Friam

Genetic algorithm for groups

Classic

List

Threaded

9 messages Options

Russ Abbott

Genetic algorithm for groups

David Sloan Wilson has been an advocate of group selection in evolution for quite a while. (And I think he's right.) What I'd like to know is whether anyone knows of any work on group selection in a (computational) genetic algorithm context.

Suppose I wanted to evolve a fleet of cars for a car rental agency. One approach would be a genetic algorithm in which the population elements were fleets, each of which is a collection of cars. Crossover would generate children fleets some of whose cars were copied from each parent.

In addition, I want to assume that the car properties themselves are evolvable. So one could, for example, crossover two cars to produce offspring cars with properties from the two parents.

This has also been called multi-level selection because evolution takes place at multiple levels at once: in this case at the fleet level and at the car level simultaneously

Is anyone aware of a framework that supports this sort of process? Or is anyone aware of any papers that describe results in this area?

Thanks.

-- Russ

-- Russ Abbott
______________________________________

Professor, Computer Science
California State University, Los Angeles

cell: 310-621-3805
blog: http://russabbott.blogspot.com/
vita: http://sites.google.com/site/russabbott/
______________________________________

============================================================
FRIAM Applied Complexity Group listserv
Meets Fridays 9a-11:30 at cafe at St. John's College
lectures, archives, unsubscribe, maps at http://www.friam.org

Eric Charles

Re: Genetic algorithm for groups

Russ, et. al,
I should send an email focusing on group selection, but instead I will point out, on a very related note, that there was a pretty nice altruism article published by some of the people on the list not too long ago ;- ) http://jasss.soc.surrey.ac.uk/9/2/4.html -- That article demonstrates that a strategy that always co-operates, but changes partners if faced with a defector, out performs strategies that only co-operate under certain circumstances (e.g., the much revered tit-for-tat). At least one of the authors knows Wilson pretty darn well, and another got to present the paper in a symposium with Wilson and got pretty good compliments.

I had a fantasy about creating a genetic algorithms version of the same program, but got side tracked on other projects. The idea was that we would start with a population of all non-co-operators non-leavers. Each would have a "chromosome" where there was a low probability it would "mutate", gaining or losing whichever ability the gene represented. Presumably it would take many, many generations for co-operation to emerge as a contender in the population. Given a limited number of generations, most ! populations would be unlikely to evolve altruism (i.e., the occasional mutation would be quickly eliminated). However, the interesting study would be too look back at those populations in which altruism DID evolved, and determine the order of events. Our hypothesis, based on the prior simulation (and the really good logic behind it) would be that leaving evolves first, then co-operation. At least, that would be the typical pattern.

It would be a really fun study, and I would be happy to help put it together. It would be done already except for two factors 1) a dispersion of the interested parties and 2) new Netlogo versions required tweaking the original program more than the remaining brain-power allowed. The last version was pretty heavily documented (admittedly by people who are not skilled at the art), so it shouldn't take a skilled programer too long to fix it up.

Anyway, already a longer email than intended,

Eric

P.S. Nick knows the group! selection stuff backwards and forwards. I can do pretty good ! schpeel too, and you should scold me for not having answered your question more exactly. The reason this is related is because group selection is only an interesting conversation (i.e., only a controversial conversation) if you are trying to use it to explain the evolution of altruism.

On Tue, Mar 9, 2010 08:52 PM, Russ Abbott <[hidden email]> wrote:

David Sloan Wilson has been an advocate of <a href="http://www.nbb.cornell.edu/wkoenig/wicker/NB4340/Wilson&Wilson2008.pdf" onclick="window.open('http://www.nbb.cornell.edu/wkoenig/wicker/NB4340/Wilson&Wilson2008.pdf');return false;">group selection in evolution for quite a while. (And I think he's right.) What I'd like to know is whether anyone knows of any work on group selection in a (computational) genetic algo! rithm context.

Suppose I wanted to evolve a fleet of cars for a car rental agency. One approach would be a genetic algorithm in which the population elements were fleets, each of which is a collection of cars. Crossover would generate children fleets some of whose cars were copied from each parent.

In addition, I want to assume that the car properties themselves are evolvable. So one could, for example, crossover two cars to produce offspring cars with properties from the two parents.

