Robots Evolve Altruism, Just As Biology Predicts

[Oneironaut Zero]

Article Thought this was pretty awesome. Figured I'd share. (See what I did there? lol)

[Xei]

Cool, although I don't really get what's happening in the video. The article says the bots were already programmed to locate and collect the disks and the experiment was just about their deciding whether or not to cooperate, yet the video apparently shows that initial generations were inept at individual locomotion, which seems unrelated to the article. There's not really enough details about the specifics to understand what is happening here...

[Oneironaut Zero]

There's a video? Lol. I'm at work. They must have blocked it. I'll have to check it out when I get home.

[MadMonkey]

Thats pretty awesome, pretty sone they are going to start to group up and have wars over food points.

[tommo]

I don't get it. They clearly programmed the robots to help the others with a certain amount of the same genes as them. Doesn't really mean.... well, anything really.

[Xei]

I don't think that's what they did at all... the point was that the robots evolved such sharing behaviours as a result of natural selection, in a magnitude according to a theoretical equation. It didn't need to be programmed into them any more than fingers had to be programmed into the first fish to crawl out of the sea.

[nina]

Awh, aren't they adorable. Little ink cartridges on wheels. I'm more inclined to agree with the criticisms of the experiment. In some ways, the rule and its accompanying theory of kin selection is contested. Some scientists have used it to extrapolate too easily from insects to people, and some researchers think it overstates the importance of relatedness. But a more fundamental issue with Hamilton’s rule is the difficulty of testing it in natural systems, where animals evolve at a far slower pace than any research grant cycle. Simulations of evolution in robots, which can “reproduce” in mere minutes or hours, have thus become a potentially useful system for studying evolutionary dynamics. And though simple in comparison to animals, Keller’s group says robot models are not too different from the insects that originally inspired Hamilton. Basically they have programmed these robots in such a way, that altruism (sharing) has a higher benefit to cost ratio. Wouldn't it be implicit then that these robots would always develop "altruism"? To me it seems like all they are doing is selectively breeding for altruism, the same way we selectively breed for larger steak cows. This isn't evolution.

[Xei]

Artificial selection is evolution, it's just not natural selection. It doesn't seem to me like altruism was programmed to have a higher benefit / cost ratio; this was just an emergent property of the environment, as it is in nature. That, I think, was the point.

[sloth]

I agree with everyone. lol Evolution is what it is. If you were to break any concept down enough it all becomes part of absurdism. Perhaps they programmed these robots in such a way that they would inevitably end up cooperating... the article really doesn't make it clear just how much of this programming or behavior evolved on its own. Either way, the environment was set up to make it inevitable that sharing is beneficial to the whole of the group, and that is what the robots ended up doing. Robots are good people.

[tommo]

Artificial selection is evolution, it's just not natural selection. It doesn't seem to me like altruism was programmed to have a higher benefit / cost ratio; this was just an emergent property of the environment, as it is in nature. That, I think, was the point. I think we need the actual study to make any real judgement on how they did it. But that's just what it seemed like to me. It's not a particularly well written article though. Too bad you have to pay for most of these things. More people making money who aren't the people who do the actual science lol

[SnakeCharmer]

I think we need the actual study to make any real judgement on how they did it. But that's just what it seemed like to me. It's not a particularly well written article though. Too bad you have to pay for most of these things. More people making money who aren't the people who do the actual science lol You don't have to pay anything to read the study. All PLoS papers are open access. From what I understood, robots don't learn to cooperate from scratch. Rather, they are given a choice between two strategies, sharing their resources (cooperation), or keeping them (selfishness) and then they evolve the optimal allocation strategy. They "reproduce" based on how well their genotypes (programs/neural networks in this case) perform in gathering the resources. It's also not a case of artificial selection because the populations evolves towards cooperation by itself when certain conditions are satisfied (when relatedness between cooperation individuals is higher than the cost to benefit ratio). And by definition of altruism, benefit can never be higher than the cost.

[Xei]

To the individual, that is..?

[SnakeCharmer]

To the individual, that is..? Yes, I didn't explain that well. The cost of the cooperative action is always higher than the direct benefit for the actor, otherwise it cannot be classified as altruistic behavior. The benefit in Hamilton's rule is the benefit for the recipient, which is transformed into benefit for the actor via relatedness between the actor and the recipient.

[DrunkenArse]

Artificial selection is evolution, it's just not natural selection. I've never understood the difference between the two. But I've never really understood natural selection itself for that matter. It's not a real phenomenon. There's only selection pressures. They add up and we can take this selection pressure and lump it in with "natural" selection and that selection pressure and lump it in with "sexual" or "artificial" selection and end up with two or three nice little piles of pressures but they're just useful and imaginary models. One is really no different than the other.

