Author Topic: Natural Selection as an abstraction for game theory  (Read 13369 times)

Offline Numsgil

  • Administrator
  • Bot God
  • *****
  • Posts: 7742
    • View Profile
Natural Selection as an abstraction for game theory
« on: December 05, 2006, 03:46:03 PM »
After a great deal of reading, this is the pet idea I've formed in my mind.  I assume it's not novel, but it was an epiphany to my sleep depraved mind.

Natural selection is almost always viewed as a property of individuals, or as Dawkins imagined, individual genes.  But I think this view is limiting because it makes explaining ecological concepts constrained.  Altruism, human nationalism, etc. become difficult constructs to resolve.  They happen on abstractions order of magnitude removed from classical natural selection.

What I suggest is abstracting natural selection to being a property of any game (as in game theory) with agent(s).  Genes compete inside an organism.  Orgranisms compete within a species.  Species compete for niches.  Nations compete with other nations.  In this abstract natural selection, the goal is not to reproduce necessarily.  The goal is to maximize the probability of your existance.  And it's not so much a goal as a consequence that if you don't, you lose.

Agents which do not do this cease to exist, replaced with agents that do exist.  For individuals this means reproduction.  The more copies of you that exist, the larger the probability that you won't be wiped out by a random death.  But reproduction doesn't really make sense from the framepoint of countries as a successful strategy.  For countries, long term survival means a strong military, good diplomacy, internal stability, and growth (either in the economy or in physical borders, etc.).

The phenomenon of natural selection is the tendancy for inept players in a game to be removed.  Natural selection tends to operate on several layers at once in the real world.  Every level that has a game going on really.  Individual levels have to balance the fact that the different players on one abstraction level might be playing as a group on another abstraction layer.  You can have several games being played on several different levels.  Higher abstraction levels, though, make the results of lower level games moot if the higher abstraction game loses.

Which is where the really interesting dynamic comes into play.  Why do slime molds cooperate in such a way that a great majority of them die to ensure the survival of a few.  Why isn't every cell selfish?  Because if they were they would lose the game on the next higher abstraction layer and the game between individuals wouldn't matter.  Slime mold colonies that have too many cheaters cease to exist.

Once something "wins" a game, that game no longer exists and the level below it becomes the highest abstraction game.  Humans do not generally play as a single species against other sentient creatures because there are no other sentient species humans are competing against.  The level of abstraction falls back down to the nation states.  If we were competing against the replitoids of Gamma 6, there would still be competition between countries, but there would also be alot of cooperation.

Note that this generalized natural selection does not offer any insights into the strategies that are used in the different abstraction layers.  It does not say how to be a good species.  It only says that poor species are removed from the game.  The different abstraction layers play games of very different rule sets.  Random mutations work on a genomic level game, but would get you killed quickly in a Risk style game.

What's interesting here is that "winning" isn't selected either for or against, except in that by removing competitors (other players) you can eventually achieve a full blown win, with your existance never questioned again on this abstraction layer (until the game starts again anyway).

A beetle that pushes out all the beetles from its niche cannot ever suffer non existance unless foreign beetles invade (the game starts over) or its destroyed on another level (the planet is hit by an asteroid and explodes).

"So what?" you might be asking.  Traditional natural selection doesn't make sense in my ex nihilo sims.  There isn't reproduction, so how can there be natural selection?  Something is still happening though.  Bots that get themselves killed are still removed from the game.  This game is interesting though because it's solitare.  Bots might interact randomly, but it's not until a bot learns to reproduce and/or eat that competition really develops between the different bots (the abstraction moves to the next layer up).  In this case it's a competition against time and mutations.  Mutations will eventually cause a behavior that will kill the bot, and it'll lose.  But since point mutations are low frequency local events, a replicator will almost guarentee its indefinite existance against point mutations.

The ultimate winning strategy for this simple game is to become a self replicator.  Then it moves to the next level: competition between individuals, where classic natural selection is very familiar.