This has also been called multi-level selection because evolution takes place at multiple levels at once: in this case at the fleet level and at the car level simultaneously

Is anyone aware of a framework that supports this sort of process? Or is anyone aware of any papers that describe results in this area?

Thanks.

-- Russ

-- Russ Abbott
______________________________________

Professor, Computer Science
California State University, Los Angeles

cell: 310-621-3805
blog: <a href="http://russabbott.blogspot.com/" onclick="window.open('http://russabbott.blogspot.com/');return false;">http://russabbott.blogspot.com/
vita: <a href="http://sites.google.com/site/russabbott/" onclick="window.open('http://sites.google.com/site/russabbott/');return false;">http://sites.google.com/site/russabbott/
______________________________________
============================================================
FRIAM Applied Complexity Group listserv
Meets Fridays 9a-11:30 at cafe at St. John's College
lectures, archives, unsubscribe, maps at http://www.friam.org

Eric Charles

Professional Student and
Assistant Professor of Psychology
Penn State University
Altoona, PA 16601

Russ Abbott

Re: Genetic algorithm for groups

Thanks, Eric. Very interesting message. But it didn't address the questions I asked.

Does anyone know of any work on a genetic algorithm system that supports group selection -- or of papers in that specific area.

Thanks.

-- Russ Abbott
______________________________________

Professor, Computer Science
California State University, Los Angeles

cell: 310-621-3805
blog: http://russabbott.blogspot.com/
vita: http://sites.google.com/site/russabbott/
______________________________________

On Tue, Mar 9, 2010 at 7:16 PM, ERIC P. CHARLES <[hidden email]> wrote:

Russ, et. al,
I should send an email focusing on group selection, but instead I will point out, on a very related note, that there was a pretty nice altruism article published by some of the people on the list not too long ago ;- ) http://jasss.soc.surrey.ac.uk/9/2/4.html -- That article demonstrates that a strategy that always co-operates, but changes partners if faced with a defector, out performs strategies that only co-operate under certain circumstances (e.g., the much revered tit-for-tat). At least one of the authors knows Wilson pretty darn well, and another got to present the paper in a symposium with Wilson and got pretty good compliments.

I had a fantasy about creating a genetic algorithms version of the same program, but got side tracked on other projects. The idea was that we would start with a population of all non-co-operators non-leavers. Each would have a "chromosome" where there was a low probability it would "mutate", gaining or losing whichever ability the gene represented. Presumably it would take many, many generations for co-operation to emerge as a contender in the population. Given a limited number of generations, most ! populations would be unlikely to evolve altruism (i.e., the occasional mutation would be quickly eliminated). However, the interesting study would be too look back at those populations in which altruism DID evolved, and determine the order of events. Our hypothesis, based on the prior simulation (and the really good logic behind it) would be that leaving evolves first, then co-operation. At least, that would be the typical pattern.

It would be a really fun study, and I would be happy to help put it together. It would be done already except for two factors 1) a dispersion of the interested parties and 2) new Netlogo versions required tweaking the original program more than the remaining brain-power allowed. The last version was pretty heavily documented (admittedly by people who are not skilled at the art), so it shouldn't take a skilled programer too long to fix it up.

Anyway, already a longer email than intended,

Eric

P.S. Nick knows the group! selection stuff backwards and forwards. I can do pretty good ! schpeel too, and you should scold me for not having answered your question more exactly. The reason this is related is because group selection is only an interesting conversation (i.e., only a controversial conversation) if you are trying to use it to explain the evolution of altruism.

On Tue, Mar 9, 2010 08:52 PM, Russ Abbott <[hidden email]> wrote:
David Sloan Wilson has been an advocate of group selection in evolution for quite a while. (And I think he's right.) What I'd like to know is whether anyone knows of any work on group selection in a (computational) genetic algo! rithm context.

Suppose I wanted to evolve a fleet of cars for a car rental agency. One approach would be a genetic algorithm in which the population elements were fleets, each of which is a collection of cars. Crossover would generate children fleets some of whose cars were copied from each parent.

In addition, I want to assume that the car properties themselves are evolvable. So one could, for example, crossover two cars to produce offspring cars with properties from the two parents.

This has also been called multi-level selection because evolution takes place at multiple levels at once: in this case at the fleet level and at the car level simultaneously

Is anyone aware of a framework that supports this sort of process? Or is anyone aware of any papers that describe results in this area?

