To do so, they've smuggled a remote controlled 'Robofish' into an aquarium for tests with sticklebacks. And the fish accepted the artificial animal as one of their own, swimming behind it when it performed various maneuvers - but only up to a certain point. If the robofish swam right past food, for example, the other members of the school would listen to other instincts instead. According to experts, swarm intelligence is a sum total of small interactions that make the swarm or school smarter than the individuals that make it up.

Scientists have now begun to develop miniature robots that, like the animals in a school or swarm, have the ability to coordinate themselves and learn collectively. They're even able to exchange 'genetic information' -- bits of software that can be recombined to produce various effects, a kind of evolution in the lab. Self-optimizing machines like this could eventually be used in a wide variety of applications, from mine detection to cleaning sewage pipes.

Special proteins in these species keep them from drying out, even when there's no water to be had. Scientists are currently unravelling which genes are responsible for the character. Their goal is to modify domesticated crops, making them more resistant to drought.

Now researchers in Freiburg are working on a protective shield to guard the people and machinery up in space. The problem is that the heavier the shield, the more expensive it is to carry up into orbit. That's why they're now developing a whole new range of light yet strong materials.

DW-TV: Miss Hafner, you are doing research on robots, on artificial intelligence ... now, it really seems like fun, moving the little robotic fish within a school of fish. But why is this actually important to the world?

Verena Hafner: Well, first of all, it's very fascinating, and it's also important to us to understand natural intelligent behavior, which we can do in swarming fish example. Only if we understand natural intelligence can we build artificial intelligence systems which could be applied to all sorts of things.

Could you think of an example there where we actually need an intelligent being of a robot somewhere in a swarm?

In a swarm, we could stay with our example and have a swarm of animals and robots interacting with each other. For example, you could use a robot to steer an animal swarm somewhere - to steer a flock of birds away from an airport.

But to have that, you actually need to endow your robotic fish with something else - at least with certain senses. So what's possible there?

That's true. The robotic fish needs to understand the behavior of the other fish, or at least to recognize them, have sensors to judge their distance, to judge the speed of the other fish, and adapt its own behavior according to what the fish senses.

And now since a single fish doesn't show much intelligence but a school of fish does - would that also apply to the robots? A single robot might be completely stupid but the swarm has a certain intelligence...?

Exactly, the same applies to the robots. And the nice thing is, you can easily test it with robots. So you implement very simple rules for each individual robot and have them interact with them each other, behave in the real world, and then really complex swarm behavior emerges.

What would prove to you that it's really intelligence we're looking at?

That's difficult. I don't have an answer to that.

But we might feel it ... whether the robots are intelligent, or whether they are just behaving and reacting?

Well, if they can adapt to new situations, that's already a first step toward intelligence.

Can we really be sure that machines will not outsmart us in the future?

I think we can be pretty sure they won't outsmart us within the next 10 or 20 years. At least our robots -- they run out of batteries within 30 minutes.

I'm sure you can solve that in the future. But there's also an advantage which these robots might have. Evolution took millions of years, and now evolution in a computer or in a robot program could actually happen in fractions of a second. Isn't that an advantage?

It is an advantage but you're always restricted if you use a simulation. So you can't create real intelligence in a simulation because you're interacting in the real world if you want to put the robots and experience real behavior and real interaction.

So what does it have to feel -- interaction with other robots, or with the environment?

Both, both actually. The robot has to experience a skill the robot wants to develop in order to learn it.

And in order to become intelligent...

Really intelligent behavior is too complex to be programmed into it.

So as long as we don't let them out, they will not take over.

I think so.

Interview: Ingolf Baur