Evolution Not Random

[mijopaalmc]

That's very nice. But, what EVERYONE wants to know, and what you conspicuously refuse to address, is whether your probability distribution for evolutionary change looks like this:

[qimg]http://upload.wikimedia.org/wikipedia/commons/thumb/d/d6/White-noise.png/180px-White-noise.png[/qimg]

...or like this:

[qimg]http://upload.wikimedia.org/wikipedia/commons/thumb/8/8c/Standard_deviation_diagram.svg/350px-Standard_deviation_diagram.svg.png[/qimg]
...or like something else -- and if so, then what???

Why does it matter?

To humor you:

The probability distribution for natural selection (like that for mutation) is probably the Poisson distribution:

 

[Fredrik]

Actually, that is one of the most persistent myths perpetuated by those who insist that evolution is non-random. While it is true that most real-world data has uncertainty, the variation in the final conditions real-world deterministic (or chaotic) systems, as well as the models that describe them, derives completely the variation in initial conditions, whereas the the variation in the final conditions real-world stochastic (or random) systems, as well as the models that describe them does not necessarily derive from variation in initial conditions.

This was false the other times you said it, and it is still false. Those of us who keep telling you that are not the ones who say that evolution is non-random. We are the ones who say that evolution has two components, one mostly random and one mostly non-random, and that it doesn't make sense to dumb it down beyond that. I personally disagree more with Articulett than with you, because some of her arguments have been irrational. Yours are just wrong. (Not all of them of course, just the most important one).

There are no deterministic systems in the real world. The reason is not that some systems exhibit determinstic chaos. (You seemed to believe that that was what I meant earlier). The reason is that quantum mechanics guarantees that identical initial conditions lead to different results, which is how you define random.

 

[kjkent1]

Why does it matter?

To humor you:

The probability distribution for natural selection (like that for mutation) is probably the Poisson distribution:

http://upload.wikimedia.org/wikiped...on_PMF.png/800px-Poisson_distribution_PMF.png

Okay, so based on your statement above then, (1) what does the x and y axis represent for the mutation distribution, and (2) same question for the selection distribution?

 

[mijopaalmc]

Okay, so based on your statement above then, (1) what does the x and y axis represent for the mutation distribution, and (2) same question for the selection distribution?

x-axis: the number k of mutations (or offspring) an individual of a specific phenotype might have

y-axis: the probability p that an individual of a specific phenotype will have k of mutations (or offspring) in a generation given the average number of mutations (or offspring) λ for that phenotype

 

[Belz...]

Actually, that is one of the most persistent myths perpetuated by those who insist that evolution is non-random. While it is true that most real-world data has uncertainty, the variation in the final conditions real-world deterministic (or chaotic) systems, as well as the models that describe them, derives completely the variation in initial conditions, whereas the the variation in the final conditions real-world stochastic (or random) systems, as well as the models that describe them does not necessarily derive from variation in initial conditions.

Mijo, it's not a myth, it's YOUR DEFINITION. If you don't like the consequences, change your definition.

 

[Skeptic Ginger]

And from my perspective, the problem isn't that there is nothing random IN the process of evolution. The problem is in describing the overall process as random. It is not.

 

[Skeptic Ginger]

post deferred pending more precision

 

[Meadmaker]

And from my perspective, the problem isn't that there is nothing random IN the process of evolution. The problem is in describing the overall process as random. It is not.

It's at least as random as welding, and I get paid to call welding a "random process".

 

[mijopaalmc]

There are no deterministic systems in the real world. The reason is not that some systems exhibit determinstic chaos. (You seemed to believe that that was what I meant earlier). The reason is that quantum mechanics guarantees that identical initial conditions lead to different results, which is how you define random.

Really? How so?

 

[cyborg]

It's at least as random as welding, and I get paid to call welding a "random process".

You get paid to correct errors.

 

[Fredrik]

Really? How so?

I have explained it at least twice already, and I think Sol has too, but OK, one more time...

Let's consider the gravitational interaction between three approximately spherical objects, e.g. a star and two planets. We would normally describe this system with a classical theory that exhibits deterministic chaos. ("Deterministic" because the final state is completely determined by the initial condition, and "chaos" because small variations of the initial conditions would lead to very different final states). This theory is accurate enough for most purposes, and it's not "random" by anyone's definition.