Thanks.

-- Russ

-- Russ Abbott
______________________________________

Professor, Computer Science
California State University, Los Angeles

cell: 310-621-3805
blog: http://russabbott.blogspot.com/
vita: http://sites.google.com/site/russabbott/
______________________________________
============================================================
FRIAM Applied Complexity Group listserv
Meets Fridays 9a-11:30 at cafe at St. John's College
lectures, archives, unsubscribe, maps at http://www.friam.org
Eric Charles

Professional Student and
Assistant Professor of Psychology
Penn State University
Altoona, PA 16601

Nick Thompson

Re: Genetic algorithm for groups

In reply to this post by Russ Abbott

russ,

I little sober reflection (no pun intended) will reveal that David Sloan Wilson's "trait group selection" is actually a mechanism for quantitative inheritance of group traits. See

http://home.earthlink.net/~nickthompson/naturaldesigns/id49.html

where you can download the pdf by clicking on the abstract

or download the pdf directly from

http://www.clarku.edu/faculty/nthompson/1-websitestuff/Texts/2000-2005/Shifting_the_natural_selection_metaphor_to_the_group_level.pdf

The full treatment of this mechanism is in DSW's Natural Selection of Populations and Communities (?) 1979, i think.

Enjoy,

Nicholas S. Thompson

Emeritus Professor of Psychology and Ethology,

Clark University ([hidden email])

http://home.earthlink.net/~nickthompson/naturaldesigns/

http://www.cusf.org [City University of Santa Fe]

----- Original Message -----

From: [hidden email]
To: [hidden email]
Cc: [hidden email]

Sent: 3/9/2010 9:12:04 PM

Subject: Re: [FRIAM] Genetic algorithm for groups

Thanks, Eric. Very interesting message. But it didn't address the questions I asked.

Does anyone know of any work on a genetic algorithm system that supports group selection -- or of papers in that specific area.

Thanks.

-- Russ Abbott
______________________________________

Professor, Computer Science
California State University, Los Angeles

cell: 310-621-3805
blog: http://russabbott.blogspot.com/
vita: http://sites.google.com/site/russabbott/
______________________________________

On Tue, Mar 9, 2010 at 7:16 PM, ERIC P. CHARLES <[hidden email]> wrote:

Russ, et. al,
I should send an email focusing on group selection, but instead I will point out, on a very related note, that there was a pretty nice altruism article published by some of the people on the list not too long ago ;- ) http://jasss.soc.surrey.ac.uk/9/2/4.html -- That article demonstrates that a strategy that always co-operates, but changes partners if faced with a defector, out performs strategies that only co-operate under certain circumstances (e.g., the much revered tit-for-tat). At least one of the authors knows Wilson pretty darn well, and another got to present the paper in a symposium with Wilson and got pretty good compliments.

I had a fantasy about creating a genetic algorithms version of the same program, but got side tracked on other projects. The idea was that we would start with a population of all non-co-operators non-leavers. Each would have a "chromosome" where there was a low probability it would "mutate", gaining or losing whichever ability the gene represented. Presumably it would take many, many generations for co-operation to emerge as a contender in the population. Given a limited number of generations, most ! populations would be unlikely to evolve altruism (i.e., the occasional mutation would be quickly eliminated). However, the interesting study would be too look back at those populations in which altruism DID evolved, and determine the order of events. Our hypothesis, based on the prior simulation (and the really good logic behind it) would be that leaving evolves first, then co-operation. At least, that would be the typical pattern.

It would be a really fun study, and I would be happy to help put it together. It would be done already except for two factors 1) a dispersion of the interested parties and 2) new Netlogo versions required tweaking the original program more than the remaining brain-power allowed. The last version was pretty heavily documented (admittedly by people who are not skilled at the art), so it shouldn't take a skilled programer too long to fix it up.

Anyway, already a longer email than intended,

Eric

P.S. Nick knows the group! selection stuff backwards and forwards. I can do pretty good ! schpeel too, and you should scold me for not having answered your question more exactly. The reason this is related is because group selection is only an interesting conversation (i.e., only a controversial conversation) if you are trying to use it to explain the evolution of altruism.