This also makes leagues have a certain logic.  The winner of a round has permanent existance rights (unless you restart a round).  The species it defeated can never again challenge the winner's existance, because it's extinct.  During this struggle, cannibalism isn't a good strategy, because while it may increase the cannibal's immediate success, it weakens its species success, and can cause the entire species (including itself) to lose to the other species.  The species that develops and allows cannibals to run free during a competition will have a higher probability of dying.

Conspec genes would probably be selected for if a league round lasted long enough for evolution to take hold.  Once this game is won, though, the abstraction collapses to the one below, between individuals, and this trait is selected against.

And the problem of big berthas we had back in the day: big berthas were the winners of the highest abstraction game: the struggle between individuals.  Once they won that, so a big bertha killed all other bots that could possibly kill it (big berthas else where don't count if they're far enough away), the abstraction falls back down to the lower level: the solitare fight against time and point mutations.

All in all it's a very useful framework for explaining alot of results that would seem to contradict the classical natural selection.  The only issue that doesn't seem clear to me are the events that need to happen to move up an abstraction layer.  Darwinbots has an abstraction ceiling.  Evolution can't seem to progress beyond competition between individuals.  And if it's artificially set up to start at a higher abstraction layer (as in the leagues), it inevitably resolves the higher abstraction and moves back down to a lower abstraction level.  I think I know why this is true for Darwinbots.  I think we solved the problem of big berthas by making waste a more real problem bots have to deal with.  I think we can raise the abstraction ceiling another notch by adding the ability to specialize.

But can this be abstracted away from our implementation?  What needs to happen in an abstraction level to cause a higher abstraction layer to form in a closed system (no immigration of other species, for example)?
« Last Edit: December 05, 2006, 03:48:09 PM by Numsgil »

Offline Jez

  • Bot Overlord
  • ****
  • Posts: 788
    • View Profile
Natural Selection as an abstraction for game theory
« Reply #1 on: December 05, 2006, 05:34:30 PM »
Ok, in my midly sleep depraved, ethanol distorted POV I shall try to  deal with some of the many points you have raised in a sensible  manner...
         
Quote from: Numsgil
Natural selection is almost always viewed as a property of individuals, or as Dawkins imagined, individual genes.
           The  selfish gene? Altruism is explained there as a species advantage, some  'people' are willing to risk their lives for the propogation of the  species as a whole, one effect one answer is not a survival strategy.
           
           
Code: [Select]
Genes compete inside an organism.            Genes make an organism? I don't see how individual genes compete unless it is organism V organism.
           
           
Code: [Select]
  Species compete for niches. Nations compete with other nations. In this  abstract natural selection, the goal is not to reproduce necessarily.
            A nation is a niche, a nation is a subspecies (in some ways) no reproduction = no evolution = you die.
           
           
Code: [Select]
But  reproduction doesn't really make sense from the framepoint of countries  as a successful strategy. For countries, long term survival means a  strong military, good diplomacy, internal stability, and growth (either  in the economy or in physical borders, etc.).            Why not? If a  country doesn't reproduce it ceases to exist, Germany during WWII had a  medal for mothers that had lots of children.
           
           
Code: [Select]
Higher  abstraction levels, though, make the results of lower level games moot  if the higher abstraction game loses .            Homo Sapieans plays the highest abstraction level? If the lower abstracition levels were removed Homo Sapiens wouldn't exist...
           
           
Code: [Select]
  Why do slime molds cooperate in such a way that a great majority of  them die to ensure the survival of a few. Why isn't every cell selfish?            Selfish genes, it is not the individual it is the species that allows the genes to exist.
           
           
Code: [Select]
Humans  do not generally play as a single species against other sentient  creatures because there are no other sentient species humans are  competing against.            We beat all the other competitors, we are now Cannibots fighting against outselves.
           
           
Code: [Select]
The  different abstraction layers play games of very different rule sets.  Random mutations work on a genomic level game, but would get you killed  quickly in a Risk style game.            Aye, but if you played against Shakespeares infinite monkeys you would lose eventually.
           
           
Code: [Select]
What's interesting here is that "winning" isn't selected either for or against            You lose, I win. Nuff said.
           
           
Code: [Select]
A beetle that pushes out all the beetles from its niche cannot ever suffer non existance            Apart  from the fact it can no longer reproduce and theirfore dies. If it is  homosexual? it might but if a beatle species removes all other beatle  species from its niche then it is immmortal 'till that niche doesn't  exist or an outside influence/species/catastrophe removes it from its  niche.I agree!
           
