Randomness in Evolution: Valid and Invalid Usage

[mijopaalmc]

Yes. Do you think anything I said contradicts anything it said?

Perhaps you could explain how the finding that speciation is dominated stochastic events is consistent with your assertion that evolution is predictable in the long run, given that the overall direction of evolution after speciation is dependent on the genetic background left by the speciation.

 

[sol invictus]

OH MY GOD, IT'S ALIVE!!

Perhaps you could explain how the finding that speciation is dominated stochastic events is consistent with your assertion that evolution is predictable in the long run, given that the overall direction of evolution after speciation is dependent on the genetic background left by the speciation.

What the heck....

I see your level of comprehension of this discussion hasn't changed at all.

"Evolution is predictable in the long run" - what does that mean to you? To me, it means that certain things can be reliably predicted in the long run (weird that it would mean that, huh?). That paper doesn't affect that conclusion.

For example: bacteria in petri dishes with some citrate. At the start of the experiment, none of the bacteria can metabolize citrate. The ability to metabolize citrate would be a strong advantage given the environment. So, let multiple colonies in multiple petri dishes live and reproduce for a long time.

Again: the claim at issue is that the theory of evolution by natural selection is predictive in the long term; that is, that it can predict some results of some experiments. So let's apply it to this. According to the theory, there is some non-zero probability per generation that a bacterium will mutate in such a way that it can metabolize citrate. Moreover, the theory tells us that that bacterium and its descendants will have an advantage in the sense that they will reproduce more rapidly than bacteria without that ability. More specifically, the theory predicts that eventually, all bacteria in that petri dish - and therefore eventually, in all petri dishes - should have the ability to metabolize citrate (modulo a few details, like that said ability doesn't come at a cost so significant it cancels out the benefit).

In principle the theory will also tell us how long we need to wait on average before that happens in a given petri dish, but to extract that prediction requires knowledge of the probability I mentioned, and that's hard to estimate (although one can certainly try). Regardless, the above is a definite, solid prediction for the long term, and we can test it.

So someone tested it. Guess what the result was (after a decade or so)?

 

[Wowbagger]

Perhaps you could explain how the finding that speciation is dominated stochastic events is consistent with your assertion that evolution is predictable in the long run, given that the overall direction of evolution after speciation is dependent on the genetic background left by the speciation.

Climate is also dependent on the background left by the weather. And, yet, we can predict larger trends in climate.

Perhaps an analogy of traffic patterns is easier. You can predict how bad the traffic is going to be around sports stadiums, by looking at the schedule of games. But, you can't predict, in advance, exactly which car is going to be on what part of the road, at any given time - even though the traffic patterns are highly dependent on which cars are on which parts of what roads.

This was NOT A PERFECT ANALOGY to biology, for several reasons. But, hopefully, it will illustrate the general idea of how something can be nonrandom, yet dependent on some random elements.

 

[mijopaalmc]

Uh...it's not even a good analogy.

As the study that was posted demonstrates, evolution is dependent of stschatic elements the majority of the time and is therefore, in general not predictable, especially over time scales over which speciation becomes important. In other words, the study study the exact opposit exact opposite of ehat you have been saying about that predictability of evolution and its resulting non-randomness.

 

[jimbob]

Wowbagger, This is the crux of the debate as far as I can see.

There is one view that evolution is generally predictable over the long term. In other words natural selection dominates, and the selection pressures are stable. If you reran the tape of evolution, you would come up with pretty similar organisms.

The other view is that there are enough complex nonlinear feedbacks in the system so that the selection pressures are subject to "random*" change over time, so that over a long enough timescale you can't make many useful predictions about what the ecosystem would look like, to say nothing about the organisms that have evolved within it.


Taken to its extreme, there might be a case for saying that the evolution of multicellular life was a fortuitous occurrence, which only occurs in a small proportion of planets where life arises. Of course there is currently no way of finding this out, at the moment, but we do know that the vast majority of biomass on Earth is not multicellular.


*And I would argue truly random as opposed to pesudorandom.

 

[Wowbagger]

Uh...it's not even a good analogy.

As the study that was posted demonstrates, evolution is dependent of stschatic elements the majority of the time and is therefore, in general not predictable, especially over time scales over which speciation becomes important. In other words, the study study the exact opposit exact opposite of ehat you have been saying about that predictability of evolution and its resulting non-randomness.

Read it carefully. The study is NOT the exact opposite of what I was saying. It is basically offering an adjusted framework by which evolution can be predicted to work.