On Tue, Mar 9, 2010 08:52 PM, Russ Abbott <[hidden email]> wrote:

David Sloan Wilson has been an advocate of group selection in evolution for quite a while. (And I think he's right.) What I'd like to know is whether anyone knows of any work on group selection in a (computational) genetic algo! rithm context.

Suppose I wanted to evolve a fleet of cars for a car rental agency. One approach would be a genetic algorithm in which the population elements were fleets, each of which is a collection of cars. Crossover would generate children fleets some of whose cars were copied from each parent.

In addition, I want to assume that the car properties themselves are evolvable. So one could, for example, crossover two cars to produce offspring cars with properties from the two parents.

This has also been called multi-level selection because evolution takes place at multiple levels at once: in this case at the fleet level and at the car level simultaneously

Is anyone aware of a framework that supports this sort of process? Or is anyone aware of any papers that describe results in this area?

Thanks.

-- Russ

-- Russ Abbott
______________________________________

Professor, Computer Science
California State University, Los Angeles

cell: 310-621-3805
blog: http://russabbott.blogspot.com/
vita: http://sites.google.com/site/russabbott/
______________________________________

============================================================ FRIAM Applied Complexity Group listserv Meets Fridays 9a-11:30 at cafe at St. John's College lectures, archives, unsubscribe, maps at http://www.friam.org
Eric Charles

Professional Student and
Assistant Professor of Psychology
Penn State University
Altoona, PA 16601

Russ Abbott

Re: Genetic algorithm for groups

Thanks, Nick,

You raise interesting and difficult points: what is a group (when it is not predefined in the problem), what are its properties, etc. I took my first pass through your paper too quickly to understand your solution.

But I'm not facing that problem. In a computational GA environment, I don't have to answer the question of what a group is. It is (trivially) a member of my population. All I'm asking is whether there are computational systems that let me work with such a computational framework and whether there are any papers about such a computational framework.

Since I'm not getting any answers to my actual questions, I'm supposing that the answer is that there aren't any such systems.

On the other hand, one might argue that what I'm really asking for is a Pitt style learning classifier system. Within that framework, one evolves collections of rules, which is basically what I'm asking for except that instead of thinking of each element of each group as necessarily a rule it is an individual more generally conceived.

-- Russ Abbott
______________________________________

Professor, Computer Science
California State University, Los Angeles

cell: 310-621-3805
blog: http://russabbott.blogspot.com/
vita: http://sites.google.com/site/russabbott/
______________________________________

On Tue, Mar 9, 2010 at 9:12 PM, Nicholas Thompson <[hidden email]> wrote:

russ,

I little sober reflection (no pun intended) will reveal that David Sloan Wilson's "trait group selection" is actually a mechanism for quantitative inheritance of group traits. See

http://home.earthlink.net/~nickthompson/naturaldesigns/id49.html

where you can download the pdf by clicking on the abstract

or download the pdf directly from

http://www.clarku.edu/faculty/nthompson/1-websitestuff/Texts/2000-2005/Shifting_the_natural_selection_metaphor_to_the_group_level.pdf

The full treatment of this mechanism is in DSW's Natural Selection of Populations and Communities (?) 1979, i think.

Enjoy,

N

Nicholas S. Thompson

Emeritus Professor of Psychology and Ethology,

Clark University ([hidden email])

http://home.earthlink.net/~nickthompson/naturaldesigns/

http://www.cusf.org [City University of Santa Fe]
----- Original Message -----

From: [hidden email]
To: [hidden email]
Cc: [hidden email]

Sent: 3/9/2010 9:12:04 PM

Subject: Re: [FRIAM] Genetic algorithm for groups

Thanks, Eric. Very interesting message. But it didn't address the questions I asked.

Does anyone know of any work on a genetic algorithm system that supports group selection -- or of papers in that specific area.

Thanks.

-- Russ Abbott
______________________________________

Professor, Computer Science
California State University, Los Angeles

cell: 310-621-3805
blog: http://russabbott.blogspot.com/
vita: http://sites.google.com/site/russabbott/
______________________________________

On Tue, Mar 9, 2010 at 7:16 PM, ERIC P. CHARLES <[hidden email]> wrote:

Russ, et. al,
I should send an email focusing on group selection, but instead I will point out, on a very related note, that there was a pretty nice altruism article published by some of the people on the list not too long ago ;- ) http://jasss.soc.surrey.ac.uk/9/2/4.html -- That article demonstrates that a strategy that always co-operates, but changes partners if faced with a defector, out performs strategies that only co-operate under certain circumstances (e.g., the much revered tit-for-tat). At least one of the authors knows Wilson pretty darn well, and another got to present the paper in a symposium with Wilson and got pretty good compliments.