           
Code: [Select]
Traditional natural selection doesn't make sense in my ex nihilo sims.  
            There isn't reproduction, so how can there be natural selection?
           Reproduction is the beginning of Natural selection, without self replication you don't have evolutionv surely?
             
   
Code: [Select]
Something is still happening though.  Bots that get themselves killed are still removed from the game.    To be replaced by non mutating (I assume zerobot) non replicating  models of the original bot... Back to Shakespeares infinite monkeys...
     
Code: [Select]
Bots  might interact randomly, but it's not until a bot larns to reproduce  and/or eat that competition really develops between the different  bots      Infinie monkeys. infinite monkeys, how do you think evolution evolved?
     
     
Quote
Mutations will eventually cause a behavior that will kill the bot, and it'll lose.
     Or cause a mutation that wins and it will fill the assigned niche; pre something better evolving.
     
 
Quote
The ultimate winning strategy for this simple game is to become a self replicator.
 
    Surely just the ultimate starting point?
   
   
Quote
This also makes leagues have a certain logic.
   Bless you.  The leagues are a more controlled extension of Zerobot evotution.
   
   
Quote
During this struggle, cannibalism isn't a good strategy
   Killing your species without reason rarely is.
   
 
Quote
Conspec genes would probably be selected for if a league round lasted long enough for evolution to take hold.
 Leagues  should be without mutation, evolution is not. Plus the chance of a  round lasting long enough to allow positive mutations to take place in such a small enviroment are slim to none.
 
  Big Berhas are either  a problem with the evolutionary part of the program or a representation  of the Cambrian? big is better part or evolution coupled with small sim size
 
 
Quote
All in all it's a very useful framework for  explaining alot of results that would seem to contradict the classical  natural selection.
 Or proving it, take your pick.
 
 
Quote
Darwinbots has an abstraction ceiling.
Yes it does, you are never going to see Home sapiens evolving!

Quote
Evolution can't seem to progress beyond competition between individuals.
Neither can DB evolution...

Quote
I think we can raise the abstraction ceiling another notch by adding the ability to specialize.
Yes we can, DB has always moved in this direction, the specialisaton of species is only small atm but it is getting better.

Forgive me being so pedantic.  

PS if you can spot why it's not quoting probperly I would be grateful.

Edit: apparently the number of quote blocks is limited.  I replaced some as code blocks to make it format properly - Numsgil

Edit: The limit on quotes is 10 in one post - Light
« Last Edit: December 05, 2006, 06:52:08 PM by Light »
If you try and take a cat apart to see how it works, the first thing you have in your hands is a non-working cat.
Douglas Adams

Offline Light

  • Bot Destroyer
  • ***
  • Posts: 245
    • View Profile
Natural Selection as an abstraction for game theory
« Reply #2 on: December 05, 2006, 06:22:59 PM »
With regards to altruism my understanding is that it is generally demonstrated between genetic relatives, the stronger the link the stonger the levels of altruism. So an organism is trying to ensure the survival of its genetic variation rather than the species as a whole.
« Last Edit: December 05, 2006, 06:51:22 PM by Light »

Offline Jez

  • Bot Overlord
  • ****
  • Posts: 788
    • View Profile
Natural Selection as an abstraction for game theory
« Reply #3 on: December 05, 2006, 06:39:21 PM »
Ty Nums.

Light; members of a species are closely genetically related, they wouldn't be a species otherwise. I understand your point but that wouldn't stop me, theoretically, rescuing a child about to be killed at the cost of my life, because they weren't directly related. Altruistic moments often aren't preceeded by consideration of genetic diversity or DNA analysis. Consider the person rescuing the cat/dog from being hit by a car.
If you try and take a cat apart to see how it works, the first thing you have in your hands is a non-working cat.
Douglas Adams