However, in an actual physical system consisting of a star and two planets, the final state is not determined by the initial condition. As I said a couple of times before, the decay of a single uranium atom (after the time we consider "initial", but before the time we consider "final") would be enough to change the final state into something completely different. So in this real-world system, the final state is not determined by the initial condition. This makes it "random" by your definition.

The same argument holds for any physical system (not just the chaotic ones). Quantum mechanical processes will always push the system off the path through phase space that the (deterministic) classical theory predicts. In chaotic systems, this will change the final state a lot. In systems without chaos, the final state will only change a little, but that doesn't matter because you defined "random" to include every system that has more than one possible final state for each initial state. So even systems without chaos in the classical description (i.e. one planet orbiting a star) are "random" according to you.

 

[mijopaalmc]

Fedrik-

My point is that, in a deterministic system, if one starts with identical initial distributions*, one will always end up with the same final distribution, whereas in a stochastic system, if one starts with identical initial distributions, one will not always end up with the same final distribution.

*I think part of the problem in my explanation was that the phrase "initial conditions" was a bit vague; it could have been a point, an interval, or both.

 

[jimbob]

I would argue that if identical initial conditions can lead to significantly* different outcomes, then the system is best described as random.

If we are talking about the evolution of intelligence, it was not inevitable even at the demise of the dinosaurs; so the evolution of intelligent life is due to a collection of chance events. This might be random, but was not "haphazard".

*This is not defined, and depends on what one considers significant.

 

[Lamuella]

evolution consists of one very random element (variation) and one distinctly and definitely nonrandom agent (selection).

It is undirected and unpredictable, but it is anything but random.

 

[Lamuella]

I would argue that if identical initial conditions can lead to significantly* different outcomes, then the system is best described as random.

how identical are we talking, here?

In addition, it is possible for something to have an unpredictable end-point and yet not be entirely "random". Take the Price Is Right game Plinko. In that game, a disc is dropped down into a series of pins and bounces around in such a way that its ending location is essentially unpredictable from the start. However, where it will end up is not entirely random. We know, for example, that the disc will not shoot up out of the plinko game. It will move downwards into one of the categories at the bottom. A large part of what we know about the way the game will end is nonrandom.

similarly, with evolution, we know that whatever course mutation and selection takes, it will end up with something that is a solution to the environmental problem

 

[mijopaalmc]

Meadmaker-

When you worked on process improvement/development in the car factories was there any discussion among you and your coworkers of whether the processes that you were trying to improve/develop were in reality random?

 

[jimbob]

evolution consists of one very random element (variation) and one distinctly and definitely nonrandom agent (selection).

It is undirected and unpredictable, but it is anything but random.

Part of this discussion has been whether describing "selection" as nonrandom is accurate.

Traits affect reproduction rates, and indeed "load the dice" but selection is inherently probabilistic, this does not mean that predictions can't be made, but the nature of the predictions is important, and has to deal with odds, untill the law of large numbers comes into play.....

Because the law of large numbers does come into play, one can talk about a selective advantage of 0.1%, and it still means something.

Here is my reasoning, hidden for brevity :

As articulett thinks this is confusing I would reiterate my question:

Originally Posted by jimbob
Does anyone here disagree with the following, and where?

1) Natural selection is probabilisic with various traits "loading the dice of selection" differently.

2) If a population is in equilibrium, the average number of breeding offspring per parent would be one.

3) The actual distribution of breeding offspring per parent is likely to be described by a poisson distribution, with lambda of one.

4) An advantageous trait will raise the average number of breeding offspring per parent above one. Thus lambda will be raised.

5) Conversely a disadvantageous trait would lower lambda.

6) The more advantageous a trait is, the more there is a selective pressure, and the more the lambda is raised.

7) A selective advantage of 10% equates to a lambda of 1.10 compared to the equilibrium lambda of 1.

ETA:

8) This is best understood as a probabilistic process for the reasons above.

I would then say that over geological timeframes, the actual environment is altered by random factors.