It seems to indicate that evolution is predictable, over timescales that speciation occurs. It's just that the precise circumstances are less predictable, since (according the study) they are triggered by rare events of reproductive isolation, rather than in a "race" against the enviornment. (It even predicts a constant rate of speciation.)

Does anyone else disagree with my interpretation?

It really doesn't impact the overall statement that Natural Selection is nonrandom, one way or the other.

 

[Wowbagger]

There is one view that evolution is generally predictable over the long term. In other words natural selection dominates, and the selection pressures are stable. If you reran the tape of evolution, you would come up with pretty similar organisms.

The other view is that there are enough complex nonlinear feedbacks in the system so that the selection pressures are subject to "random*" change over time, so that over a long enough timescale you can't make many useful predictions about what the ecosystem would look like, to say nothing about the organisms that have evolved within it.

Both are correct, to certain degrees.

The ULTIMATE, broader causes and effects of evolution would be the same, if you reran the the tape of evolution. In a broad sense, you would likely get organisms that fill relatively similar niches in relatively similar manners.

There are just sooo many different ways one can fill a niche, and physics restricts the sorts of niches we would find in any given ecosystem.

However, this implies that the PROXIMATE, more specific causes and effects will likely be very different. (Though, we would also expect a certain amount of "coincidental convergence" of some superficial features, too.)

If we looked at the life forms of an alternative "run" of evolution, we would find that they would look very different from what we are used to, but will probably function and behave, in a broad sense, in recognizable ways.

For example, host/parasite relationships would probably work the same. Though, the existence of a parasite might not be recognized until it is spotted moving to a new host. (As is usual for novel parasites, even on Earth.)

Given enough time, multicellular life would likely develop, since its occurrence follows from host/parasite and other related models. Though, the exact manner in which it happens could differ, as we see in the variety of ways that sponges work.

 

[Wowbagger]

This thread will come to a bad end.

Looks like you were wrong. This thread will come to NO end.

 

[sol invictus]

And I'd say this is one of the more succinct deconstructions of the semantics involved that I've seen, sol.

Thanks.

Doesn't look like it did much good, though.

 

[recursive prophet]

Originally Posted by recursive prophet
And I'd say this is one of the more succinct deconstructions of the semantics involved that I've seen, sol.

Thanks.

Doesn't look like it did much good, though.

Planted seeds take a while to blossom sol. It took me 10 pages on that other thread to concede you were right in your analysis of the terminology on page one. Remember your smoke detector analogy? That planted the seed for me, but it took 10 pages for me to know what I knew after pondering your multiple choice questions.

I hope to soon send a PM to susu.exp and Marios at RDF before they turn out the lights there, suggesting they join the discussion here. I can't begin to remember all the arguments, but there were a passel of very bright people that basically agreed with mijo (mjpam there) wrt the non-deterministic overall nature of NS, and why it is important to view it from that perspective. Allele drift in small groups and reproduction success were two of the major factors they mentioned, as I recall. My_wan was the only one from here that jumped the fence, and he put on quite a show. Very compelling; many new rebuttals that resonated, but to little effect. He had-not sure if it's still ongoing-a challenge to the wikinition of random that was pretty awesome. Really made a lot of very new slants on this ambiguous concept. Wittgenstein's 'beetle in a box' came to mind reading it.

As always I'm far from certain which POV is more; relevant; pragmatic; or accurate. But it does seem to stimulate a lot of interesting discussion, and I'm glad this thread did a Lazarus even be it all pure glasperlenspiel.

If you have a chance I'd enjoy reading your comments on jimbob's last entry. And again, I'd really appreciate your thoughts on the XpoTurbine.

 

[mijopaalmc]

OH MY GOD, IT'S ALIVE!!



What the heck....

I see your level of comprehension of this discussion hasn't changed at all.

"Evolution is predictable in the long run" - what does that mean to you? To me, it means that certain things can be reliably predicted in the long run (weird that it would mean that, huh?). That paper doesn't affect that conclusion.

For example: bacteria in petri dishes with some citrate. At the start of the experiment, none of the bacteria can metabolize citrate. The ability to metabolize citrate would be a strong advantage given the environment. So, let multiple colonies in multiple petri dishes live and reproduce for a long time.

Again: the claim at issue is that the theory of evolution by natural selection is predictive in the long term; that is, that it can predict some results of some experiments. So let's apply it to this. According to the theory, there is some non-zero probability per generation that a bacterium will mutate in such a way that it can metabolize citrate. Moreover, the theory tells us that that bacterium and its descendants will have an advantage in the sense that they will reproduce more rapidly than bacteria without that ability. More specifically, the theory predicts that eventually, all bacteria in that petri dish - and therefore eventually, in all petri dishes - should have the ability to metabolize citrate (modulo a few details, like that said ability doesn't come at a cost so significant it cancels out the benefit).