I had a fantasy about creating a genetic algorithms version of the same program, but got side tracked on other projects. The idea was that we would start with a population of all non-co-operators non-leavers. Each would have a "chromosome" where there was a low probability it would "mutate", gaining or losing whichever ability the gene represented. Presumably it would take many, many generations for co-operation to emerge as a contender in the population. Given a limited number of generations, most ! populations would be unlikely to evolve altruism (i.e., the occasional mutation would be quickly eliminated). However, the interesting study would be too look back at those populations in which altruism DID evolved, and determine the order of events. Our hypothesis, based on the prior simulation (and the really good logic behind it) would be that leaving evolves first, then co-operation. At least, that would be the typical pattern.

It would be a really fun study, and I would be happy to help put it together. It would be done already except for two factors 1) a dispersion of the interested parties and 2) new Netlogo versions required tweaking the original program more than the remaining brain-power allowed. The last version was pretty heavily documented (admittedly by people who are not skilled at the art), so it shouldn't take a skilled programer too long to fix it up.

Anyway, already a longer email than intended,

Eric

P.S. Nick knows the group! selection stuff backwards and forwards. I can do pretty good ! schpeel too, and you should scold me for not having answered your question more exactly. The reason this is related is because group selection is only an interesting conversation (i.e., only a controversial conversation) if you are trying to use it to explain the evolution of altruism.

On Tue, Mar 9, 2010 08:52 PM, Russ Abbott <[hidden email]> wrote:

David Sloan Wilson has been an advocate of group selection in evolution for quite a while. (And I think he's right.) What I'd like to know is whether anyone knows of any work on group selection in a (computational) genetic algo! rithm context.

Suppose I wanted to evolve a fleet of cars for a car rental agency. One approach would be a genetic algorithm in which the population elements were fleets, each of which is a collection of cars. Crossover would generate children fleets some of whose cars were copied from each parent.

In addition, I want to assume that the car properties themselves are evolvable. So one could, for example, crossover two cars to produce offspring cars with properties from the two parents.

This has also been called multi-level selection because evolution takes place at multiple levels at once: in this case at the fleet level and at the car level simultaneously

Is anyone aware of a framework that supports this sort of process? Or is anyone aware of any papers that describe results in this area?

Thanks.

-- Russ

-- Russ Abbott
______________________________________

Professor, Computer Science
California State University, Los Angeles

cell: 310-621-3805
blog: http://russabbott.blogspot.com/
vita: http://sites.google.com/site/russabbott/
______________________________________

============================================================ FRIAM Applied Complexity Group listserv Meets Fridays 9a-11:30 at cafe at St. John's College lectures, archives, unsubscribe, maps at http://www.friam.org
Eric Charles

Professional Student and
Assistant Professor of Psychology
Penn State University
Altoona, PA 16601
============================================================
FRIAM Applied Complexity Group listserv
Meets Fridays 9a-11:30 at cafe at St. John's College
lectures, archives, unsubscribe, maps at http://www.friam.org

Jochen Fromm-4

Re: Genetic algorithm for groups

In reply to this post by Eric Charles

Is group selection only interesting if you are using it to explain the
evolution of altruism? Can you explain this further? Is it because the
different levels of selection collide in this case?

-J.

----- Original Message -----
From: ERIC P. CHARLES
To: Russ Abbott
Cc: The Friday Morning Applied Complexity Coffee Group
Sent: Wednesday, March 10, 2010 4:16 AM
Subject: Re: [FRIAM] Genetic algorithm for groups

[..] group selection is only an interesting conversation (i.e., only a
controversial conversation) if you are trying to use it to explain the
evolution of altruism.

============================================================
FRIAM Applied Complexity Group listserv
Meets Fridays 9a-11:30 at cafe at St. John's College
lectures, archives, unsubscribe, maps at http://www.friam.org

Nick Thompson

Re: Genetic algorithm for groups

In reply to this post by Russ Abbott

In great haste: group selection is of interest in explaining group design.
Group design may look like "altrusim" to a dyed in the wool individual
selectionist, but more importantly, it is just group design.

