Altruistic robots produced through evolution

[Iason Ouabache]

http://www.tgdaily.com/general-sciences-features/48222-altruistic-robots-produced-through-evolution

A Swiss team has applied Darwinian selection to robot development, producing robots that can walk, cooperate and even hunt each other.

"Just a few hundred generations of selection are sufficient to allow robots to evolve collision-free movement, homing, sophisticated predator versus prey strategies, coadaptation of brains and bodies, cooperation, and even altruism," say the Ecole Polytechnique Fédérale de Lausanne and University of Lausanne researchers.

"In all cases this occurred via selection in robots controlled by a simple neural network, which mutated randomly."

The input neurons of the neural network were activated by the robot's sensors, and the output neurons controlled its motors.

Each robot had a different 'genome', describing different connections between neurons. This resulted in unique behaviour and fitness - how fast and straight it moved, for example, or how often it collided with obstacles.

At the beginning, the robots had random values for their genes, leading to completely random behaviours.

But Darwinian selection was then imitated, by choosing the genomes of the robots with the highest fitness to produce the next generation.

To do this, genomes were paired, and random mutations such as character substitution, insertion, deletion, or duplication were applied.

The team found that within 100 generations, robots were able to move without collisions in a maze.

When 'breeding' for predator behaviour, a range of strategies evolved, including lying in wait and circling the walls.

And, astonishingly, the robots were even able to evolve altruistic behaviour, in a task that involved pushing tokens around. Some could be pushed single-handed, earning the robot one 'fitness point'; others required two robots, gaining the whole group one point.

It was found that groups of unrelated robots - those with randomly differing genomes - invariably took the selfish approach and went for the small tokens. But those with similar genomes generally pushed the larger tokens, cooperating to raise the fitness of the whole group - and thus reducing their own chances of 'winning'.

"These examples of experimental evolution with robots verify the power of evolution by mutation, recombination, and natural selection," conclude the authors.

"The ability of robots to orientate, escape predators, and even cooperate is particularly remarkable given that they had deliberately simple genotypes directly mapped into the connection weights of neural networks comprising only a few dozen neurons."

[Jasper]

That is the coolest motherfucking thing ever.  Holy hot dicks from hell.

[Jasper]

What needs to happen next is evolving robot programs for specific purposes. 

This is damn cool.  I mentioned that, right?

I want to research the evolutionary pressures that helped us evolve consciousness and apply those to the brain breeding process. 

Oh damn.

I want this. 

[Triple Zero]

evolutionary programming is balls.

awesome balls.

(meaning, really good)

[LMNO]

AWESOME.  BALLS.

[Captain Utopia]

And, astonishingly, the robots were even able to evolve altruistic behaviour, in a task that involved pushing tokens around. Some could be pushed single-handed, earning the robot one 'fitness point'; others required two robots, gaining the whole group one point.

I get confused where the line between mutually beneficial and altruistic is drawn at the species level.  At a genetic level, isn't it all selfish?

For example, I think a better test for altruism would be if helping another bot (for simplicity, of the same species - not a widely differing genome) push their token didn't necessarily grant a direct benefit to the individual who was helping.

[Triple Zero]

AWESOME.  BALLS.

Yes. I should write some evolutionary programming things.

Maybe this is what I should do with all this reading abotu SVG I've been doing lately.

[Reginald Ret]

And, astonishingly, the robots were even able to evolve altruistic behaviour, in a task that involved pushing tokens around. Some could be pushed single-handed, earning the robot one 'fitness point'; others required two robots, gaining the whole group one point.

I get confused where the line between mutually beneficial and altruistic is drawn at the species level.  At a genetic level, isn't it all selfish?

For example, I think a better test for altruism would be if helping another bot (for simplicity, of the same species - not a widely differing genome) push their token didn't necessarily grant a direct benefit to the individual who was helping.

example is not altruism.
example is selfish.
robot that tries to push 2robot tokens gets more points than if the same robot didn't try to push 2robot token.
if robot benefits, robot not altruistic.
indirect benefits count.
it would be altruism if the robot did not benefit, but other robots did.
evolutionary pressure can't create altruism. because pressure only exists if there is benefit.

for the rest:
AWESOME. BALLS.