In principle the theory will also tell us how long we need to wait on average before that happens in a given petri dish, but to extract that prediction requires knowledge of the probability I mentioned, and that's hard to estimate (although one can certainly try). Regardless, the above is a definite, solid prediction for the long term, and we can test it.

So someone tested it. Guess what the result was (after a decade or so)?

I'm not sure why those who argue that evolution is non-random keep holding up Lenski's research as a demostration that evolution is non-random. Only one of the twelve original, identical, replicate populations evolved the cit+ phenotype, and the vast majority of replicate populations taken from different generations failed to re-evolve the cit+ phenotype.

How exactly do these results demonstrate that evolution is in general predictable over the long-run?

 

[sol invictus]

I'm not sure why those who argue that evolution is non-random keep holding up Lenski's research as a demostration that evolution is non-random.

I suppose that's because, despite literally years of discussing this, you still fail to comprehend the most fundamental aspects of it.

Only one of the twelve original, identical, replicate populations evolved the cit+ phenotype, and the vast majority of replicate populations taken from different generations failed to re-evolve the cit+ phenotype.

How exactly do these results demonstrate that evolution is in general predictable over the long-run?

Did you read what I wrote? I described precisely how Lenski's results demonstrate that. Is there something in there that wasn't sufficiently clear?

The fact that one population evolved and fixed the cit+ phenotype is an massive success of the theory, because the odds of such a thing happening by chance (or in any other theory) are essentially zero. That experimentally tested successful prediction by itself proves - to the extent proof is possible in science - that evolution is both predictive and correct, and predictable in the sense I described.

Of course there are mountains of other experiments and fossil records that prove the same thing.

As a side note: why did only one population out of twelve evolve that trait over the course of the experiment? As I said that's something one could predict too (as a probability distribution, obviously) given knowledge of the time-scale for the mutation and/or the probability per generation for it to occur. As I described in my previous post, such probabilities are difficult to estimate a priori, but in any case they aren't necessary for and do not affect the success of the prediction that the trait will eventually evolve and fixate.

 

[mijopaalmc]

I suppose that's because, despite literally years of discussing this, you still fail to comprehend the most fundamental aspects of it.

Such a bold assertion with absolutely no evidence.

Did you read what I wrote? I described precisely how Lenski's results demonstrate that. Is there something in there that wasn't sufficiently clear?

Yes, I read what you wrote.

The thing that is not sufficiently clear is how you get from "there is some non-zero probability" to "non-random". The existence of probabilities is, by definition, sufficient grounds for calling evolution a stochastic (or random) process.

The fact that one population evolved and fixed the cit+ phenotype is an massive success of the theory, because the odds of such a thing happening by chance (or in any other theory) are essentially zero. That experimentally tested successful prediction by itself proves - to the extent proof is possible in science - that evolution is both predictive and correct, and predictable in the sense I described.

And, once again, "predictable" doesn't mean "non-random". If it did, processes where events are independent, identically and uniformly distributed wouldn't be random, as the proportion of successes to total trials can be predicted (with error).

Of course there are mountains of other experiments and fossil records that prove the same thing.

And I don't dispute the evidence. I just dispute the redefinition of widely accepted scientific term in order to get the evidence to conform to your preferred philosophical opinion.

As a side note: why did only one population out of twelve evolve that trait over the course of the experiment? As I said that's something one could predict too (as a probability distribution, obviously) given knowledge of the time-scale for the mutation and/or the probability per generation for it to occur. As I described in my previous post, such probabilities are difficult to estimate a priori, but in any case they aren't necessary for and do not affect the success of the prediction that the trait will eventually evolve and fixate.

So stating predictions in terms of probabilities is now how "non-random" is defined?

 

[Uncayimmy]

As a side note: why did only one population out of twelve evolve that trait over the course of the experiment? As I said that's something one could predict too (as a probability distribution, obviously) given knowledge of the time-scale for the mutation and/or the probability per generation for it to occur. As I described in my previous post, such probabilities are difficult to estimate a priori, but in any case they aren't necessary for and do not affect the success of the prediction that the trait will eventually evolve and fixate.

If I understand you correctly, what I think you are saying is this:

In any given species there are existing variations as well as new mutations. They have some probability of happening. Let's say the variation is a resistance to a toxin that normally kills most people. We know such variations exist in humans because we all don't react the same way to bee stings, snake bites or ingesting poisons.

