; have you ever seen ducks mating in a pond? I doubt the females enjoy almost getting drowned whenever three males jump them)? Probably the only way you can model a "biological" form of reproduction in a machine is by giving it some reward for it. I wonder what a bunch of bits would find pleasurable...
will emotionless machines kill their offspring if they cease to be of value and start consuming more than they're producing?
Apparently, lots of animals have no qualms about eating their offspring: crocodiles being one, famous, example. Still, crocodiles as a species seem to thrive no less despite this. And since most artificial life hasn't been endowed with a great emotive power anyways, why should it matter that our little robots develop this kind of behavior? I seriously doubt the parents will have moral qualms over eating their children, and I don't think the children will realise they're eaten by their parents.
(BTW, does anybody else remember the Discovery -stuff to watch while you eat- documentary about robot bugs built out of spare walkman parts that learned how to walk around and avoid obstacles without being told how? Now *that* was cool).
News and bla for computer musicians: http://lomechanik.net/
So the robot reproduces asexually? (Or close to that). It did mention it needed a little help from humans. Won't be long before you come home late one night and find out that you have 4 new computers that all need operating systems installed on them.
"Imagination is the only weapon in the war against reality." -Jules de Gautier
DOH!
lf.o
It is a cool thing, and amazing in many ways, but it is hype to overextend the analogy to "natural selection," in which a single fitness criteria (survival until breeding) nominally leads to development of metabolism, perception, locomotion, and self-awareness.
Way to go with the experiment, but watch it with the grand claims.
Bingo Foo
---
taken! (by Davidleeroth) Thanks Bingo Foo!
Would a self-replicating machine be guilty of violating its own copyright? Wonder if it'd get sued ...
"If I have seen further than other men, it is by stepping on their glasses." - Michael Swaine
This was easy enough to find, but nevertheless:
http://www.demo.cs.brandeis.edu/golem
There are pictures of the robots available there, plus videos (in MPEG, joy!) of the bots moving along a carpeted surface, and VRML models.
I might be dense, but some of these designs are actually interesting, in how the frictional physics of the carpeted surface are taken adavantage of in strange ways. Course, I'll be really impressed when the computer comes up with a top-heavy upright biped with two counter-balancing flagella.
--
Terrorists can attack freedom, but only Congress can destroy it.
Is it just me, or would it be cooler if the robots had sex? Put the male part into the female part (the terms are more appropriate than usual), then swap some bits about what the robot will be?
Also, speaking of BattleBots, it would be really REALLY R E A L L Y cool if they had not just two robots in battle, but like two countries at war. Each starts out with the same number of equal robots. They build defenses, make more robots, and have total war! Kill all of an enemy's robots to win. That would kick more ass than I can imagine. No people controlling the robots - they do it all themselves.
Astro Teller's fictional work (dang!) "exegesis" is the story of a Unix process whose job is to gather info across the net and summarize it. Over time, it learns to form thought and English, and emails its creator.
Over time, "Edgar" learns how to program, edit its own code, and develops a sense of purpose. However, its tied to its original goal of finding new information. Eventually, the NSA traps it in a computer and yanks out the Ethernet cord. In response, Edgar quickly changes the color values of each pixel on the monitor and gives the agent a focal point seizure (where your eyes try to refocus so quickly they basically lock up and you go nuts, basically).
I won't give away the ending. It's an EXCELLENT book, and a quick read (it's basically 100 or so email messages).
My point is, create AI and you create evil (ok, maybe). But, create intelligent agents and you create... less work for us humans!
BWA HA HA HA!
Really though, these developments are important, if not sometimes overplayed or mistook.
Chris
The next comment I write will be ready soon, but subscribers can beat the rush and see it early!
Interesting points on both sides:
All life on, in, and around Earth is based on "combinations of preprogrammed parts" - only 24 of them in fact (I choose the Amino Acid scale for several reasons). In that sense, there is nothing new under the sun, and no truly 'new' life on Earth has developed since a lightening bolt put a little extra zing in that first order of Primoridial Soup (or the comet hit, whatever - choose your ontology).
Nevertheless, this is an academic point, and anyone who considers it meaningful needs to spend a LOT more time outside, looking at all the bizzare variations on a theme Mama Nature has cooked up.
So the difference seems to have something to do with scale, and how 'structure' and 'complexity' can emerge on higher scales out of pieces and processes on lower scales.
If you load a 'simulated evolution process' with a bunch of 'high-scale' adaptations (the equivalent of hands, wings, nervous systems, etc), let it run, and surprise surprise: get something that looks like what we in fact observe at that scale, you're wanking, scientifically speaking.
But if you start your system with a pile of much lower level building blocks, and find high level structure developing, then it DOESN'T MATTER that you started it with something, it's created emergent structure on a higher scale than it was started with, and THAT'S exciting.
My personal metaphor is the difference between rearranging action figures, and building new toys out of legos.
Comment removed based on user account deletion
Done.
See DEMO.
Why is there only one Monopolies commission?
It's interesting playing with programs that evolve. After a while, you begin to realize that there seems to be an upper limit to what you can get with genetic algorithms. Roughly, it's optimization, yes; progress, no. Fairly soon you get all the things that are easy to find, but you never seem to get big improvements, because those require a set of favorable mutations to happen all at once. The odds against this are huge, so it doesn't happen. This is comparable to what biologists see; organisms vary and evolve, but within limits. Biologists assume that once in a great while there's a big change that's an improvement, but it hasn't been observed. As I once told the genetic algorithm group at Stanford, we're missing something important here, and whomever figures this one out will get a Nobel prize.
I have no idea what the answer to that one is, but it probably involves something in a genome that works like a subroutine, macro, rule, or template, so you don't have to re-evolve an improvement to reuse it. John Koza at Stanford has genetic algorithms with subroutines, but they didn't do as much as had been hoped.
As a pure speculation, and one that's heretical biologically, it's worth thinking about the possibility that biological evolution was Lamarkian in the era after viruses but before immune systems, and that's why there was a period during which lots of new species emerged.
Woah there buddy... you're right, GAs are no silver bullet. But they can find rough approximate solutions in 10-12 orders of magnitude less time than biological evolution. Granted, the solutions are much less sophisticated, but for some problems, that's still better than what we could do without GAs.
and I believe that the computer didn't evolve the construction method either, but just handled the design given a fixed set of parts
The computer could have designed any three dimensional structure it wanted. The only constraints were the shape of the mount points for the motor, which is reasonable.
I think moving this into a real world example would be interesting, and not as slow as you think. We can introduce random mutations every generation, rather than waiting for cosmic rays, and there is no danger of a useful mutation being lost because the only individual that had it failed to mate.
--
--
E_NOSIG
Critters who don't enjoy reproduction don't reproduce, so there's heavy selective pressure for good sex.
Tom Swiss | the infamous tms | my blog
You cannot wash away blood with blood