[khh]

I've never understood the difference between the two. (...) One is really no different than the other. I disagree. The way I see it natural selection is something that just happens because of how the world works, while "artificial selection" is when the selection is guided by intelligence. The two are only the same if you believe in a god.

[Xei]

Yeah, I think the distinction is pretty... distinct. No more vague than any other human concept. In any case it's a useful explanatory tool; hence we keep it.

[DrunkenArse]

If they're so distinct, then why is a wasp only pollinating the flowers that most closely resemble female wasps not considered to be artificial selection. We could perfectly well consider the male wasp's genes to be selecting the individual plants that produce flowers that look the most like female wasps. The answer is because it's not that one magic monkey that lives at the center of the universe doing it and that's what makes "artificial" selection deserve it's own little category. We might say that "artifical" selection is "consciously" directed and that "natural" selection isn't but I have problems with that. First, the conscious mind doing the directing is the product of genetic expression. The wasp deciding to only attempt to mate with flowers that look like female wasps is the product of genetic expression. What's the difference? What does the criteria for consciousness add? Second, once we dispence with the illusion of the role of consciousness playing any deciding role in the distinction, we are left with "artificial" selection being "directed" as the distinction. But natural selection works exactly like "artificial" selection. It's directed towards the ability to propagate one's genes into future generations. What is so useful about the distinction anyways? I fail to understand it.

[Xei]

I think the point is that artificial selection has intention behind it. No other animal actually has conception of the existence of evolution, and knowledge of how to use it to their own ends. Natural selection is blind, responding only to the environment, whereas artificial selection has a specific goal in mind. I think it's quite clear that they are distinct processes, and artificial selection does various things which wouldn't make sense in natural selection... that makes the distinction useful.

[DrunkenArse]

The problem with that is that humans were engaging in "artificial selection" for thousands of years (at the very least) before we had any inkling of evolution. I also can't parse the fragment "responding only to the environment". At any rate all one does in artificial selection is change a portion of the environment, namely that portion of the environment that consists of the other species interactions with our own. So it's still "natural" selection that actually modifies the organism. Consider a miniature poodle. The thing should be hawk food. But an alien from another planet (that understood evolution) would be able to infer that it evolved in an environment where being cute and helpless conferred an evolutionary advantage. In this case, the genes passed through an environment where only the smallest and cutest animals got to breed. I fail to see how that's any different from any other phenomena that can be explained in terms of natural selection. There are ants farming fungi in some rain forest somewhere that understand evolution as well as we did when we domesticated the dog.

[tommo]

The problem with that is that humans were engaging in "artificial selection" for thousands of years (at the very least) before we had any inkling of evolution. I also can't parse the fragment "responding only to the environment". At any rate all one does in artificial selection is change a portion of the environment, namely that portion of the environment that consists of the other species interactions with our own. So it's still "natural" selection that actually modifies the organism. Consider a miniature poodle. The thing should be hawk food. But an alien from another planet (that understood evolution) would be able to infer that it evolved in an environment where being cute and helpless conferred an evolutionary advantage. In this case, the genes passed through an environment where only the smallest and cutest animals got to breed. I fail to see how that's any different from any other phenomena that can be explained in terms of natural selection. There are ants farming fungi in some rain forest somewhere that understand evolution as well as we did when we domesticated the dog. I'd say you are correct. It's the same thing as the fact that synthetic chemicals are just as natural as any other chemicals. I suppose the adjective "artificial" is just indicative of the fact that humans generally think they are separate from nature. But In any case it's a useful explanatory tool; hence we keep it.

[nina]

