Paper 1 review (Jonas Klink)

From: jklink@u.washington.edu
Date: Wed Dec 01 2004 - 00:36:11 PST

  • Next message: Vaishnavi Sannidhanam: "Paper 1 Review"

    The evolutionary origin of complex features
    Richard E. Lenski, Charles Ofria, Robert T. Pennock & Christoph Adami

    One-line summary
    The paper argues for verification of Charles Darwin’s theory on complex biological features evolving from simpler ones, by using digital organisms as a simulation platform.

    Main ideas
    The basis for the setting of the experiment(s) is founded on meeting three different conditions: replication, mutation and competition. These three conditions can be out together into the two main ideas of the paper; evolution of digital life in an artificial environment (includes replication and competition) and progressing from simple to complex (by the influence of mutation).

    The replication process itself spawns the competition and mutation; simulated through the need for energy (processing time) to perform a self-copy, which is erroneous with a certain probability. Evolution of logical functions (simulating biological features) is awarded with an additional amount of energy; exponentially proportional to the complexity of the function (thereby striving to duplicate the survival of the fittest from Darwin’s famous works).

    The artificial setting of the experiment allows tracking of the evolution of certain complex functions (in this case EQU; equality) through time, and also adds the ability to alter the ancestral trail by interchanging the effect of deleterious or beneficial steps, and study the relative outcome. By varying the reward amongst the simpler logical functions (and also fully removing it; thereby obliterating the evolution of EQU), it is claimed that different combinations of simpler functions are needed to evolve the more complex ones; thereby proving the accuracy of Darwin’s theory.

    Flaws
    One major flaw in the setting of this experiment and the following conclusions are the fact that this is in fact just a simulation in an artificial biological population of organisms. Therefore, it is not including vital characteristics as potentially beneficial traits transferred between populations (by holding the population size fixed and not allowing migration) and sexual reproduction between the truly fittest individuals. By restraining the environment to such a degree, the jump from generalizing the final conclusions to the real world seems awfully big.

    A second flaw is a combination of features of this setting, concerning the replication and competition process. By just directly replacing an individual in the fixed grid of the population, a fitter individual can potentially (in many cases) be randomly replaced by the weaker one, thereby not accurately depicting a real biological setting. Also, how the rewards for the pre-selected simpler functions was assigned is also a bit unclear, since it seems to rely a lot on the conclusions of the experiments as established facts (by assuming simpler functions are worth less than complex ones).

    Open questions and improvements
    The biggest question of this argument is to me if this kind of search (comparable to a local search with a start in the ancestor organism, a goal in having the complex function EQU and randomization (mutations) and rewards for generating good successors (fitness)) can be applied to other search spaces than the purely biological one. What makes this search problem different from many other settings is for example that in many cases, an obviously bad (deleterious or downhill) step can lead directly to a state where the search agent suddenly is in a global maximum. Also, the evolving of some positive traits might completely obliterate others, with a trade-off between search instances being present.

    Another open question is for this kind of setting to overcome the flaws of the simplification of the biological environment, to make this setting useful for researching and predicting the behaviour of other traits in a real population in nature. If this can be achieved, and a viable model established, many of nature’s fascinating complexities, like e.g. the effect and usefulness of chaos in nature (as well as in closely related search environments) can be fully explored.


  • Next message: Vaishnavi Sannidhanam: "Paper 1 Review"

    This archive was generated by hypermail 2.1.6 : Wed Dec 01 2004 - 00:36:11 PST