Offline Sprotiel

  • Bot Destroyer
  • ***
  • Posts: 135
    • View Profile
Natural Selection as an abstraction for game theory
« Reply #4 on: December 05, 2006, 06:44:19 PM »
Quote from: Jez
Code: [Select]
But  reproduction doesn't really make sense from the framepoint of countries  as a successful strategy. For countries, long term survival means a  strong military, good diplomacy, internal stability, and growth (either  in the economy or in physical borders, etc.).            Why not? If a  country doesn't reproduce it ceases to exist, Germany during WWII had a  medal for mothers that had lots of children.
You've clearly misunderstood Numsgil's point here. Germany didn't reproduce (well, you could say it did, since there used to be 2 Germanies, but one died...), it's the Germans that did. Similarly, I haven't reproduced yet (= I don't have children), but I'd be dead by now if all my cells had stopped reproducing at some point.

Offline Light

  • Bot Destroyer
  • ***
  • Posts: 245
    • View Profile
Natural Selection as an abstraction for game theory
« Reply #5 on: December 05, 2006, 06:48:27 PM »
When I was talking about genetic relatives I was being slightly confusing in that I meant blood relatives, parents children, uncles, cousins etc. I wasn't trying to say that altruism only happens between blood relatives just that it is more likely.

Offline Sprotiel

  • Bot Destroyer
  • ***
  • Posts: 135
    • View Profile
Natural Selection as an abstraction for game theory
« Reply #6 on: December 05, 2006, 06:51:34 PM »
Quote from: Jez
Light; members of a species are closely genetically related, they wouldn't be a species otherwise. I understand your point but that wouldn't stop me, theoretically, rescuing a child about to be killed at the cost of my life, because they weren't directly related. Altruistic moments often aren't preceeded by consideration of genetic diversity or DNA analysis. Consider the person rescuing the cat/dog from being hit by a car.
Human behaviour isn't genetically constrained, so you can't easily compare it with biological examples.

Offline Jez

  • Bot Overlord
  • ****
  • Posts: 788
    • View Profile
Natural Selection as an abstraction for game theory
« Reply #7 on: December 05, 2006, 07:03:51 PM »
Quote from: Sprotiel
You've clearly misunderstood Numsgil's point here. Germany didn't reproduce (well, you could say it did, since there used to be 2 Germanies, but one died...), it's the Germans that did. Similarly, I haven't reproduced yet (= I don't have children), but I'd be dead by now if all my cells had stopped reproducing at some point.
I'm not talking about the country per se, more the poputation of a country. that that makes a country is its population.
Geographical boundaries are defined by (the cultural definitions of) the population.
I haven't any children, if the populace of my country (England) didn't have any children either then England as a nation would disapear (baring imigration or DBs repopulation of dead bots).
When your cells stop reproducing you will be dead, it's one of those things to look forward too.  
If you try and take a cat apart to see how it works, the first thing you have in your hands is a non-working cat.
Douglas Adams

Offline Jez

  • Bot Overlord
  • ****
  • Posts: 788
    • View Profile
Natural Selection as an abstraction for game theory
« Reply #8 on: December 05, 2006, 07:12:53 PM »
I agree Light;

You are more likely to rescue your child that is trapped beneath a car than someone elses. My perception of altruism is slight larger. Anyone (of a decent/alturistic nature) will try to rescue anyone elses child trapped under a car. (+some cross species baby theory I can't remember atm).

Sprotiel;

If we are not genetically constrained (our mind is an evolution of our genetics AFAIK) then what are we constrained by? I thought we were biological entities therefore it is natural to assume that our makeup defines our existance and actions.
If you try and take a cat apart to see how it works, the first thing you have in your hands is a non-working cat.
Douglas Adams

Offline Light

  • Bot Destroyer
  • ***
  • Posts: 245
    • View Profile
Natural Selection as an abstraction for game theory
« Reply #9 on: December 05, 2006, 07:22:47 PM »
I think a lot of it depends on the level of risk to the person being altruistic against how much the other person means to them, as to whether someone does an altruistic act or not, but its probably more of a sub conciesous thought process

Offline Numsgil

  • Administrator
  • Bot God
  • *****
  • Posts: 7742
    • View Profile
Natural Selection as an abstraction for game theory
« Reply #10 on: December 05, 2006, 07:28:23 PM »
Quote from: Jez
The  selfish gene? Altruism is explained there as a species advantage, some  'people' are willing to risk their lives for the propogation of the  species as a whole, one effect one answer is not a survival strategy.