N

Nicholas S. Thompson
Emeritus Professor of Psychology and Ethology,
Clark University ([hidden email])
http://home.earthlink.net/~nickthompson/naturaldesigns/
http://www.cusf.org [City University of Santa Fe]

> [Original Message]
> From: Jochen Fromm <[hidden email]>
> To: The Friday Morning Applied Complexity Coffee Group <[hidden email]>
> Date: 3/10/2010 3:36:12 PM
> Subject: Re: [FRIAM] Genetic algorithm for groups
>
> Is group selection only interesting if you are using it to explain the
> evolution of altruism? Can you explain this further? Is it because the
> different levels of selection collide in this case?
>
> -J.
>
> ----- Original Message -----
> From: ERIC P. CHARLES
> To: Russ Abbott
> Cc: The Friday Morning Applied Complexity Coffee Group
> Sent: Wednesday, March 10, 2010 4:16 AM
> Subject: Re: [FRIAM] Genetic algorithm for groups
>
> [..] group selection is only an interesting conversation (i.e., only a
> controversial conversation) if you are trying to use it to explain the
> evolution of altruism.
>
>
>
> ============================================================
> FRIAM Applied Complexity Group listserv
> Meets Fridays 9a-11:30 at cafe at St. John's College
> lectures, archives, unsubscribe, maps at http://www.friam.org

Eric Charles

Re: Genetic algorithm for groups

In reply to this post by Jochen Fromm-4

Jochen, and any others interested,

Roughly:
Evolutionary biologists, if they are interested in group selection at all, are interested because it might explain traits difficult or impossible to explain by reference to individual selection. If we define altruism as a trait that lowers an individuals evolutionary fitness relative to others in the population, then altruism is THE thing that is difficult to explain in terms of individual selection. (That definition, of course, captures what me normally mean by altruism in plain-language conversations, and at least some other stuff.) Once upon a time, people explained altruism as the result of selection acting at the group level, and this was largely discredited - It is crazy to say that lemmings evolved to run off cliffs, because the group benefits as a result.

Wilson offers something in between the standard modern individual-centric model and the crude old group selection arguement. He breaks the population of organisms into sub-groups. He then shows that traits which work against a group member relative to within-group organisms can still be advantageous relative to out-group organisms. That is, if a group contains several altruists, then the whole group grows at a fast enough rate that it doesn't matter that within-group the % of altruists is shrinking. (Between generations you need to shuffle.) Thus, altruism can evolve if you have a multi-level selection model, and define altruism relative to within group peers. If you commit the "averaging fallacy" and just average across groups, it looks like plain Jane natural selection.

Eric

P.S. I'm attaching a proof of concept worksheet I whipped up for one of Nick/my classes on the subject. Be sure to notice in the first 2) That in Cs do worse than Ns in both groups when you look at the groups individually! For the first step 3, the point is that having Cs in your group provides an advantage to the other group members, thus the group with more Cs, has a higher survival rate overall. I don't remember why 'C' stands for 'callers' and 'N' for 'non-callers', but at any rate 'C' is the trait that looks altruistic when you do the multi-level analysis.

On Wed, Mar 10, 2010 03:36 PM, "Jochen Fromm" <[hidden email]> wrote:

Is group selection only interesting if you are using it to explain the

evolution of altruism? Can you explain this further? Is it because the

different levels of selection collide in this case?

-J.

-----
Original Message ----- 
From: ERIC P. CHARLES
To: Russ Abbott
Cc: The
Friday Morning Applied Complexity Coffee Group
Sent: Wednesday, March 10,
2010 4:16 AM
Subject: Re: [FRIAM] Genetic algorithm for groups

[..]
group selection is only an interesting conversation (i.e., only a

controversial conversation) if you are trying to use it to explain the

evolution of
altruism.



