from 2007. Long but worth the watch. Let’s discuss below?
176 thoughts on “Polyworld”
Leave a Reply
You must be logged in to post a comment.
The Skeptical Zone
"I beseech you, in the bowels of Christ, think it possible that you may be mistaken."
from 2007. Long but worth the watch. Let’s discuss below?
You must be logged in to post a comment.
Rich,
Long, indeed! I did start watching but couldn’t get past the intro. Any chance, having watched the presentation, you could maybe summarize some of the content or link to a meaty bit or two?
Haha, that’s a good one Alan!
And here I thought no one at theskepticalzone had a sense of humor.
Hilarious!
Not worth the watch unless you watch it.
It’s a simulation of both morphology / instincts and brian wiring. Fitness function is survival and reproduction. Some interesting emergent behaviours and also information theory used to measure the complexity of the virtual critters.
Perhaps a certain Gary should take a look.
OK, I’ll give it another try and skip the intro.
I like the bit where Griffith confirms there is no way to quantify intelligence (about 40 minutes in)
Alan Fox,
I like the bit where the speaker tries to complete one coherent sentence. Its such an interesting challenge. So Richard-like.
phoodoo,
Aw bless, Phoodoo. Your lack of comprehension is not his lack of coherence.
Evidence, please. You haz link to codez?
You see, we have to remain skeptical, because people who post here say their program measures fitness but when we look at the actual code we discover it assigns fitness.
I’m probably just dumb. No doubt keiths will be along soon to confirm that, what with his tremendous mind-reading skills.
But if “the fitness function is survival and reproduction,” why does one even need a fitness function? All one needs is a random sampling function.
Isn’t that how evolution really works?
🙂
True. I have not wached the video. Ring one up for our resident retail clerk!
Does the video provide a link to the code?
Which of my questions do you think you’ve answered, and why?
If some members of a population leave more offspring than other members of that population, why can’t you just take a random sample from the offspring?
Who needs a fitness function if you have survival and reproduction, unless it is the case that without that fitness function all members would produce offspring according to the same probabilities?
Maybe Richardthughes just doesn’t understand evolution or simulations that claim to use evolution. But this is The Skeptical Zone, so no one is likely to believe that!
LoL!
Are you dense? A quick google will provide you with multiple open source options for this. Not that you’d be able to comprehend it. You’re becoming JoeG with dipshit go-to lines: “Do you have the code”, and Phoodoo for opinions on things you’ve not even examined. *Golf clap*
Fitness can be stochastic. http://theskepticalzone.com/wp/evodice-part-1-fitness/
Sorry, I thought it was obvious. Are you just now figuring that out?
That wasn’t the question. I was not doubting that if every member of a population left the same number of offspring (on average) that they were all equally fit.
Why do you need a fitness function if fitness is stochastic? Why do you need a fitness function at all? Just take random samples.
Unlike me, you are not dense. So you will answer. Right?
You really don’t understand anything, do you? How are we to determine which organism get to survive and procreate?
I understand that I am posting at TSZ, not at UD. Perhaps I’m not as dense as you think. 🙂
Who cares? If all the organisms leave the same number of offspring then it’s all up to chance.
Even if some organisms leave more offspring than others you can still randomly sample and their offspring will be more likely to appear. That’s how evolution works.
There’s no need to determine which organisms get to survive and reproduce. Unless you’re writing a WEASEL program.
Let me know if you need me to explain that to you.
Mung,
I’d love your explanation Mung, because I suspect you have some hilarious misconceptions. Have at it, be expansive. I have my needle and thread ready for my sides.
Intelligent Design. Isn’t it obvious?
From 2007. So almost certainly false.
Imagine that you encounter an evil Creationist quoting from something written before 2007.
Tragic trollery. Trying to help you is a waste of time: http://theskepticalzone.com/wp/complexity-explorer/#comment-111990
Let us know when you’re finished Old Munging.
You never attempted this:
Is is being wed to your ideology, or mental limitations or something else that stops you learning? This is not a good forum for your personal development, Mung.
OMD he said “cheating!:
https://github.com/polyworld/polyworld
Yes, Frankie/JoeG, yes.
Why don’t you write a sim and find out! Or is that too much like actual work for you FMM, oh sorry, Mung.
You say that presumably as a joke but are in the tent with people who seriously believe that is a legitimate answer. Perhaps you should save your mockery for those people. Your people.
Its worse than this Mung. The speaker actually gives it away unknowingly. he makes the same fatal flaw in thinking that we see from Joe, and Alan and DNA Jock.
The speaker says, there has to be a limit, just like there has to be a limit in evolution, as to how long things can live and reproduce, or else you might get the unfortunate result of the least fit individual reproducing the most!
Get it? There is a definition of fitness which is separate from those that reproduce the most. That definition can vary according to the whim of the programmer. In this case the most fit seems to be the ones who give them the designs they feel are most clever apparently.
Same old story, just another search function as always.
phoodoo,
Why don’t you type out exactly what he said so we can compare to your misrepresentation above?
That’s the same thing. How does it assign fitness? By measuring reproductive success.
That’s what it means to measure fitness. You have a given definition of it, and then you observe an organism and collect data and once you have this data (how many offspring does it produce on average), you assign a fitness-value to the organism. The observing-part is what you would call a measurement.
In a simulation, the simulation is directly producing the data, so it doesn’t have to be “observed” in the same sense. Once this data has been obtained, it can be used to assign fitnesses to organisms.
This makes literally no sense. It is literally impossible, by definition, for the least fit individual to reproduce the most.