But this doesn't explain it within the context of natural selection.  Natural selection doesn't work on species, as the idea goes.  Species don't have genetic material.  Natural selection works on individuals.  Or if you believe Dawkins and the current paradigm, genes.
           
Quote
Genes make an organism? I don't see how individual genes compete unless it is organism V organism.

This is the idea of Dawkin's "Selfish Gene" work.  Here's wiki on the subject.  Competition is probably too strong a word, but the idea is that genes work to maximize the numbers of themselves.  This involves at the very least a little competition, since DNA length does have an upper limit on physical size.
           
           
Quote
A nation is a niche, a nation is a subspecies (in some ways) no reproduction = no evolution = you die.Why not? If a  country doesn't reproduce it ceases to exist, Germany during WWII had a  medal for mothers that had lots of children.

But Nazi Germany is dead.  The children of those mothers and their decendants are not.  Members of a nation reproducing does not mean the nation is reproducing.  We don't have hundreds of little Nazi Germanies running around Europe.  Countries don't reproduce.  They are born some what spontaneously, the grow and survive for a while, then they die.  Quite independantly of the people in them.  If you play Civ4 for instance, there is only grow, conquer, diplomacy, research and/or death.
           
           
Quote
Homo Sapieans plays the highest abstraction level? If the lower abstracition levels were removed Homo Sapiens wouldn't exist...

Lower levels aren't removed.  They're resolved (meaning you won).  We are the only homonid species alive today.  We have no major predators.  Not only do we not have a species rivaling our niche, we can build new niches as we spread through the magic of fire and air conditioning.  I'd say we won the lower abstraction levels against genetic fidelity from point mutations and niche competition.

It's similar to The heirarchy of needs.  If a new game starts on a rung below where you currently are, the abstraction level collapses.  Civilization would collapse, for instance, if we ran out of food (as it often does, such as Russia before the Bolshevik revolution), civilization would collapse until that food is reinstated.  The group behavior is only possible if we are all winning our game against death.

           
Quote
Selfish genes, it is not the individual it is the species that allows the genes to exist.
Again, species aren't the core evolving agent in classical natural selection.  Species cannot evolve traits.  Only individuals can.  Or that's the idea anyway.
           
Quote
We beat all the other competitore, we are now Cannibots fighting against outselves.
 Exactly.  We won on one abstraction level, and we moved to the next one up: infighting between groups of humans.  Each higher abstraction level represents a splintering of the sole survivor from the rung below.  But generally this splintering changes the rules of the game.
           
Quote
Aye, but if you played against Shakespeares infinite monkeys you would lose eventually.

If the game is to write works of Shakespeare, I would win because I can write English and make sentences and have even outread some of those monkeys   If somehow the resources were limited, so that every book I write damages the monkeys, eventually I'll outcompete the monkeys.  I can write better than a bunch of monkeys ( I hope!)  Eliminating the monkeys means I never have to fight them again, since they'll be dead.  I'll win this abstraction layer, and a new one might form somehow from me splintering against myself.

This analogy doesn't really work I think  
           
Quote
Apart  from the fact it can no longer reproduce and theirfore dies. If it is  homosexual? it might but if a beatle species removes all other beatle  species from its niche then it is immmortal 'till that niche doesn't  exist or an outside influence/species/catastrophe removes it from its  niche.I agree!

Exactly.  What I'm trying to build is a framework for understanding why things work the way they do.  The beetle species would probably further speciate.  Why?  Because they've advanced to a new abstraction layer.
           
Quote
Reproduciton is the beginning of Natural selection, without self replication you don't have evolution surely?

Indeed.  This is what got me thinking.  What's happening in an ex nihilo sim with no reproducers?  Is it entirely random?  No, because some bots will die.  All bots will eventually die in fact.  The odds are 100% that a point mutation will eventually kill them.  But some bots last longer than others.  Eventually an old bot might learn to reproduce and avoid the local problems of point mutations.  Its DNA has won the most basic abstraction layer and can now maintain a high fidelity.