PS i know people will be difficult about this 'altruism can't evolve' point so let me quote a respected expert:

indirect benefits count.

[Jasper]

Doesn't make this any less awesome.  Who cares if it's not really real altruism?  The point is that these guys have figured out how to breed robotic AI.

[Kai]

I thought this was pretty cool, until I read the journal article and realized the title said "Darwinian Selection" aka Natural Selection. What's going on in this is artificial selection. In other words, the robots aren't selected by environment, or predation, or actual fitness (ie ability to produce offspring). Humans do the selecting, discarding those with the "worst performance".

[Jasper]

Yeah, it's closer to animal husbandry.  Again though, are labels a big deal?  This is a great idea.

[Reginald Ret]

Let me quote a respected expert:

AWESOME. BALLS.

[Captain Utopia]

Yeah, it's closer to animal husbandry.  Again though, are labels a big deal?  This is a great idea.

It is awesome, although Artificial Life (rather than AI) has a semi-quack history, it's a fascinating way to play around with emergence.  I would say labels are important, however, otherwise how do you hope to understand what's really going on?

And, astonishingly, the robots were even able to evolve altruistic behaviour, in a task that involved pushing tokens around. Some could be pushed single-handed, earning the robot one 'fitness point'; others required two robots, gaining the whole group one point.

I get confused where the line between mutually beneficial and altruistic is drawn at the species level.  At a genetic level, isn't it all selfish?

For example, I think a better test for altruism would be if helping another bot (for simplicity, of the same species - not a widely differing genome) push their token didn't necessarily grant a direct benefit to the individual who was helping.

example is not altruism.
example is selfish.
robot that tries to push 2robot tokens gets more points than if the same robot didn't try to push 2robot token.
if robot benefits, robot not altruistic.
indirect benefits count.

I agree this far.

it would be altruism if the robot did not benefit, but other robots did.

The behaviour which is influenced by the genetic code is what is being used to drive the selection mechanism though.  So if robot 1 chooses to help robot 2 instead of choosing a more lucrative activity - it will end up lower on the league table as an individual, but the genes which prompted that behaviour will be higher up on the genetic league table.  I hope I'm not mangling it too much to use the metaphor of individual vs. manufacturer leaderboards in Formula One racing.

A problem I have is how did the robots identify those with similar genomes?  If it's an innate sense then that's pretty much cheating.

So an individual can still benefit from an altruistic act, as long as there is a strong expectation that they benefit less than they would have otherwise done if the individual had pursued a more selfish goal.


Although, by both our definitions, playing the lottery is altruistic. 

evolutionary pressure can't create altruism. because pressure only exists if there is benefit.

I think you have to separate the individual (performs the actions) from the genetic (exists in multiple individuals and makes some actions more likely than others).   Otherwise, altruism doesn't exist and is a meaningless concept.

[Rococo Modem Basilisk]

I thought this was pretty cool, until I read the journal article and realized the title said "Darwinian Selection" aka Natural Selection. What's going on in this is artificial selection. In other words, the robots aren't selected by environment, or predation, or actual fitness (ie ability to produce offspring). Humans do the selecting, discarding those with the "worst performance".

I'd argue that this is splitting hairs. At what point does selection cease being natural? That's an old point, though, and it's probably been rehashed too much.

AWESOME BALLS

I concur.

[Triple Zero]

I thought this was pretty cool, until I read the journal article and realized the title said "Darwinian Selection" aka Natural Selection. What's going on in this is artificial selection. In other words, the robots aren't selected by environment, or predation, or actual fitness (ie ability to produce offspring). Humans do the selecting, discarding those with the "worst performance".

No not humans, computers do. This whole experiment could have run at night without humans present, until they come back the next morning and find a baby skynet in their lab. But you are correct in that the environment/"natural" selection does not.

The artificial life simulation Tierra does have something more akin to natural selection. In that simulation machine code programs have to evolve their reproduction routines themselves as well*, so the ones that do this most efficient and fast reproduce more. That would not be artificial selection, right? In this simulation, only the mutation is artificial, a tiny percentage of random bitflips in computer memory.

*except the initial organism, which is clumsily hand-coded, but replaced very quickly by more advanced automatically evolved species. but that's somewhat similar to the idea of earth life originating from a meteorite with some tiny single cellular proto fungus on it.