From: Jiun-Hung Chen (jhchen@cs.washington.edu)
Date: Wed Dec 01 2004 - 11:13:36 PST
The evolutionary origin of complex features
Richard E. Lenski, Charles Ofria, Robert T. Pennock and Christoph Adami
Review by Jiun-Hung Chen
Summary
They try to explain the origin of complex organismal features using digital organisms and
show how complex functions can originate by random mutation and natural selection.
Most important ideas
Using digital organisims in explaining evolutionary theory and other domains is a very important idea, especially when we does not have complete knowledge, information and evidence.
In the following, I describes the reasons why I think it is important. First, it is reasonable and practical. Furthermore, it is easy to control parameters and see how they effect results.
After all, it is possible that all biological evidence disappears. In addition, simulating a real organism in real environments may take billions of years and some good luck to develop a useful feature, let alone money. Instead, using digital organisims is very efficient compared to billions of years. In summary, I totally agree the last sentence in this paper, "they are particularly well suited to problems that are difficult to study with organic forms owing to incomplete information, insufficient time and the impracticality of experiments."
Largest flaws
The largest flaw is that rewards are independent of many important factors such as time, environment and species. I think their reward function is too ideal to be realistic. Instead, I believe that rewards should depend on many factors such as time, environment and species. For example, eyes are important for diurnal animals but ears can be more important than eyes
for them at night and for nocturnal animals. From this simple example, it is easy to understand rewards should be dependent on many factors.
The other main flaw is that they assume rewards are known and can be manually designed based on the complexity of an action. However, we do not know the "real" reward function in "real worlds". Because they do not try all possible reward functions, I think their results are special cases. So, I can not believe them fully. A more rigor way is to prove complex features are "fixed points" in dynamic systems which satisfy their assumption.
Open research questions
Simulating digital organisms under more realistic conditions merits further research.
For example, a more realistic world for digital organisms can be the following.
As time changes, environments and rewards change accordingly. In addition to energy, many other factors influence evolution of complex features and reproduction, too. For example, as authors mentioned, sex might accelerate the evolution of complex features. Hence, adding sex in simulation can be very interesting. These directions are all toward simulating digital organisms under more realistic conditions. Furthermore, to prove that complex features are "fixed points" in dynamic systems which satisfy their assumption is also important.
This archive was generated by hypermail 2.1.6 : Wed Dec 01 2004 - 11:13:37 PST