============================================================
FRIAM Applied Complexity Group listserv
Meets Fridays 9a-11:30 at cafe at St. John's College
lectures, archives, unsubscribe, maps at http://www.friam.org

Eric Charles

Professional Student and
Assistant Professor of Psychology
Penn State University
Altoona, PA 16601

Averaging Fallacy.doc (58K) Download Attachment

Russ Abbott

Re: Genetic algorithm for groups

In reply to this post by Russ Abbott

I agree that it's an important subject. And I agree completely with you and DSW -- although I think his "Truth and Reconciliation" series on the Huffington Post a while ago on the subject was a bit overboard. I liked his Evolution for Everyone book quite a lot. (See, for example, this post for some pointers.)

-- Russ Abbott
______________________________________

Professor, Computer Science
California State University, Los Angeles

cell: 310-621-3805
blog: http://russabbott.blogspot.com/
vita: http://sites.google.com/site/russabbott/
______________________________________

On Thu, Mar 11, 2010 at 12:39 AM, Nicholas Thompson <[hidden email]> wrote:

Russ,

Well, let's see what DSW suggests! I think he is in Australia, or something, at the moment, but he is better about email than any human of his degree of busy-ness I know, so we might get an answer.

Look. I don't even pretend to be able to participate in a conversation in which "computational framework" is the subject. However, if one is looking for a set of rules that that relates degree of possession of trait in the groups of one generation to the possession of the traits in the subsequent generation, quantitative inheritance (aka=trait group selection) is such a rule-set.

All sorts of luck with your project. I think it's a really important issue to sort out. I happened to have a conversation with a Famous Person Who Ought To Have Known Better yesterday who stoutly informed me that kin selection was the only way to go. I have not thought particularly hard about it recently, but it seems to me that the answer is ultimately philosophical. Is it the case that we can make a choice about how we talk about the subject, or is it the case that we HAVE to include group effects in the conversation to understand what is going on. As I think David would say, whether the group members are related or not is a red herring. The fact is that differential group effects is a vector (one of several) that effects the relative frequency of alleles. IMHO.

Sorry to keep getting you off your topic, but my conversation with the FPWOTHKB and your introduction of the topic here got me all excited again.

Nick

Nicholas S. Thompson

Emeritus Professor of Psychology and Ethology,

Clark University ([hidden email])

http://home.earthlink.net/~nickthompson/naturaldesigns/

http://www.cusf.org [City University of Santa Fe]
----- Original Message -----

From: [hidden email]

To: [hidden email];[hidden email]

Sent: 3/10/2010 7:24:00 AM

Subject: Re: [FRIAM] Genetic algorithm for groups

Thanks, Nick,

You raise interesting and difficult points: what is a group (when it is not predefined in the problem), what are its properties, etc. I took my first pass through your paper too quickly to understand your solution.

But I'm not facing that problem. In a computational GA environment, I don't have to answer the question of what a group is. It is (trivially) a member of my population. All I'm asking is whether there are computational systems that let me work with such a computational framework and whether there are any papers about such a computational framework.

Since I'm not getting any answers to my actual questions, I'm supposing that the answer is that there aren't any such systems.

On the other hand, one might argue that what I'm really asking for is a Pitt style learning classifier system. Within that framework, one evolves collections of rules, which is basically what I'm asking for except that instead of thinking of each element of each group as necessarily a rule it is an individual more generally conceived.

-- Russ Abbott
______________________________________

Professor, Computer Science
California State University, Los Angeles

cell: 310-621-3805
blog: http://russabbott.blogspot.com/
vita: http://sites.google.com/site/russabbott/
______________________________________

On Tue, Mar 9, 2010 at 9:12 PM, Nicholas Thompson <[hidden email]> wrote:

russ,

I little sober reflection (no pun intended) will reveal that David Sloan Wilson's "trait group selection" is actually a mechanism for quantitative inheritance of group traits. See

http://home.earthlink.net/~nickthompson/naturaldesigns/id49.html

where you can download the pdf by clicking on the abstract

or download the pdf directly from

http://www.clarku.edu/faculty/nthompson/1-websitestuff/Texts/2000-2005/Shifting_the_natural_selection_metaphor_to_the_group_level.pdf

The full treatment of this mechanism is in DSW's Natural Selection of Populations and Communities (?) 1979, i think.

