Morphogenesis: Shaping Swarms of Intelligent Robots

[metabog]

They probably won't become self aware and decide to kill everything at all!

[CertainAsTheSun]

Thanks for putting this video up. I tried to access the video on your website, but it was difficult to download or play the video there. I'm a student and I'm going to share this with my Bio class :]

[speedproductions797]

morphogenesis is the key!

[doceigen]

A better example of 'swarm intelligence' can be had if you watch people cram into an skyscraper's elevator after lunch.

[michalchik]

Bees and ants are not in any sense simple. Their DNA is as complex as ours and they are capable of feats of navigation, pattern recognition, dexterity and communication that no robot can com close to. Their complex social behaviors are not a result of "swarm intelligences" but complex behaviors encoded in their DNA. These guys need to learn more about insects before they can claim they are emulating them. If "swarm intelligence" exists this is not it.

[Houshalter]

@michalchik, DNA codes for RNA, RNA codes for protien, protien forms neurons, neurons form different structures, and these different structures form the complex behaviour you see. All we want is the end result, so its really unnessasary to examine the entire DNA and figure out how it works. Most of its junk anyways. Swarm intellegence is just a few simple rules resulting in complex behaviour, something easy enough to do, but the more advance behaviour you want, the more complex it gets.

[michalchik]

No swarm in intelligence is wishful thinking based on the fallacious assumption that there are a few simple rules that govern the complex behavior of natural swarms. If there are a few simple rules, someone should have been able to write them down by now.

Billions of basepairs of DNA are coded into 10's of thousands of genes, which are turned into hundreds of thousands of types of mRNA and snRNA, that produce 10's of thousand of proteins that interact in countless complex ways.

[Houshalter]

Yes its complicated but thats because its had millions of years to evolve. If your talking about ants countless people HAVE already come up with a set of rules. Just look up ant simulations.

[michalchik]

The simulations are very simplified models that are not empirically validated. My exact point is that ant behavior is the product of 100's of millions of years of evolution on a global scale, not the "emergence of complex intelligent behaviors from a few simple rules" That is a red herring. We are still discovering new pheromones, and acoustic signals, and all these signals are incorporated into different behaviors by context. The rules are very, very complex and finely tuned in ways we don't

[Houshalter]

@michalchik, Millions of years of evolution has given ants some added complexity, but if you look at the simulations you'll see they're close to real ant behavior. I used to work with a simulator called "darwinbots". It demonstrates both swarm behavior, but also evolution through RMNS.

[michalchik]

I am not saying your research isn't interesting or valuable. What I object to is the phrase "you'll see they're close to real ant behavior." By what metric? I know of barometers, to measure pressure, voltmeters to measure voltage. There is not antometer to measure how antish behavior is nor close to the natural world your simulated environment is. Why? Because those things are incredibly complex and no one has any meaningful way of reducing them to a set of measurable parameters.

[Houshalter]

@michalchik, That would be cool to have an "antometer", but no, im not a researcher, and these simulations are really just for observing any kind of emergent and evolving behaviours. So far no ones created a successful "antbot" although a few good programmers came close. The simulatoin actually works against ants because its essentially a discreet flat plane. The only complexity in the enviroment comes from interaction with other organisms, just like a real ecosystem.

[michalchik]

It may look antlike to a casual observer the same way a a wax figure may look human to a casual observer, but that is only because you look at both in a controlled simplified environment in a casual way. Ant's cope with widely fluctuating temperatures, soil chemistry, poisonous plants, individual and colony predators, contagious diseases, uneven and collapsing terrain, photodegradation of pheromones, wind, mechanical trail disruption, irregular size food, food the fights back and on (continued)

[Houshalter]

@michalchik, look, im not saying ant behaviour isn't complex. I used to stare at them all the time and watch them run around. They're very complex, and often inefficient at actually searching for food and then getting other ants to come look for it to. I got a fly swatter out and swated a bunch of flys that were bothering me. I gave the ones i got to the ants. You wouldn't believe how visous these little guys are to dead and dying flies. They carried dozens of them down in seconds.

[michalchik]

There are very good reasons why we have computers and robots that play chess well or work on assembly lines, but we don't have robots that do tasks that average humans can do such as build roads, drive trucks, demolish houses, etc... Chess and factory floors are minimal environments that are highly regular and predictable. In the real world things are not. Anything human or ant without huge repertoire of complex behaviors gets stepped on.

[Houshalter]

Ants might be complex, actually they are, but theres a limit. There are only a few thousand neurons in their brain, much less than even other insects. They have been to focus of research because they are consistent, easy to observe, and do often seem to follow a few, albiet complex, set of rules. Look at it this way; their behaviour is more complex than say a plant or a dust mite, but far less complex than a large vertabrate, like a fish or human.

[michalchik]

This technology is interesting and probably will be important but "swarm intelligence" is a complete misnomer.

We have known for a long time that complex behaviors can be executed in a series of simple steps. That is what an algorithm is. What is key is how those steps are organized and what rules are used. That is where intelligence comes in and all the intelligence in this case is coming form the programmers looking at individual problems and deciding on rules.

Cooperative robotics.

[absher786]

@michalchik

Michael, I read yours Houshalter's comments and found the conversation very interesting. For some points, I'd agree with you, while agree with Houshalter for the other ideas.

Basically, if programmers have been designing individual algos based on individual cases, then there is no doubt that all their effort is baseless. To devise the rules by which it works, it is essential for every researcher to look at as many scenarios as possible.

[absher786]

@michalchik

Moreover, I'd also like to point to the fact that Data Mining is a kind of technology in Computing whereas we try to solve the same kinds of problems. I guess, the "Rules" for the Intelligence of a Swarm could be determined using this technology, if we devise a way to capture these natural swarm behaviours and represent them as Data against a period of time.

[dabeshu]

lol i had to figure this shit out the hard way. ive been thinking about this on a constant basic, with no outside information, for about 5 months now. its an obsession

[alain314]

Very interesting. Your examples focuse on action, but do you also study information acquisition and transmission? I try to figure out the perspectives of a 'swarm-distributed' knowledge. Not far from our human culture I guess. Please, post some references if you have. Thanks a lot.

[cartoon806]

Whoooaaah!

[N2KC]

This is a great video.

[DeimosSaturn]

Doesn't it scare you guys that swarm nano bots could rapidly evolve and emerge as a singularity event that causes the extinction of all life that does not serve the bots? What about human swarm intelligence? I much prefer a life as a Borg assimilating life rather than totally being extinct.

[aptennap]

That's fiction, and will never happen.

[jman301747]

dude seriously ...go get laid