|
Author
|
Topic: Evolution Simulations
|
Mike Baughman
Member
Member # 270
|
posted 10. May 2002 11:38
Although I'm not a biologist I have been interested in the attempts to do computer models of evolution, such as the MESA and ev programs.
It seems that these programs in their attempt to demonstrate whether or not you can build complex specified information through a deterministic chance/selection algorithm are lacking in an important aspect: they approach a target which is more or less arbitrary in that it doesn't correspond to anything semantically useful. That is to say, CSI has to fit a pattern that has "meaning", which in biological as well as computer systems is expressed as function.
Computer programs and organisms both express function of some kind by means of executing "code" - chemically/electrically in organisms, electronically in computers.
It may be feasible to set up a more realistic model (although still vastly more simplistic than biological reality) whereby a functioning program could be "evolved" to see if it can gain actual function solely through deterministic variation/selection routes without any predetermination. What I envision is building an "organism" consisting of several routines. Each organism would have an "alive" switch. The routines would execute as long as the "alive" switch is on:
1. Metabolism - a loop which executes continually the following subroutines.
a) maintenance: a routine to search for and "eat" some bit string (ASCII 'F' for "food"?). If the search fails within a certain number of iterations, the organism dies (stops executing).
b) growth: a routine to find needed bit strings and store them internally for use in adding structures.
2. Reproduction - a routine that is executed once after the organism has lived for a while (i.e. executed some set number of iterations of "metabolism"):
a) This routine would build a copy of the main metabolism loop using bit strings stored internally by "growth".
b) Some sort of variation is applied: random mutation (allowed in the target addresses of instructions but not the operation code); duplication of subroutines (gene duplication) (co-optation could be a side effect of duplication); elimination of subroutines (gene reduction); lateral transfer (possibly by having "growth" able to subsume an entire functioning subroutine - I'm not sure exactly how this would work.)
c) After building the copy, the parent program then turns on the child's "alive" switch and it begins executing separately from the parent.
The simulation would proceed by launching a certain number of organisms in an environment full of bit strings that might be useful as "food". As resources are depleted (turn bit strings to zero when they've been eaten) organisms will die. When they die, their "alive" switch is turned off and they become part of the environment, available for other organisms to eat.
Obviously this simulation would require a vast area to execute in; it might require multiple I-streams to work.
I'm not sure what such a program would really prove. My hunch is that no "new" functions would emerge. It would then be liable to criticism from evolutionists on the grounds of not being sophisticated enough or an accurate model; ID theorists on the other hand, if it did "evolve", might point to how much intelligence is already designed in up front. But as criticisms were incorporated into tuning such a program, it might eventually be useful as a crude indicator of the feasibility or lack thereof of the deterministic variation/selection mechanism.
Any suggestions or comments out there?
IP: Logged
|
|
warren_bergerson
Member
Member # 262
|
posted 10. May 2002 14:56
Mike,
I think your suggestion of actually attempting to simulate Darwinian evolution is a useful one. Current evolutionary theory depends to a very large extent on the assumption that if Darwinian mechanism can produce very simple change, then over time these processes could explain all observed change. No one seems to have actually attempted to determine 1)how much or how many changes are associated with a particular change and 2)can Darwinian mechanisms actually explain all the changes that occur. To actually attempt to simulate ‘real evolution’ on even a simplified basis would be very enlightening.
The analysis I have been doing suggests or predicts the result your experiment will produce. I predict that not only will or organisms be unable to evolve new functions, but in an even slightly complex environment, your organisms will be unable to survive.
My ‘prediction’ is based on a concept called adaptive change. Darwinian theory, at least in its simplest form, suggests that an organism adapts to a single set of environmental conditions. Overtime the organisms or species moves toward an optimal relationship to the environment in which it lives. If the environment changes, then the organism will change or evolve to adapt to the new environment.
The adaptive change concept asserts that the environment in which an organism must survive involves many, many combinations of conditions and if the organism is to survive, it must have appropriate adaptive solutions to each of the environmental conditions and the organism must be able to generate the appropriate adaptive solution at the right time.
Before you get your organism to evolve new functions, you will need to get it to survive a few million different combinations of conditions. To even begin addressing this issue you will have to start adding new processes and mechanism to the Darwinian process.
It is possible to explain the complexity that exists in biological systems using mechanistic processes, but you need processes and mechanism that are far more powerful and far faster than the Darwinian processes.
