4 ms·

Ah, so that's the other part of your misconceptions. As far as I can see, the point made in the (first part of) the book is that informed algorithms (on averag

Ah, so that's the other part of your misconceptions. As far as I can see, the point made in the (first part of) the book is that informed algorithms (on average) always outperform random searches and that genetic algorithms are akin to random searches. The first part is true, the second part isn't. Evolution isn't random, mutation is random. Selection is very much not random, it is informed by the environment. This drastically reduces the size so-called "search space". Using the term "search" for the outcomes of evolution is a curious choice anyway. Evolution doesn't search for anything in particular, it produces adaptation to an environment. It's a process without end. As for computer simulations, check out the simulated biped that evolved hopping like a kangaroo: https://www.youtube.com/watch?v=kQ2bqz3HPJE https://www.youtube.com/watch?v=kQ2bqz3HPJE

yters 7y ago
The book is not about biological evolution. They are analyzing artificial life simulations like Avida and Terra. Different topic than what we are discussing in the other thread. The fundamental point is that information is conserved, so if artificial life sims produce information, it must have been inserted into the code at some point, and it is possible to track down these particular insertion points, which the authors do. In information theory the conservation of information theorem goes by various names, such as the data processing inequality and independence conservation. Anyways, I highly recommend the book. Everything is clearly laid out for your own analysis and application.
- gridlockd 7y ago
  > The fundamental point is that information is conserved, so if artificial life sims produce information, it must have been inserted into the code at some point, and it is possible to track down these particular insertion points, which the authors do. Fair enough. Let's apply this to the video I linked. The authors certainly didn't add "information" about what walking or hopping looks like into the code explicitly. It just emerged. Following your logic, the hopping behavior can't be new information, it must've been contained in the rules of the simulation. The simulation itself is made up mostly of an off-the-shelf physics library, a couple of muscle-skeleton arrangements (authored by hand), and a fitness function measuring the distance traveled by activating those muscles through weights (the learning variables). Given this environment, walking, hopping and running behaviors invariably emerge. Fascinating! Adapting this to our world, the laws of physics and the initial arrangement of matter contain all the "information" required for life to eventually emerge in the universe. This leaves you with a "god of the gaps" opportunity: An intelligent designer must've designed those laws and made that arrangement of matter so that it all works out. This is however entirely compatible with Darwinian models. I don't have an issue with that kind of hypothesis, except that it is unclear what the existence of such a god/designer would imply for our particular lives here on earth.
  - yters 7y ago
    Have you worked much with evolutionary algorithms? It also seems you think that evolutionary algorithms are somehow analogous to biological evolution. To what degree do you think that is true?
    - gridlockd 7y ago
      > Have you worked much with evolutionary algorithms? Not really, like I said, there's usually better algorithms than genetic algorithms. They're fun to play around with, though. > It also seems you think that evolutionary algorithms are somehow analogous to biological evolution. To what degree do you think that is true? The general principle - selection and random mutation - is the same.
      - yters 7y ago
        What gives you so much confidence in GAs? Also, I am surprised you've followed this discussion for so long. Most people fire off some canned phrases and then disappear.
        gridlockd 7y ago
        > What gives you so much confidence in GAs? Do I sound confident about them? I've said multiple times that they usually aren't very good and that usually you have a better algorithm. Usually, you can apply intelligent design. The fundamental difference between us is probably that you'd look at life in nature and say something like: "This species is so remarkable, it must've been designed!". I wouldn't jump to that conclusion. It's only remarkable if it emerged naturally. If it was designed, then I would have to point out all sorts of obvious flaws in the design.
        yters 7y ago
        I haven't said anything about design. You are making assumptions. I said Darwinism appears to be completely useless when it comes to practical application, and has been replaced by other mechanisms that are identified in the biological data. In this new thread I said there is a conservation of information law that the book authors apply to artificial life simulations. I am talking about what we know through mathematics and quantitative empirical analysis. Regarding GAs, the conservation of information applies to how much information we can expect from a single random sample. If we can sample from the entire space of possible solutions, then there is a chance of sampling the optimal solution. The conservation of information applies to the necessary input to the algorithm in order to bring the chance to a certain level.

yters 7y ago

The book is not about biological evolution. They are analyzing artificial life simulations like Avida and Terra. Different topic than what we are discussing in the other thread. The fundamental point is that information is conserved, so if artificial life sims produce information, it must have been inserted into the code at some point, and it is possible to track down these particular insertion points, which the authors do. In information theory the conservation of information theorem goes by various names, such as the data processing inequality and independence conservation. Anyways, I highly recommend the book. Everything is clearly laid out for your own analysis and application.

gridlockd 7y ago
> The fundamental point is that information is conserved, so if artificial life sims produce information, it must have been inserted into the code at some point, and it is possible to track down these particular insertion points, which the authors do. Fair enough. Let's apply this to the video I linked. The authors certainly didn't add "information" about what walking or hopping looks like into the code explicitly. It just emerged. Following your logic, the hopping behavior can't be new information, it must've been contained in the rules of the simulation. The simulation itself is made up mostly of an off-the-shelf physics library, a couple of muscle-skeleton arrangements (authored by hand), and a fitness function measuring the distance traveled by activating those muscles through weights (the learning variables). Given this environment, walking, hopping and running behaviors invariably emerge. Fascinating! Adapting this to our world, the laws of physics and the initial arrangement of matter contain all the "information" required for life to eventually emerge in the universe. This leaves you with a "god of the gaps" opportunity: An intelligent designer must've designed those laws and made that arrangement of matter so that it all works out. This is however entirely compatible with Darwinian models. I don't have an issue with that kind of hypothesis, except that it is unclear what the existence of such a god/designer would imply for our particular lives here on earth.
- yters 7y ago
  Have you worked much with evolutionary algorithms? It also seems you think that evolutionary algorithms are somehow analogous to biological evolution. To what degree do you think that is true?
  - gridlockd 7y ago
    > Have you worked much with evolutionary algorithms? Not really, like I said, there's usually better algorithms than genetic algorithms. They're fun to play around with, though. > It also seems you think that evolutionary algorithms are somehow analogous to biological evolution. To what degree do you think that is true? The general principle - selection and random mutation - is the same.
    - yters 7y ago
      What gives you so much confidence in GAs? Also, I am surprised you've followed this discussion for so long. Most people fire off some canned phrases and then disappear.
      - gridlockd 7y ago
        > What gives you so much confidence in GAs? Do I sound confident about them? I've said multiple times that they usually aren't very good and that usually you have a better algorithm. Usually, you can apply intelligent design. The fundamental difference between us is probably that you'd look at life in nature and say something like: "This species is so remarkable, it must've been designed!". I wouldn't jump to that conclusion. It's only remarkable if it emerged naturally. If it was designed, then I would have to point out all sorts of obvious flaws in the design.
        yters 7y ago
        I haven't said anything about design. You are making assumptions. I said Darwinism appears to be completely useless when it comes to practical application, and has been replaced by other mechanisms that are identified in the biological data. In this new thread I said there is a conservation of information law that the book authors apply to artificial life simulations. I am talking about what we know through mathematics and quantitative empirical analysis. Regarding GAs, the conservation of information applies to how much information we can expect from a single random sample. If we can sample from the entire space of possible solutions, then there is a chance of sampling the optimal solution. The conservation of information applies to the necessary input to the algorithm in order to bring the chance to a certain level.