|
Author
|
Topic: Constructing a relevant Avida experiment
|
Royal
Member
Member # 1060
|
posted 11. July 2004 16:10
[Micah Sparacio]: quote:
RBH has recommended that IDists take advantage of the Avida environment and construct some Avida world to test various ID hypothesis.
[Royal]: If neo-Darwinian Theory is to be falsifiable in principle via Avida runs, the relevant experiments must permit the following:
(a)There must be gradations of rewards for the same logic function. This is simply a biological fact. Mutations can lead to functional but sub-optimal performance. I’ll leave out the details here.
For n alternative mutants with fitness level X (i.e., a logic function), there are many times more functional mutants with fitness level between 0,95X and 1.0X; and there are many times more functional mutants with fitness level between 0,9X to 0,95X than the preceding level. Eventually the mutants are non-functional. The penantly comparison of a mutant is not against the optimal variant but against the average fitness of a population at that point in time.
(b) Very small genomes only are relevant, and therefor genome truncation in the absence of logic functions needs to be by a factor proportional to genome length. (This factor is a parameter the researcher would set).
One starts the experiment with all organisms possessing multiple logic functions. More complex logic functions are available via random mutations, but the statistical probability needs to be determined. The Avida runs in the literature fall absurdly at the lower end of Bill Dembski’s “events of intermediate probability” which can reasonably be expected to occur by chance. I showed in my own thread that Avida logic functions are ridiculously easy to produce by chance as compared to novel and more complex cellular functions.
The researchers can then choose settings at will. Here is what is going to result, according to the parameters one arbitrarily chooses:
(1) New complex logic functions will evolve: use high rewards for new logic functions; very little indispensible genetic material which can be destroyed by mutations; high mutation rates; little or no genome truncation; and many logic functions which are easy to produce.
(2) The net effect will be loss of logic functions: use low rewards, much indispensible material, low mutation rates, significant genome truncation, and more complex logic functions which are difficult to produce. Facilitate genome degradation gracefully by assuming small penalties for sub-optimal mutants, but that these are far more numerous; assume there are several such degradation steps.
I would be very pleased if Avida runs could be made to model real organisms closely enough to determine what the net effect of random mutations plus natural selection is. I believe the answer is: small adjustments to environmental changes by slight modifications of existing proteins and biochemical networks can occur; trivial at best increases, in very rare cases, of novel complex biological functions could occur; and much degradation and loss of pre-existing functions will be found.
IP: Logged
|
|
Royal
Member
Member # 1060
|
posted 11. July 2004 16:43
[RBH 21. April 2004]: quote:
One of Truman's criticims, in fact, is related to that issue - the 'intermediates' are held to be too close together in code sequence space - and that's a question well worth examining quantitatively.
I address this matter to some extent on the other Avida thread. For NOT and NAND I use random instructions in the starting sequence instead of nop-C. No rewards, and long runs. I determined the proportion of NOT and NAND found vs. the number of different mutants produced. I repeated as a function of starting genome length. (What was found: extrapolation to protein lengths implies about every random polypeptide would have to have some useful function. This is hopelessly unrealistic).
The next step is similar: start with all genomes having one or both the NOT, NAND logic functions, determine the proportion of more complex logic functions generated vs. all mutants.
From http://www.iscid.org/boards/ubb-get_topic-f-6-t-000465-p-4.html
RBH wrote: quote:
To take a fast look at the last suggestion that Truman made (#5, reduce the selective advantage for performing more complex logic functions) I freed up two fast machines last night and made three evolutionary runs with three different randomization seeds. The runs were of a maximum of 250,000 updates with the following fitness function [Function: My Computational Merit (Lenski Computational Merit)]:
NOT: 1.25 (2) NAND: 1.5 (2)
AND: 2 (4) OR_N: 2 (4)
OR: 2.5 (8) AND_N: 2.5 (8)
NOR: 3 (16) XOR 4 (16)
EQU 4 (32)
[Royal]: The results after repeating this experiment several times is easy to predict. The logic functions are so ridiculously close together (as least in biological terms), that I showed in the other thread one can even throw out NOT and NAND altogether (reward = 0 !!), lower the other rewards generously and complex logic functions still evolve. It is simply an artefact of computer logic functions, it has no cellular relevance.
Incidently RBH's rewards, such as 3, 4 are powers of 2. The relative fitnesses are about 2^^3 and 2^^4 instead of 0.001 or 0.01 as is commonly assumed. These kinds of values would be relevant only in the cases of humanly produced poisons designed to kill bacteria. Some members would survive by already having had fortuitious mutations which just happened to save their necks, er, lives.
Nevertheless, I also showed that using biologically representative rewards and fewer stepping stones prevented any logic functions from fixing even after a million updates.
IP: Logged
|
|
|
Printer-friendly view of this topic