I have developed an idealized model of such processes. I call the model a ‘Life Force Simulator’. I would be glad to share a copy of it if you are interested.
In their basic forms, both Darwinian evolution and ID suggest that the processing which defines a complex organism occurred before the organism came into being. The adaptive approach, by contrast, suggests that that is impossible. The adaptive approach suggest that most of the complexity in a complex organism is created or reinvented during the organisms lifetime. Most of the ‘intelligence’ needed to create such a complex organism exists within the organism. Both ID and Darwinian theory can, of course, by made consistent with the adaptive approach by assuming that evolution or external designers created the ability to create complex designs, rather than the complex designs themselves.
Again I think the experiment you propose is a good one. I would also suggest that it wouldn’t take a very complex model to demonstrate that the simple forms of both ID and Darwin won’t work. From that useful starting point, it may be possible to obtain more precise, and more easily testable forms of both types of theory.
IP: Logged
|
|
John Bracht
Member
Member # 5
|
posted 10. May 2002 15:57
Mike,
Reading your post reminded me of a book I've read. I went back and looked at "An Introduction to Genetic Algorithms" by Melanie Mitchell (MIT press, 1996), where she describes scientific modelling with evolutionary algorithms. In particular, she describes an experiment modelling an ecosystem which sounds very similar to what you are describing. The model is called "Echo", invented by John Holland, Terry Jones, and Stephanie Forrest.
This model utilizes a 2-D lattice containing different types of "resources" which represent sources of energy and are represenated by letters of the alphabet. The world has "agents" in it which have a genotype consisting of a set of rules governing the interactions with other agents and stipulating how the organism looks to the outside world (the phenotype). At each timestep the agent either interacts with other agents or, (if it doesn't have any resources at its current location) moves to another location. When organisms encounter each other they can trade, fight, or mate.
It seems that this model is so new that researchers are still comparing it with the real world ecosystems, looking at the relative distributions of various genome types to see whether they agree with the wild. No detailed research findings were reported in the book. Here are some references on Echo:
Holland J.H. 1994. Echoing emergence: Objectives, rough definitions, and speculartions for ECHO-class models. IN G. Cowan, D. Pines, and D. Melzner, eds, Complexity: Metaphors, Models, and Reality. Addison-Wesley.
Jones, T., and Forrest, S. 1993. An Introduction to SFI Echo. Working Paper 93-12-074, Santa Fe Institute.
Forrest, S., and Jones, T. 1994. MOdeling complex adaptive systems with Echo. In R.J. Stonier and X.H. Yu, eds., Complex Systems: Mechanism of Adaptation. IOS Press.
There were other similar but less complicated systems that Mitchell described for exploring the Baldwin effect, sexual selection, and evolutionary innovation.
So--it seems that other people have had similar ideas to yours and testing of such models is happening but is still in the preliminary stages.
John Bracht
IP: Logged
|
|
warren_bergerson
Member
Member # 262
|
posted 11. May 2002 09:41
John,
JB: So--it seems that other people have had similar ideas to yours and testing of such models is happening but is still in the preliminary stages.
I am somewhat surprised at the above comment. It is well known that GA or mutate-select models can only optimize on one set of conditions at a time. It is also well known that the survival and reproduction of an organism requires solutions to multiple adaptive problems.
It therefore follows that Darwinian evolutionary processes could not possibly or logically operate without additional functionality. Unless or until the additional functionality is specifically defined, there is no scientific theory to test.
The purpose of complex simulations like the one suggested by Mike is not to ‘test Darwinian theory’ since at present no such explicitly formulated scientific theory exists. The purpose of complex simulations is to attempt to identify and/or discover what functionality would be required to formulate a specific testable scientific theory of evolutionary change.
IP: Logged
|
|
Mike Baughman
Member
Member # 270
|
posted 13. May 2002 14:46
Thank you for the comments and resources. I hadn't heard of the Echo simulation before - it sounds interesting.
A "realistic" evolutionary model needs to quantify in terms of chance+law not just that complexity can increase, but that the resultant complexity is useful ("specified"). As Warren has suggested, I expect that the difficulties of getting the "organisms" to survive, let alone get "better", will prove insurmountable without introducing intelligence.
IP: Logged
|
|
|
Printer-friendly view of this topic