One way to detect the prevalence of these variations in a species such as bacteria would be to isolate the gene responsible and then sample a buttload of bacteria looking for this gene. That's time consuming and expensive I would think.

Another way to do it is pretty much like the experiment described. Try to grow colonies of the bacteria in the presence of the toxin with other control colonies growing in the "normal" environment. The controls will reproduce and grow at a certain rate. Those colonies in the toxic environment with no resistance will die off. Those colonies with a variation that allows them to survive in the toxic environment will experience a dying off of the unlucky bacteria while the lucky ones survive to reproduce. The overall population will drop followed by an increase as the bacteria with the variation grow in numbers.

Mathematically I'm sure you could figure out the prevalence of the variation/mutation. What's interesting is at the same time you are also demonstrating evolution at work. It also demonstrates that evolution is random. There's no guarantee that the variation will appear before the population has died off.

As a simple example think of rolling a "population" of three dice. The "toxin" is a hammer that smashes any die that doesn't land with the two dots facing up. If the three die are 4, 5, and 1, then the entire population of three dice is wiped out when the hammer arrives. If the population is 2, 2, and 2, that entire population survives to reproduce.

Mathematically we can predict that given N populations of three dice with the number of dots appearing equally just how many populations have one or more 2s in them. So while it's predictable, it's still random. At any given time there might be no populations with a 2 in them, so the whole species gets wiped out.

Then again, I'm just a layman trying to explain what might be an oversimplified way of looking at this.

 

[sol invictus]

Such a bold assertion with absolutely no evidence.

The evidence is in your posts.

Yes, I read what you wrote.

The thing that is not sufficiently clear is how you get from "there is some non-zero probability" to "non-random".

If you read what I wrote, why are you asking that? Do you see anywhere in my post where I even used the term "non-random" or "random"?

And, once again, "predictable" doesn't mean "non-random". If it did, processes where events are independent, identically and uniformly distributed wouldn't be random, as the proportion of successes to total trials can be predicted (with error).

And once again, did you read what I wrote? If so, what are you talking about?

And I don't dispute the evidence. I just dispute the redefinition of widely accepted scientific term in order to get the evidence to conform to your preferred philosophical opinion.

Frankly, I don't give a **** about semantics - and I will not waste my time arguing over the definitions of terms.

So stating predictions in terms of probabilities is now how "non-random" is defined?

Once again....

 

[sol invictus]

If I understand you correctly, what I think you are saying is this:

In any given species there are existing variations as well as new mutations. They have some probability of happening. Let's say the variation is a resistance to a toxin that normally kills most people. We know such variations exist in humans because we all don't react the same way to bee stings, snake bites or ingesting poisons.

One way to detect the prevalence of these variations in a species such as bacteria would be to isolate the gene responsible and then sample a buttload of bacteria looking for this gene. That's time consuming and expensive I would think.

Another way to do it is pretty much like the experiment described. Try to grow colonies of the bacteria in the presence of the toxin with other control colonies growing in the "normal" environment. The controls will reproduce and grow at a certain rate. Those colonies in the toxic environment with no resistance will die off. Those colonies with a variation that allows them to survive in the toxic environment will experience a dying off of the unlucky bacteria while the lucky ones survive to reproduce. The overall population will drop followed by an increase as the bacteria with the variation grow in numbers.

Mathematically I'm sure you could figure out the prevalence of the variation/mutation. What's interesting is at the same time you are also demonstrating evolution at work. It also demonstrates that evolution is random. There's no guarantee that the variation will appear before the population has died off.

As a simple example think of rolling a "population" of three dice. The "toxin" is a hammer that smashes any die that doesn't land with the two dots facing up. If the three die are 4, 5, and 1, then the entire population of three dice is wiped out when the hammer arrives. If the population is 2, 2, and 2, that entire population survives to reproduce.

Mathematically we can predict that given N populations of three dice with the number of dots appearing equally just how many populations have one or more 2s in them. So while it's predictable, it's still random. At any given time there might be no populations with a 2 in them, so the whole species gets wiped out.

Then again, I'm just a layman trying to explain what might be an oversimplified way of looking at this.

I'll repeat what I said long ago: if "random" is synonymous with "probabilistic", or if it simply means that some aspects of the process in question are not deterministic or cannot be predicted with 100% certainty, then it is random that smoke detectors go off when smoke pours into a room(most smoke detectors rely on the decay of radioactive atoms, which is about as truly "random" - at least by those definitions - as you can get).

I'm pretty sure that vast majority of people, including scientists, would agree that that statement is totally absurd. Since I'm not very interested in writing a dictionary entry (I'm interested in the science of evolution), I'd rather just avoid the term "random" entirely.