It "evolved" in the sense that it adapted to its environment.  But there was no reproduction except as a "win" condition.  What do you make of this?  By definition it didn't evolve, and was not involved with Natural selection.  But something did happen.  Order did develop from chaos.  Thinking of it as a game gives a framework for it to make sense.  It doesn't make any sense using just Natural selection, since it's out of natural selection's problem domain.
             
Quote
To be replaced by non mutating (I assume zerobot) non replicating  models of the original bot... Back to Shakespeares infinite monkeys...

Replaced yes, but not non mutating.  It's replaced by a younger bot.  The inevitable appearance of a replicator, and a replicator's inevitable conquering of time (replicators will continue to exist into perpituity) only makes sense if you think of it as a game (again, a game as in game theory).
     
Code: [Select]
Infinie monkeys. infinite monkeys, how do you think evolution evolved?I'm not disputing evolution.  I'm expanding its problem domain, abstracting it away from the processes inherant in life and trying to show it as a natural and very, very simple part of existance.  Things that don't survive don't survive.  For any slice of time quanta, we will see things around us that possess the capability to stave off entropy and death, to varying degrees.  Things that instantly die don't exist in great numbers.

It's a very simple premise that puts natural selection as its understood into a large framework, from which conclusions can be drawn.  It's a proto theory.
     
Code: [Select]
Or cause a mutation that wins and it will fill the assigned nich; pre something better evolving.
This is a subtle point.  Winning isn't really the objective.  Things happen when you win, sure, but your goal isn't to win.  You're just surviving.  If you "win", and conquer the present mode of death, you are upgraded to a new game with new rules and a new way to die.

There's always another way to die.  Death is strangely universal.
     
 
Code: [Select]
    Surely just the ultimate starting point?
Of the next abstraction level, yes.  But for the game of "survive point mutations", you've reached the end.  That there's been life on Earth uninterrupted for billions of years attests to that.

   
Code: [Select]
Killing your species without reason rarely is.
This is less true than you think.  Evolution sims always have the conspec recognition genes break, causing cannibalism.  Always, always, always, except maybe under extremely specific conditions.  The environment just doesn't favor this.  Yet if you are cannibalistic, it hurts you in the leagues.  How do we resolve this?  If you view it as layers of games, it makes perfect sense.  Why should a strategy that worked in checkers work in risk?  They are new games, with new rules.
   
Code: [Select]
Leagues should be without mutation, evolution is not. Plus the chance of a round lasting long enough to allow positive mutations to take place in such a small enviroment are slim to none.
Right.  But lets imagine for a moment that you set up a competition between two species that takes so long that mutations and evolution start to kick in.
 
Code: [Select]
Big Berhas are either a problem with the evolutionary part of the program or a representation of the Cambrian? big is better part or evolution coupled with small sim size
Undoubtedly if our goal is to create the stability real life has (and it is) this is a problem.  But it's also a fact.  A valid datum.  Evolution can lead to extinction even in Muller Ratchet free universes.  Natural selection works with generations.  How do you resolve (and by resolve, I mean include) the idea of big berthas into the idea of natural selection.  You could call them an aberration or pathological case, pat yourself on the back and move on, but isn't that really missing the point?

Code: [Select]
  Or proving it, take your pick.
They don't contradict.  Classical natural selection is a subset of the abstract natural selection.  Like special and general relativity.  My point is: does natural selection as we understand it immediately tell you about the possibility for altruism?  What about sentience?  We can explain these things as arising from natural selection, and certainly it does, even if indirectly, but does that help you make predictions about what's going to happen?

It's like explaining a string and pulley using quantum mechanics.  Undoubtedly the properties of the string and pulley arise from the properties of quantum mechanics, but that doesn't help you build or design a pulley system, or predict easily what one will do.
« Last Edit: December 05, 2006, 07:29:53 PM by Numsgil »

Offline Numsgil

  • Administrator
  • Bot God
  • *****
  • Posts: 7742
    • View Profile
Natural Selection as an abstraction for game theory
« Reply #11 on: December 05, 2006, 07:35:30 PM »
Certainly altruism arises because you're helping your fellow mates.  Sheep generally don't decide to be altruistic towards wolves and give themselves over.