Because the degree of fitness IS A MEASURE OF PREPRODUCTIVE SUCCESS. That would mean that, in so far as it is reproducing the most, it is the MOST fit organism. Why? Because that is what fitness measures, reproductive success. The greater reproductive success, the more fit the organism. So the most reproducing organism is the fittest one.
You are probably making the basic mistake of conflating fitness (reproductive success) with ability to survive under harsh and changing conditions.
Yes Mung, well done. That’s the two prongs of drift and selection. In small populations, drift can overwhelm selection.
But once you get into the tens of thousands to millions in population size, the variation you get from random sampling will be overwhelmed by selection.
Rumraket,
Oh my goodness Rumraket. You are amazing.
You have reached the level of understanding of Richard.
Moved one comment to Guano. Please address the content not the character.
If I understand Richard correctly, his question is in the context of a simulation. You’re correct that a random sampling of offspring resulting from differential reproductive success will model what we observe in biology, but there still needs to be a way to determine the relative rates of reproduction.
In a standard EA there most certainly is.
Please share your alternative algorithm that models known evolutionary mechanism.
Probably?
No.
Patrick,
There aren’t any that model unguided evolution, ie natural selection and drift. All EAs model evolution via intelligent design.
Is it then, in your view, even possible to write a simulation of unguided evolution?
So far we have one person who says no.
But I think Rumraket got this one right and Robin is wrong.
In a simulation you can program in a goal or ideal and a way to measure how close to or how far a genotype is to/from that goal or ideal and assign (not measure) a fitness value that is then used to determine contribution of offspring to the next generation.
You don’t measure proportion of offspring and use that to assign a fitness because you don’t care about it after the fact, you care about it before the fact, so that you can actually influence how many offspring a particular genotype leaves.
Let’s see evolutions code!
I don’t know why you think this would pose any sort of challenge for one of my obvious talents. But first I’d need a rigorous definition of evolutionary algorithm. Got one?
Evolution – change over time.
Algorithm – a fixed and immutable series of steps to accomplish a goal or purpose.
Looks to me like the phrase “evolutionary algorithm” is an oxymoron.
Do you want me to write an algorithm that changes over time and has no purpose or goal? I think I am uniquely qualified to do that.
I don’t think Frankie does follow up questions….
Well, maybe you could write a simulation of the evolutionary process then, if the word ‘algorithm’ is giving you troubles.
I would say however, that you can write an algorithm that immutably and unchangeably follows a series of of steps, and those steps involve:
1: Randomly generate variation in a simulated organism.
2: Test that variation in a simulated environment.
3. Use the most sucessful variants (those that leave the most offspring) as a basis for the next round of step 1.
4. Repeat.
The algorithm itself is fixed, it always proceeds according to these steps. It doesn’t have any end-goal. It’s not trying to reach any particular outcome. The algorithm does not try to alter itself, so it IS an algorithm. Yet it will be simulating change over time without any set goal.
No, evolution and agorithm is not intrinsically contradictory. The individual components of evolutionary change can be simulated algorithmically. See how that works?
By the way, regarding the definition of algorithm Mung supplied:
“Algorithm – a fixed and immutable series of steps to accomplish a goal or purpose.”
In my above post, the “goal or purpose” of the algorithm is not to evolve anything in particular (so the simulated evolutionary process would not have any goal, like for example a WEASEL-type program has), it is merely “to simulate a goal-less evolutionary process using a set of fixed and immutable steps”.
So there you go. Such programs have been written by the way. One of them is called Avida.
The calculation of fitness is a measurement. Storing the results of that calculation is an assignment in the context of a program, but that’s just an implementation detail. I’m not sure what your point is here.
Differential reproductive success is essential to evolution. The relative reproductive success of each member of a simulated population must therefore be calculated. This calculation is with respect to a fitness function. The word “fitness” in that name is related to but not the same as “fitness” measured by counting offspring. If it’s just the word that bothers you, we can call it the differential reproductive success function instead.
So if a differential reproductive success measurement is used to determine the relative odds of each member of a population reproducing and then the number of offspring is used to calculate fitness, that would be an acceptable algorithm to you? If not, how specifically would you code an EA to meet whatever criteria you feel are so important?
Patrick,
In such a scenario, which entities are the most fit: those that produce the most offspring, or those with the highest reproductive success calculation?
That depends on the definition of fitness you’re using. Since reproductive success is stochastic (depending on the selection model), it’s possible that an individual with a higher relative reproductive success measure might leave fewer offspring than one with a lower measure. In that case the two definitions could yield different answers.
Patrick,
What two definitions?
The relative reproductive success measure and the biological relative number of offspring measure.
It’s possible with both in silico and real world environments for the best, most shining, Harrison Bergeron of a population to fail to leave any descendants. That individual may have the shortest traveling salesman path or the biggest muscles or the fastest speed and still get caught in a (virtual) tar pit. Nonetheless, it’s not unreasonable to consider that individual the most fit in that population.
It’s also not unreasonable to consider Harrison’s neighbor who hung around the fertile females while Harrison was out being fit and hence left far more descendants in the next population as most fit. It all depends on how you want to use the information.
So, what is your point?
Over many generations, it’s what alleles and genes become prevalent or fixed in a population. In sexually reproducing populations, it is possible for a sickly individual to have many descendents and for a healthy individual to leave none.
It is possible to observe that a certain allele confers some benefit — an objective benefit — but go extinct anyway, for any number of reasons. War, disease, giant meteors.
Most viable individuals, given favorable circumstances, can produce more offspring than necessary for replacement. I suspect that the only objective measure of fitness is how close an allele is to fixation.
A different definition for every scenario.
Of course not.
Do you know how fitness is measured in the real world?