quote:Writing in Nature, the robot's creators say their experiment shows the ability to reproduce is not unique to biology.
Their long-term plan is to design robots made from hundreds or thousands of identical basic modules.
These could repair themselves if parts fail, reconfigure themselves to better perform the task they have been set, or even to make extra helpers.
So far the robots, if they can be called that, consist of just three or four mobile cubes.
Each unit comes with a small computer code carrying a blueprint for the layout of the robot, electrical contacts to let it communicate with its neighbours, and magnets to let them stick together.
By turning and moving, the cubes can pick up new units, decide where they belong, and stack them alongside each other to make new devices.
In a little more than a minute, a simple three-cube robot can make a copy of itself.
If we can produce "food pellets" for them - some kind of battery they can plug into themselves - shouldn't they start evolving?
posted
I don't think they're really reproducing until they create the little cubes that they are made of.
Posts: 751 | Registered: Apr 2005
| IP: Logged |
posted
Ok, so they need a module to pick up and plug in batteries (which could just be added to the environment to stand in for the sun) and a way to make the little cubes. Anything else?
Posts: 26071 | Registered: Oct 2003
| IP: Logged |
posted
Also, if they only reproduce by making new robots identical to the blueprint they carry, then they won't be evolving. But if they can alter that design as needs arise, then there's potential for nanorobotic evolution, definitely.
--Enigmatic & The Nanites (is a good name for a band)
Posts: 2715 | Registered: Apr 2005
| IP: Logged |
posted
I think electronic evolution is the only way we will ever see a true AI. But I think it is much more likely to happen in a virtual environment, like an evolving computer virus for instance.
One major difference is that you can have over a billion "generations" in a day in a virtual world.
Posts: 5656 | Registered: Oct 1999
| IP: Logged |
posted
I think that being able to make those little cubes is *far* beyond anything we are close to today.
Posts: 751 | Registered: Apr 2005
| IP: Logged |
quote:Also, if they only reproduce by making new robots identical to the blueprint they carry, then they won't be evolving. But if they can alter that design as needs arise, then there's potential for nanorobotic evolution, definitely.
The key to evolution is that it's not designed. The changes are random. The selection of which changes survive is not. As long as the copy procedure screws up every now and then, evolution will happen. Of course, it might kill all the robots if there's not enough of them to start.
quote:I think that being able to make those little cubes is *far* beyond anything we are close to today.
Depends on what they are.
quote:I think electronic evolution is the only way we will ever see a true AI. But I think it is much more likely to happen in a virtual environment, like an evolving computer virus for instance.
One major difference is that you can have over a billion "generations" in a day in a virtual world.
quote: Avida is an auto-adaptive genetic system designed primarily for use as a platform in Digital or Artificial Life research. In lay terms, Avida is a digital world in which simple computer programs mutate and evolve.
Avida allows us to study questions and perform experiments in evolutionalry dynamics and theoretical biology that are intractable in real biological system.
Very cool. I keep meaning to download and play with it.
posted
I did a couple simple Alife projects in my artificial intelligence classes in college. Doing the research I came across Avida (if I remember correctly) but the scope of my project (and time limits) didn't allow for any experimenting. My main AI project was developing a predator/prey ecosystem in a virtual world. Ended up being a lot of fun. I was going to have both species mutate and exchange their behavioral algorithms, but that was way out of the project's scope.
But anyway, Alife is a pet interest of mine, and I would love to do more research. Maybe now that I have more free time...
Posts: 5656 | Registered: Oct 1999
| IP: Logged |
posted
There is no reason in principle why evolution couldn't be guided by the evolving species. Clearly it doesn't happen that way in real life : We have no way of reaching in to mutate our own DNA. But if we had, it would still be evolution - now selecting for the ability to design survivor traits and implement them in genes. (I am reminded of an OSC story, actually - I Put My Blue Genes On.)
However, such a trait would presumably require true AI, which is itself an unsolved problem. A more likely source of evolution would be the random corruption of the robots' memory banks from cosmic radiation, much as happens to biological species. An objection to this is that the robots could easily be designed with multiple memory banks, and take a vote, thus throwing out any corrupted memory locations - basic error checking. Short of providing an infinite number of memory banks, this does not completely prevent mutations - all five, or seven, or n, might get corrupted in the same memory location - but the expected time to happen could be made arbitrarily long, up to and including the lifetime of the Solar System.
Actually, now I think about it, what about the copying process itself? You cannot very well run seven different copying algorithms at the same time, moving a bit only if all seven agree! If a bit was corrupted as it was being moved, you might get mutations more quickly. Indeed, I believe most mutations in biological species are copying errors rather than radiation faults.
Another issue is that one could design the robots to be mutable rather than stable. Suppose one wrote the copying algorithm to introduce random errors in each generation? Of course, this would quickly select for mutated algorithms without that inherent design flaw! Unless, of course, you introduced parts of the 'genome' that mutated, and parts that didn't.
However, I think we are far from seeing this in real robots. Clearly they do not have the ability to mine ore, refine copper and aluminium, make plastic, and put everything together into their own components. A sufficiently wealthy research lab could supply a large amount of cubes, I suppose, but why bother? If you want to study evolution, computer sims are faster and biological species have more real-world applications.
Posts: 10645 | Registered: Jul 2004
| IP: Logged |
quote:There is no reason in principle why evolution couldn't be guided by the evolving species. Clearly it doesn't happen that way in real life : We have no way of reaching in to mutate our own DNA. But if we had, it would still be evolution - now selecting for the ability to design survivor traits and implement them in genes. (I am reminded of an OSC story, actually - I Put My Blue Genes On.)
Fair enough. I guess I really meant that design wasn't required for evolution.
quote:A sufficiently wealthy research lab could supply a large amount of cubes, I suppose, but why bother? If you want to study evolution, computer sims are faster and biological species have more real-world applications.
I know - I actually used a genetic algorithm to do some travelling problems once. It turned out to be unnecessary because I realized I could reduce the number of possibilities to a small enough number to just run them all, but it was cool to develop.
But it's much more fun to watch the robots.
Posts: 26071 | Registered: Oct 2003
| IP: Logged |
Printer-friendly view of this topic
![[Razz]](tongue.gif)
