Survival of the fittest

[Jez]

Here's a thought;

After a recent discussion following the post about Richard Dawkins and why bots devolve instead of evolve, plus a recent discussion about how the 'best bot' selection button works why not have an option to kill the worst bot? It seems to me, in my blissful cloud of ignorance, that enviromental pressures have a lot to do with the successful evolution of 'fit bots'.

While the intention is to add various enviromental pressures to the sims, these are perhaps a long term goal requiring a fair amount of work. Would it not be easier, certainly in the short term, even if we allowed for selection of what defined the 'worst bot' in the same way that there was talk of being able to define the parameters for the 'best bot', to allow for an arbitary cull of those misbegotten and unwanted offspring?

Perhaps a time or population controlled cull?

I realise that this might tread the icy thin waters of eugenics but it does, at this point in space, seem a reasonable way to emulate enviromental pressure without needing an pro-active or aggressive enviroment.

[shvarz]

There is really no reason to kill "the worst bot", because if it is indeed the worst, then it will die off anyway.

In addition, even in theory there is no way to define which bot is the best or the worst.  The only truly objective way is to look at the "end of time" and see whose offsprings survived that far and whose did not.  But at that point all the bots are dead and there is no need to kill anyone

[Numsgil]

A relatively simple way to define the "best" bot is the bot with the most loaded nrg in itself and its surviving offspring.  This is a relatively easy way to balance big berthas and cancerous bots.

But that doesn't help when you're selecting for "worst" bot.  Instead of selecting and culling a "worst" bot, I would have a disaster event, with two kinds of disasters:

1.  The first kills off any bot within some random location and radius.  This would simulate local disasters, such as a volcano.

2.  The second jacks up costs for several cycles (say, 100 cycles).

The idea is that the first kills off K-selected species while the second kills off r-selected species.

That sort of random death should provide the selective pressure to eliminate poor phenotypes.

[Zinc Avenger]

I've often tried to think of ways of artificially (or perhaps semi-arbitrarily) increasing the fitness of a population of bots without editing dna, but this catastrophe-style method seems a bit harsh. Personally, I think bots have a hard enough time surviving as it is, and superior fitness is no guarantee of survival, so adding in new ways to kill them at this stage could be counter-productive, leading to selection of the lucky rather than the best. Oh, and yes, I do know that's how it works in real life!
 
My current leading idea, which is still in the early stages of consideration, is to scale each bot's point mutation rates by its unfitness (as fitness decreases, point mutation rates increase). Bots at the low-fitness end of the population get more point mutations, which will lead the bot to either break down altogether and die, or just maybe evolve that little step necessary to bump its fitness up.

Also under this scheme, the fittest bots will not point-mutate away from their current level of fitness. However, as the population's overall fitness increases the earlier high-fitness bots would find themselves dropping down the fitness tables for the population and start to mutate.

As I said, this idea hasn't been given my usual week of casual pondering before it gets its airing so please forgive any ragged edges!

Of course, the simplest method for artificially increasing the fitness would be to duplicate the current leading bot every 1000 cycles or so, but that could lead to bots losing their .repro commands when they become fit.

[EricL]

I feel compelled to point out that in any evo sim where selection is in operation (zerobot sims where a first replcator has yet to emerge and take over don't count) ALL of the extant bots are likely to differ from each other by only a few base pairs at most with many of them being genticially identical.  They are all the same species.  The whole concept of determining "best" is rather silly in such populations - they are all 'best' in that they are all alive and relative to the vastly larger number of extinct ancestors, they all are fittest.   If you just want to select viable representitive individual, all you really want is a bot that didnt mutate that generation (since the probablity a mutation is deleterious is far greater than it being beneficial).  Other that that, one bot is really no more fitter than another - there just isn't enough genetic difference for the concept to be meaningful.  Fitness is a species-level (a genome-level) concept, not an individual-level concept and in a sim where there is essentially only a single genome (ignoring autotrofs) all bots are pretty much equally fit.