But that doesn't answer the question of why it doesn't "devolve" back into selfishness.  If my cousin saves my life at the cost of his, that's good for me.  If I let my cousin die to save myself, that's also good for me.  I'm "cheating" for profit.  Clearly there's incentive to do so.

Why don't the cheaters take over?  Why doesn't the system of altruism collapse?  It's an easy phenomenon to reproduce.  Heck, it happens all the time in Darwinbots.  Why are their still altruists in nature?

If you look at it as a strategy for personal genetic victory, it makes no sense.  If you look at it as a strategy for a game that's another abstraction layer up, between species or groups within a species, it makes perfect sense.
« Last Edit: December 05, 2006, 07:36:37 PM by Numsgil »

Offline Light

  • Bot Destroyer
  • ***
  • Posts: 245
    • View Profile
Natural Selection as an abstraction for game theory
« Reply #12 on: December 05, 2006, 07:48:27 PM »
If you play one round of prisoners dilema the best option is to cheat. But if you play many rounds and you start cheating the other person retaliates by cheating and you both lose. A few years back someone organised a competition for people to submit programs to play prisoners dilema, the one that did the best was tit for tat, that basically followed what the previous programs actions. In nature its not just about one altruistic action between 2 individuals, its about the interaction of many. If you cheat someone, other people will see that and act acordingly. There were some examples in the selfish gene, I think of bats sharing food to others that hadn't manage to catch anything that night and that they would remember bats that hadn't shared, and not give them food when they were going hungry. And I think there was also another one about birds and ticks. I guess its kind of regulation by the group as a whole.

Offline Numsgil

  • Administrator
  • Bot God
  • *****
  • Posts: 7742
    • View Profile
Natural Selection as an abstraction for game theory
« Reply #13 on: December 05, 2006, 07:52:51 PM »
I don't think all altruism in nature is tit for tat.  Certainly some of it is, but not all organisms have the capacity to remember the past actions of the other members in its group.

This is a good article about this very topic! Article.

Offline EricL

  • Administrator
  • Bot God
  • *****
  • Posts: 2266
    • View Profile
Natural Selection as an abstraction for game theory
« Reply #14 on: December 05, 2006, 08:23:36 PM »
I'm going to have to smoke some ganja or something and reread Num's post several times to comment fully on Num's therory, which I will do as soon as I'm done coding.  But to the specific topic of altruism, I suggest peopel read or re-read the definite work on the subject, Dawkins Selfish Gene.  

Genes, not organisms, are the unit of selfishness in the Dawinian sense of the word.  Genes are selfish.  They compete with other alleles.  They cooperate with other genes where cooperation benifits them.  One result of this cooperation is in building gene machines called organisms.

One obvious strategy for selfish genes to use to maximise their survival is to program their gene machines, the organisms in which they find themselves, to be selfish.  There are many examples where survival of the invidial organism favors survival of the genes that ride inside it.  But different circumstance sfavor different tactics.  There are many other circumstances, not particular rare, in which genes ensure their own survival by influencing organisms to behave altrusitic.  Favoring genetic kin in which the probabiltiy is high that those gene machines contain copies of the same genes is a well known and well studied strategy for maximizing gene survival and frequency in the gene pool in which it finds itself and in which it competes.

Reciprocal altruism is the other well known and well studied type of altrusim with a well worked out Darwinian rationale.  Robert Trivers first worked this out and it is often expressed in the mathematics of game theory.   Symbiosis is an example of reciprocal altruism between species.

In highly social animals such as humans, altruistic behaviour can also play a role in sexual selection.  That is,selfish genes have found that encouraging their robots to be altruistic in certain circumstances increases their survival and frequency in the gene pool by increasing such things as social reputation.

Cheating is only stable at a certain frequencies within the population.  The 100% "always be selfish" is stable but so are other strategies such as "by nice, reciprocate niceness, punish badness".  In game theory, these are often given names like Tit-for-Tat or Retaliator and Reciprocator.  As an example, in a population dominated by altrusistic reciprocators who punish badness, no exclusivly nasty individual will do well.
Many beers....