For example: (should an asteroid be in a chaotic orbit, then random events will influence that orbit significantly, and possibly set it on a collision course withe Earth) When, where and whether this happens would have been affected by random events.

The entire environment gets remoulded by such events, meaning that the selection pressures will have altered randomly over this timescale.

ETA: and a worked version with realistic numbers:

I like to think of it as loading the dice, but still so that there is chance about the outcome... (a deleterious mutation loads the dice heavily agianst survival).

Do you see my point, even if you don't accept it?

Actually I think of it modulating the mean value for the number of reproducing offspring per parent, and thus the chance of the traits being transmitted or surviving for a particular number of generations. (Assuming a poisson distribution, which I would argue is probably adequate.)

Thinking about the actual odds, the stability or otherwise of populations and the brood-sizes, you can conclude that any individual organism at birth is unlikely to breed.

Barn owls, for example, have a clutch-size of about 5, (many sources, for example here), mates at one year old, tends to live for 1-5 years inthe wild, and 15-20 years in a more benign environment, they can have two broods in a year.

A successful pair of barn owls could easily produce 25 chicks over that 5 years. However the population is declining in many parts of the world.

This means that on average, fewer than two chicks per pair will produce breeding offspring. The odds of a barn-owl chick surviving to breed could be *about* 1:12.

It is thus far more likely for a deleterious to be removed than for an advantageous one of a similar effectiveness to be retained.

Suppose a chick had a trait that doubled its chances of successfull reproduction; it would now have a 1:6 chance of successful reproduction. This particular trait, arising in a single individual is 6x more likely to die out in one generation than survive.

Suppose a chick had a trait that halved its chances of reproduction; it would now have a 1:24 chance of successful reproduction. This trait, arising in a single evolution, is 24x more likely to die out within a single generation than to survive.

If the advantageous trait conferred a 11.5x advantage compared to the lack of trait, then the odds of the chances of survival for one generation would be 11.5:12=23/24, and the odds of the trait being removed would be 1:24, the same as the odds of the original deleterious trait surviving.

Given a large enough population, and enough time, of course some beneficial traits will survive, and once established, will spread through the population.

Again, how likely this is for a particular trait, can best be explored with a probabilistic treatment.

 

[jimbob]

how identical are we talking, here?

In addition, it is possible for something to have an unpredictable end-point and yet not be entirely "random". Take the Price Is Right game Plinko. In that game, a disc is dropped down into a series of pins and bounces around in such a way that its ending location is essentially unpredictable from the start. However, where it will end up is not entirely random. We know, for example, that the disc will not shoot up out of the plinko game. It will move downwards into one of the categories at the bottom. A large part of what we know about the way the game will end is nonrandom.

similarly, with evolution, we know that whatever course mutation and selection takes, it will end up with something that is a solution to the environmental problem

If we are talking about a chaotic system like the weather, where a single ion caused by a radioactive decay would eventually have a significant effect; then I would say absolutly identical, assuming that quantum events are truly random, which seems to be the current consensus.


I agree that evolution is "not entirely random", and especially the bit about that a solution will evolve, however what form the solutions will take is not inevitable. (Of course there are some solutions or "niches" that are filled on many different occasions, flight, or sight being two, and large grazing animals, and their predators being other examples). I doubt that it would be possible to quantify the probability of a particular niche being occupied or a particular set of solutions evolving...

 

[articulett]

evolution consists of one very random element (variation) and one distinctly and definitely nonrandom agent (selection).

It is undirected and unpredictable, but it is anything but random.

The vast majority of people on this post agree as well as the writers of the Science article, Richard Dawkins and all experts in the field of evolution. Mijo and Jimbob, however, seem bent on convincing themselves and others that it's meaningful to call it random. Oddly enough, Intelligent Design proponent, Michael Behe, has this same quirk. No matter what you say or how many experts tell them that they are being unclear and misleading, they still insist on calling evolution random. Don't get your hopes up on changing the situation. They've imagined that they are clearer than the experts despite the fact that no one else seems to think so.

 

[Meadmaker]

Meadmaker-

When you worked on process improvement/development in the car factories was there any discussion among you and your coworkers of whether the processes that you were trying to improve/develop were in reality random?

None whatsoever.