I have yet to see someone evolve a stable sim which supports multiple heterotopic species in proximity, where choosing one genome as 'fitter' than another might make some sense.  I'm using the term 'species' here in it's genetic sense, meaning two bots are of different species if they exhibit significant genetic distance in their genomes, even if they share a common ancestor.  Yes, in an asexually reproducing population, every bot is in some sense it's own species, but our sims are generally too small and too simple to support an acestrial line diverging from and yet coexisting with other extant lines.   Said another way, in every sim I've looked at where there isn't forced, physical isolation, ALL the bots in the sim share a common ancestor that is a surprisingly small number of generations in the past.  Note that I said 'stable' and 'in proximity'.  Yes you can get specieation when there is isolation, but that disappears when they are brought into proximity.  I've yet to see a true ecosystem where mutlple species in proximity have some sort of stable, niche-determined relationship which allows speciation in proximity to be stable (I.e. one is preditor, one is prey or perhaps one is herbavore and one is scavenger, etc.).   Until such time, the whole concept of fittest is suspect IMHO.

[Jez]

There is really no reason to kill "the worst bot", because if it is indeed the worst, then it will die off anyway.

I wish that were true, but it's a petri dish experiment, a eutopian existance where manna pops into existance if somebodies pigged the last meal. There are no sharks or tigers waiting in the wings to semi-selectively cull the populations. If my definition of 'worst bot' included cancerous/cannibal/sterile berthas then they are quite unlikely to die off.

In addition, even in theory there is no way to define which bot is the best or the worst.

I can think of definitions of bad behaviour that might have a positive impact on evolution;
*must have 1 baby every x cycles
*must obtain x amount of nrg every x cycles
*must not gain more than x amount of nrg from friends every x cycles

I would have a disaster event,

Could be a good idea, it's not a herd facing the selective pressure of being ambushed by a pride of lions though. I was just trying to think of a way to increase those selective pressures that most animals face for our bots, rather than the unselective pressure that the inhabitants of Pompei ended up facing.

scale each bot's point mutation rates by its unfitness

Interesting idea, does face the same problem that shvarz suggested selecting for worst bot faced. If it is going to be a problem to select one worst bot how are you going to have a sliding scale to rate a population? I would like to see the affect that an idea like that had on a population.

[EricL]

If my definition of 'worst bot' included cancerous/cannibal/sterile berthas then they are quite unlikely to die off.

DB is all about natural selection, not artifical selection thus there is only one judge of fitness and that is survival and reproductive success over time.  Our own definitions or preconceptions of which adaptations we like or dislike, expect or don't expect is unimportant.  Selection will favor whatever strategy works.

Cancerous population explosions and sterile berthas are temporary aberations, usually due to a single deleterious mutation in a reproduction gene.  I say deleterious because they ultimatly terminate the ancerstrial line.  Sooner or later, they will die off without descendents, the former due to nrg exhasution, the latter as soon as something evolves to take them on.  (If you don't want to wait for berthas to get preyed upon, try a age cost that ramps up slowly.)

As far as canibalism, why shouldn't selection favor it?  Kin selection only matters when there is diversity within a population.  Most evo sims have very little (see my above post) thus the evolutionary value of guaging kin relatedness is close to zero - all bots are close relatives.  We only perceive cannibalism as abnormal because we start our sims with hand-coded bots that have a conspecific recognition gene and then we act surprised when this gene gets selected against.  What woud be truly interesting is the evolution away from canniblism in a population 100% evolved from zerobots. This would require the evolution of some notion of species and selection for conspecfic recognition.  Now that would be something.

[Jez]

I understand, probably not the best idea I've ever had   it would be artificial selection. I tend to think of DB as a petri dish experiment, there aren't that many outside pressures on the bot, very often they aren't forced to follow narrowly defined survival strategies. I was considering 'worst bot' as a way to focus bots mutations onto a more closely defined path.

Other methods of creating outside pressure on the bots would be better.

[Numsgil]

I would say that "fitness" isn't something that can be defined particularly well by a 1 Dimensional number for a single individual of a species.  Fitness seems to be a multifaceted vector that defines nodes in the phylogenic tree.  Any fitness calculations really only make sense to me as a way to measure positions in a phylogenic tree against a very specific metric.  Having bots evolve against these metrics is ultimately not going to be a fruitful excercise.  The bots will twist the rules to their benefit, but not in ways we'll find particularly meaningful.

