From: Indriyati Atmosukarto (indria@cs.washington.edu)
Date: Tue Nov 30 2004 - 23:33:48 PST
The evolutionary origin of complex features
Richard E. Lenski, Charles Ofria, Robert T. Pennock and Christoph Adami
This paper presents a road map to see how the evolutionary sequence of complex functions unfolds in its entirety through random mutation and natural selection, without missing any links. The authors created a population of identical digital organisms using a program called Avida. At the start, each digital organisms was incapable of solving logic problems but mutation in the offsprings altered the nature of the organisms and in some cases results in ones that could perform complex functions.
An important finding in the paper is how they observe that it is impossible for a population of digital organisms to solve difficult logic problem if the program only awarded rewards for complex problems. But if the organisms were also rewarded if they performed simpler functions, the results changed dramatically. The organisms were able to evolve and bridge the gap and eventually solve the complex logical problems. The digital organism solved the complex problems by modifying existing structure and functions of their ancestors which were used to solve the simple problems. Another important point is the fact that some of the essential mutation actually arose through hitchhiking detrimental mutations. Hence, this should raise an alert that in the real biology world we should not just reject mutation because even though it looks like it has some detrimental impact in the short term, we may in fact be missing the overall big picture.
Reading the paper somehow leaves the impression that the experiments were designed in a way that the results were expected. Higher computational merits were given to the complex function, which shows that the awards were not arbitrary but carefully chosen hence the mutation of the organisms to solve EQU should not come as a surprise. The set of random rewards is extraordinarily large and a majority of them would actually not have resulted in convergence of complex functions. It seems that to evolve more complex functions, the authors would have to specify it in advance, which leads to the second flaw of the paper in that the evolution of these digital organisms may not truly reflect the lifelike biological evolutionary progress.
Possible research directions would involve conducting the study on sexual replication which was mentioned in the paper, and trying out more possible rewards to show a more random behavior to elevate the suspicion that the experiments are just following a prepared path thus simulating a more lifelike evolution.
We can classify this problem as a local search problem such as genetic algorithms where the fitness function of the algorithm is actually the computational merit awarded to the organisms and the search space could be the space of every possible state. It seems that the search incorporates some sort of hill-climbing however rather than always moving up, the search actually sometimes moves sideways or even step down. This is because some of the mutation were harmful in the short term but survived the forces natural selection and ended up paying a crucial role in the development of the complex functions.
This archive was generated by hypermail 2.1.6 : Tue Nov 30 2004 - 23:33:48 PST