I've posted before about the fact that my current fellowship is in the area of EvoDevo, and that sometimes people (like Fodor) come up with flaky definitions or claims about that. The workshop this past weekend was great, and really clarified for me a bunch of the issues, both what the exciting new ideas are, but also on the more historical / philosophy of science level why some people take the strange directions.
To make sure my posting makes sense, I want to start with the obvious, which is basic natural selection (NS). To get evolution, you need:
The reason is because of something called evolvability. The idea is that as selection keeps happening for billions of years and through enormous environmental variation, it doesn't just select for things like longer legs or shorter noses, but also for traits that make it easier to change quickly if the environment does. So an example Marc Kirschner brought up was a recent paper on Darwin's finches. It turns out all the variation in beak size is controlled by just two genes -- one for how long it is and one for how thick it is. When you look at these birds heads, their heads are totally different so that the beaks can fit on. But that didn't require separate mutations / evolution. Rather, because development has become such a powerful, complicated process that changing one gene results in all the necessary extra things happening. In fact, these changes will happen even if there is no genetic change, just if say the mother is in a strange environmental situation, maybe she is hungry. The embryo will do the best it can with the resources it has to make a complete, functioning organism.
This is what the EvoDevo people (like the director of my institute, Gerd Müller) are studying, all the developmental mechanisms that make evolution more powerful and go faster. We learned about all kinds of things like that, including epigenetics. We also learned about multi-level selection. There was a paleontologist from the University of Chicago, David Jablonski gave this amazing talk that really showed it didn't matter very much what cool body plans you evolved if your entire clade gets wiped out just for being at the wrong latitude when a major environmental transition happens.
So I'm not sure, but I think this is why some people think selection isn't that big a deal. It's because they see more and more things accounted for by other mechanisms, basically more impressive mechanisms for generating variation, so they think (logically) that selection accounts for less and less. In some ways that's true -- for example, that there are now a lot more mechanisms we could teach kids in biology to help explain how life on earth got so complicated in "only" four billion years. But from my perspective, as an engineer, in a way evolvability just makes selection more powerful. It just means that in relatively few generations of variation & selection you can get much bigger changes.
In fact, I realized at some point while talking to Marc Kirschner that the ideas he was trying to put forward in his book about biology were incredibly similar to the ideas I was trying to put forward in my PhD. He was saying that evolution could go faster because of these mechanisms, I was saying you needed to build mechanisms / modules to make developing AI go faster. In evolution, some people really want to believe this means selection can go away, because they don't like selection's implications (it's not at all egalitarian.) In AI, some people really want to believe that various mechanisms or algorithms mean the problem of building / developing AI will go away and it will build itself, because AI development is done by people and is a hard task people can get wrong. All those people are wrong. AI development and evolutionary selection will never go away, but they will get what they do done faster and faster with better mechanisms.
I by no means mentioned all the interesting talks! But I thought it was worth blogging quickly about what I learned. On Monday I also learned some very cool stuff about monogamy after having gone to the KLI for Ethology again and talking to Richard Wagner. (Will got to meet the keas too.) Mostly I've been working on a grant with Benedikt Herrmann this week, hopefully that will end & I can submit a couple journal articles & get back to doing my new models -- hopefully before I have to prepare my talk and poster for the next scientific meeting I am going to, 1 August.
To make sure my posting makes sense, I want to start with the obvious, which is basic natural selection (NS). To get evolution, you need:
- More young individuals born in each generation than winds up breeding in the next generation.
- Some variation between those individuals, which has to be heritable (that is, something that affects these individuals' children)
- A process of selection that makes it more likely that individuals with some kinds of variations wind up producing the next generation. This is the famous "survival of the fittest", but note that "fittest" only really means "most likely to reproduce".
The reason is because of something called evolvability. The idea is that as selection keeps happening for billions of years and through enormous environmental variation, it doesn't just select for things like longer legs or shorter noses, but also for traits that make it easier to change quickly if the environment does. So an example Marc Kirschner brought up was a recent paper on Darwin's finches. It turns out all the variation in beak size is controlled by just two genes -- one for how long it is and one for how thick it is. When you look at these birds heads, their heads are totally different so that the beaks can fit on. But that didn't require separate mutations / evolution. Rather, because development has become such a powerful, complicated process that changing one gene results in all the necessary extra things happening. In fact, these changes will happen even if there is no genetic change, just if say the mother is in a strange environmental situation, maybe she is hungry. The embryo will do the best it can with the resources it has to make a complete, functioning organism.
This is what the EvoDevo people (like the director of my institute, Gerd Müller) are studying, all the developmental mechanisms that make evolution more powerful and go faster. We learned about all kinds of things like that, including epigenetics. We also learned about multi-level selection. There was a paleontologist from the University of Chicago, David Jablonski gave this amazing talk that really showed it didn't matter very much what cool body plans you evolved if your entire clade gets wiped out just for being at the wrong latitude when a major environmental transition happens.
So I'm not sure, but I think this is why some people think selection isn't that big a deal. It's because they see more and more things accounted for by other mechanisms, basically more impressive mechanisms for generating variation, so they think (logically) that selection accounts for less and less. In some ways that's true -- for example, that there are now a lot more mechanisms we could teach kids in biology to help explain how life on earth got so complicated in "only" four billion years. But from my perspective, as an engineer, in a way evolvability just makes selection more powerful. It just means that in relatively few generations of variation & selection you can get much bigger changes.
In fact, I realized at some point while talking to Marc Kirschner that the ideas he was trying to put forward in his book about biology were incredibly similar to the ideas I was trying to put forward in my PhD. He was saying that evolution could go faster because of these mechanisms, I was saying you needed to build mechanisms / modules to make developing AI go faster. In evolution, some people really want to believe this means selection can go away, because they don't like selection's implications (it's not at all egalitarian.) In AI, some people really want to believe that various mechanisms or algorithms mean the problem of building / developing AI will go away and it will build itself, because AI development is done by people and is a hard task people can get wrong. All those people are wrong. AI development and evolutionary selection will never go away, but they will get what they do done faster and faster with better mechanisms.
I by no means mentioned all the interesting talks! But I thought it was worth blogging quickly about what I learned. On Monday I also learned some very cool stuff about monogamy after having gone to the KLI for Ethology again and talking to Richard Wagner. (Will got to meet the keas too.) Mostly I've been working on a grant with Benedikt Herrmann this week, hopefully that will end & I can submit a couple journal articles & get back to doing my new models -- hopefully before I have to prepare my talk and poster for the next scientific meeting I am going to, 1 August.
Comments