So allow me to clarify some of the issues I have with this study. I realize that I must be missing something though, for were it that simple, they wouldn't have wasted time and money doing the study (I assume). So hopefully one of you uber smart people can illuminate me on what I am misunderstanding. To me it seems as though they are using a very narrow (probably scientific) definition of altruism which links altruistic behavior with relatedness. The fact that your genes (or a portion of them) will have a higher percent chance of being passed on if you share with close relatives. Something like 10% chance of your survival, but 15% chance for someone related to you...then you would give them your food so that they could survive. Apparently it's just a mathematical question, and it's programmed into these little wheelers. The part I have issue with is the human control, the artificial and directed selection that only the most fit will reproduce (aka be cloned) and the least fit will drop out of the gene pool. Well that's fine, except for the fact that your new populations are going to be more and more inbred! I would like to know the inbreeding coefficient in this experiment, and if only clones were used, or also modified genetic mutation wheelers as well. It matters because...if altruism is a characterization of relatedness, then of course these robots would develop "altruism" in the very narrow sense of the word, because they are all HIGHLY inbred!! They're all related, so of course they will give up their food to help one another, because it helps their genes to survive as well. (Which, by real world genetics, would also mean that they are highly unfit! I mean, we know that inbreeding can produce blue people...yes BLUE) This sort of computer situation does NOT extrapolate to the real world. When they tried to do things like this with insects, they discovered many problems, such as you can only increase the size of an insect to a certain amount before size begins to hinder reproduction, and the insects reverted back to their optimal size. There are way too many variables to be taken into account in real nature, that this computer simulation just doesn't account for. And for that reason I have to agree with pretty much everyone else in that this study is interesting and might provide future benefit, but I don't think it does anything to prove that robots naturally evolved to develop altruism. Perhaps someone can help me understand what I am missing here?

[SnakeCharmer]

To me it seems as though they are using a very narrow (probably scientific) definition of altruism which links altruistic behavior with relatedness. They use the standard definition of altruism that is used in studies of social evolution, based on the cost of performing the altruistic act. This definition has nothing to do with relatedness. Relatedness comes from Hamilton's rule which tries to explain how altruism evolved, or better yet, what are the conditions that will make cooperative populations resistant to invasion of non-cooperating individuals. There are other factors that can promote cooperation that are not directly linked to relatedness, e.g. spatial structure in the population (it allows cooperators to be positioned closely to each other) or policing (where non-cooperators are punished in some way). The fact that your genes (or a portion of them) will have a higher percent chance of being passed on if you share with close relatives. Something like 10% chance of your survival, but 15% chance for someone related to you...then you would give them your food so that they could survive. Apparently it's just a mathematical question, and it's programmed into these little wheelers. The ability to share or not to share was programmed, but not the rationale on when it is beneficial to do so. The populations evolved to the state where they would behave altruistically when certain conditions were fulfilled. The part I have issue with is the human control, the artificial and directed selection that only the most fit will reproduce (aka be cloned) and the least fit will drop out of the gene pool. Well that's fine, except for the fact that your new populations are going to be more and more inbred! That's true. But the study is designed as a simple test of Hamilton's rule, not as an all-encompassing study of population genetics. I would like to know the inbreeding coefficient in this experiment, and if only clones were used, or also modified genetic mutation wheelers as well. There's a section where they add the mutations to the experiment, and get qualitatively the same results. Again, the study is not meant to capture all aspects of reality. It matters because...if altruism is a characterization of relatedness, then of course these robots would develop "altruism" in the very narrow sense of the word, because they are all HIGHLY inbred!! As said before, altruism is defined only in the sense of cost and benefit. And what you seem to think is obvious, is actually Hamilton's rule. It's just one of those ideas that seem obvious in hindsight, but it's actually quite important. Furthermore, as obvious as it is, it took almost 50 years to test it in an experimental system. To quote one blogger: results of this study are surprisingly unsurprising. They're all related, so of course they will give up their food to help one another, because it helps their genes to survive as well. (Which, by real world genetics, would also mean that they are highly unfit! Real world genetic studies of this idea have also been performed, and the results showed that cooperating populations were not highly unfit, they were fitter than non-cooperators and fitter than mixed populations. This was done with populations of pathogenic bacteria that cooperate and use public goods to initiate successful infections. This sort of computer situation does NOT extrapolate to the real world. It's not meant to. It's an experimental test of a simple model. And for that reason I have to agree with pretty much everyone else in that this study is interesting and might provide future benefit, but I don't think it does anything to prove that robots naturally evolved to develop altruism. Perhaps someone can help me understand what I am missing here? You are complaining about the right things. On the other hand, I think you misunderstood the purpose and the impact of the study, possibly because it was misinterpreted in the article that was linked in the first post. You are right, it doesn't show that robots naturally evolved to develop altruism. It shows that altruism will be an evolutionary stable strategy when relatedness is higher than the cost to benefit ratio.

[nina]

Thank you. That effectively clears it up.

[tommo]

So allow me to clarify some of the issues I have with this study. I realize that I must be missing something though, for were it that simple, they wouldn't have wasted time and money doing the study (I assume). Literally millions of dollars have been spent (like the most recent one was 3.2 million euros) on studies to see whether mobile phones cause cancer. When there's absolutely no mechanism that could possibly cause this to happen. Plus they are still saying they aren't sure, even when it's clear that there is absolutely no link. So expect more money and time to be spent.