Enjoy,

N

Nicholas S. Thompson

Emeritus Professor of Psychology and Ethology,

Clark University ([hidden email])

http://home.earthlink.net/~nickthompson/naturaldesigns/

http://www.cusf.org [City University of Santa Fe]

----- Original Message -----

From: [hidden email]

To: [hidden email]
Cc: [hidden email]

Sent: 3/9/2010 9:12:04 PM

Subject: Re: [FRIAM] Genetic algorithm for groups

Thanks, Eric. Very interesting message. But it didn't address the questions I asked.

Does anyone know of any work on a genetic algorithm system that supports group selection -- or of papers in that specific area.

Thanks.

-- Russ Abbott
______________________________________

Professor, Computer Science
California State University, Los Angeles

cell: 310-621-3805
blog: http://russabbott.blogspot.com/
vita: http://sites.google.com/site/russabbott/
______________________________________

On Tue, Mar 9, 2010 at 7:16 PM, ERIC P. CHARLES <[hidden email]> wrote:

Russ, et. al,
I should send an email focusing on group selection, but instead I will point out, on a very related note, that there was a pretty nice altruism article published by some of the people on the list not too long ago ;- ) http://jasss.soc.surrey.ac.uk/9/2/4.html -- That article demonstrates that a strategy that always co-operates, but changes partners if faced with a defector, out performs strategies that only co-operate under certain circumstances (e.g., the much revered tit-for-tat). At least one of the authors knows Wilson pretty darn well, and another got to present the paper in a symposium with Wilson and got pretty good compliments.

I had a fantasy about creating a genetic algorithms version of the same program, but got side tracked on other projects. The idea was that we would start with a population of all non-co-operators non-leavers. Each would have a "chromosome" where there was a low probability it would "mutate", gaining or losing whichever ability the gene represented. Presumably it would take many, many generations for co-operation to emerge as a contender in the population. Given a limited number of generations, most ! populations would be unlikely to evolve altruism (i.e., the occasional mutation would be quickly eliminated). However, the interesting study would be too look back at those populations in which altruism DID evolved, and determine the order of events. Our hypothesis, based on the prior simulation (and the really good logic behind it) would be that leaving evolves first, then co-operation. At least, that would be the typical pattern.

It would be a really fun study, and I would be happy to help put it together. It would be done already except for two factors 1) a dispersion of the interested parties and 2) new Netlogo versions required tweaking the original program more than the remaining brain-power allowed. The last version was pretty heavily documented (admittedly by people who are not skilled at the art), so it shouldn't take a skilled programer too long to fix it up.

Anyway, already a longer email than intended,

Eric

P.S. Nick knows the group! selection stuff backwards and forwards. I can do pretty good ! schpeel too, and you should scold me for not having answered your question more exactly. The reason this is related is because group selection is only an interesting conversation (i.e., only a controversial conversation) if you are trying to use it to explain the evolution of altruism.

On Tue, Mar 9, 2010 08:52 PM, Russ Abbott <[hidden email]> wrote:

David Sloan Wilson has been an advocate of group selection in evolution for quite a while. (And I think he's right.) What I'd like to know is whether anyone knows of any work on group selection in a (computational) genetic algo! rithm context.

Suppose I wanted to evolve a fleet of cars for a car rental agency. One approach would be a genetic algorithm in which the population elements were fleets, each of which is a collection of cars. Crossover would generate children fleets some of whose cars were copied from each parent.

In addition, I want to assume that the car properties themselves are evolvable. So one could, for example, crossover two cars to produce offspring cars with properties from the two parents.

This has also been called multi-level selection because evolution takes place at multiple levels at once: in this case at the fleet level and at the car level simultaneously

Is anyone aware of a framework that supports this sort of process? Or is anyone aware of any papers that describe results in this area?

Thanks.

-- Russ

-- Russ Abbott
______________________________________

Professor, Computer Science
California State University, Los Angeles

cell: 310-621-3805
blog: http://russabbott.blogspot.com/
vita: http://sites.google.com/site/russabbott/
______________________________________

============================================================ FRIAM Applied Complexity Group listserv Meets Fridays 9a-11:30 at cafe at St. John's College lectures, archives, unsubscribe, maps at http://www.friam.org
Eric Charles

Professional Student and
Assistant Professor of Psychology
Penn State University
Altoona, PA 16601

============================================================
FRIAM Applied Complexity Group listserv
Meets Fridays 9a-11:30 at cafe at St. John's College
lectures, archives, unsubscribe, maps at http://www.friam.org