However, I do see great merit in providing stronger selective pressures.  Bots have it too easy, and it shows.  Ultimately, I think we should introduce the idea of a "random carnivore".

Every X cycles, the program loads up several individuals from various league positions into the program.  They'll go around wrecking up the place for, say, 1000 cycles, or until 70% of the evolving bot population has been wiped out.  Then they'll all die, and their corpses will float around for food.

Ultimately I think an evolving population needs some extreme selective pressure at some extremely rare times in its history in order to shake things up enough to take the next step.  See Toba catastrophe theory, which points to a large scale die back event that occurs at virtually the same evolutionary moment that Homo sapiens sapiens arrives on the scene.  Coincidence?  Well, the data is pretty flimsy still so maybe.  But the idea stands as plausibly valid.  The key is punctuated equilibrium, and you need something to do the punctuating.  Something big!

As I've said before, I think it tends to be more the unsurvival of the unfittest instead of survival of the fittest.  The rewards at the top are less rewarding than the punishments at the bottom are punishing.  Killing off huge portions of the population from time to time is important in pushing a species forward.  Or backwards.  Or anywhere other than where it is at any rate.

[Jez]

The unsurvival of the unfittest instead of survival of the fittest.  The rewards at the top are less rewarding than the punishments at the bottom are punishing.

I like that! Also the idea of punctuated equilibrium as a form of evolutionary pressure.

Metric pressure = bad, dropping a 'pride of lions' on bots heads every now and then = good!  

Isn't it true that every time we have had a 'new species explosion' it has followed a large scale die back?

I am also familiar with the theory of species extinction following Homo Sapiens spread across the planet, and continuing, unfortunately, to this day. Homo Sapiens is a natural catastrophe, Gaea maybe shot herself in the foot by not having a bigger 'Toba Catastrophe' and wiping us out a bit more completely!

[Numsgil]

Haha.

Mother Earth knew her mistake and tried to do us in with a giant volcano strategically placed for maximum damage.  Unfortunately her plan backfired.  "Is that all you got?"

[EricL]

Population bottlenecks in sexually reproducing species work because they reduce genetic diversity.  In a smaller population, alleles fixate.  Diversity is reduced.  Where before there were blue eyes and brown eyes, now everyone has blue eyes.  Any subsequent evolution on that reduced population starts from a less diverse gene pool.  Any new allele that crops up for a gene due to a mutation has fewer alleles to compete with.  The probability that it will be advatageous relative to the single or few remaining alleles for that gene is larger that it would be if it had to compete against many alleles.  Speciation accelerates relative to a larger population.

But this is only inportant for sexually reproducing organisms.  It is the constant geneflow between individuals and sub popualations within the larger sexually reproducing population that keeps multiple alleles from fixating in the first place.  Bottlenecking asexually reproducing organisms doesn't mean anything since there is (usually) no gene flow between them.

I'm all for adding new and different types of selective pressures but it won't do much if there is no diversity in the sim upon which selection can operate.  We first need to get to the point where geneticly different organisms can coexist for a long time in proximity in the same sim.  We must get to the point where every hetertropic organism in the sim doesn't share a recent common ancestor.  

To this end, I think what we really need is a richer environment that offers more niches to occupy I.e. rocks to hide under, lakes to swim in, richer morphs so we can get parasites, symbioates, etc.   It will take time and cpu cyles and larger sims with more organisms, but bottom line, we are lacking environmental diversity.

[Zinc Avenger]

Well then, what we need is a bunch of teleporter-linked sims with different environmental settings. I kept running into "error 76, path not found" after ten minutes or so with teleporters, so I'm currently trying some troubleshooting before I submit a bug report. But if teleporters work for anyone else, then perhaps that would be a good avenue to pursue.

[Numsgil]

I get that error too when I use my Darwinbots.exe autorunner utility.  It must have to do with Hard Drive IO or something along those lines.