Here are some (very) old posts of mine on that:

So mijo - are smoke detectors random?

You realize this simple question totally demolishes any lame vestige of a point you might have had, right?

You have three choices:

a) you agree with the statement "smoke detectors are random" (totally absurd), or

b) you agree with the statement "smoke detectors aren't random" (then neither is evolution), or

c) you argue that neither statement suffices because smoke detectors have both random and non-random elements, in which case it is crushingly obvious to everyone that the same applies to the vastly more varied and complicated phenomenon of evolution.

Pick your poison, mijo.

 

[sol invictus]

Oh, and one more since I was looking back:

No, it is not at all like the smoke detector.

Really?

What I call "random" is that only one out of the twelve populations evolved the ability to metabolize citrate despite the fact that the populations were identical at the beginning of the experiment and the fact that they evolved in the same evirnoment.

Yes, evolution takes time.

Even if all the populations eventually evolved the ability to metabolize citrate, the evolution of that trait would still be random in so far as the populations evolved that trait at different times. I've said it before, and I'll say it again: convergence does not imply non-randomness.

So you think the fact that these bacteria evolved the ability to metabolize citrate when put in a citrate rich environment is random.

That is just as ridiculous as your earlier contention that a smoke detector is random because it relies on the (truly random) radioactive decay of unstable atoms. You are using the word in a way no one else does - including mathematicians. You were unable to supply a single reference that defines the word your way, and you never responded when I checked several references - including a standard text on prob. and stats. - none of which define it your way.

The statement that "evolution is random", full stop, is (to be blunt) stupid.

 

[Vorticity]

My long response to this thread:

Although I've been having this argument on these forums for a number of years now, I haven't participated much in this thread, mostly due to my firm belief that it is futile.

There are two groups of people here that have fundamentally different sets of axioms about the meaning of the word random. These two groups will never see eye-to-eye since they are coming from completely different reference frames, from which each of their assertions ("evolution is random", "evolution is non-random") are true by definition.

I'm not going to get into the argument again, except to note my (doubtless controversial) observation that, in these arguments, people (like myself) who have graduate or undergraduate level educations in probability theory and/or statistics (statisticians, mathematicians, mathematical physicists, etc) generally tend to agree that evolution is "random" by the mathematical definition of the word. Everyone else tends to disagree. Doubtless individual exceptions could be found, but this is my observation of the general trend.

For a look at a thread that ended up going the other way, see this thread: http://www.internationalskeptics.com/forums/showthread.php?t=50550 in which it is noted (for example):

If Tai said "Mathematically speaking, evolution is a random process.", then few of us would seriously disagree.

My short response to this thread:

KILL IT WITH FIRE!

 

[mijopaalmc]

sol invictus-

Your assertion "that eventually, all bacteria in that petri dish - and therefore eventually, in all petri dishes - should have the ability to metabolize citrate" is unfalsifiable, because you can always claim that we simply haven't waited long enough. In fact, you essentially have when you responded, "Yes, evolution takes time" to my observation that not all of the cultures had evolved the cit+ phenotype. The crucial thing that you are seemingly ignoring is that it take an infinite amount of time for you to be certain that all the cultures will have evolved the cit+ phenotype. Thus, in any finite universe and for any finite experiment, there is a non-zero probability that you will observe at least one of the cultures will not have evolved the cit+ phenotype. This is, of course, perfectly consistent with evolution being a random (stochastic) process and completely inconsistent with evolution being a non-random (deterministic) process.

As to your obsession with smoke detectors, it may be absurd to assert that "it is random that smoke detectors go off when smoke pours into a room" without further explanation. It is perfectly acceptable, and possibly preferable, to make such a statement if you wanted explore why, for instance, americium-241 is used instead a nucleide with a longer half-life (e.g., uranium-238). Such an explanation may not be appropriate for a "how things work" talk on smoke detectors at a science museum, but it maybe perfectly suited for a lecture on the non-medical applications of radioactive decay in an advanced undergraduate engineering seminar. In essence, my answer to your question is "it depends on the context", which, while not the most satisfying answer, is the most straight-forward answer I can give.

 

[Vorticity]

sol invictus-

Your assertion "that eventually, all bacteria in that petri dish - and therefore eventually, in all petri dishes - should have the ability to metabolize citrate" is unfalsifiable...

But there is a greater issue here.

Even if it were the case that eventually, all petri dish populations develop the ability to metabolize citrate (which would not surprise me), this would in no way prevent evolution from being a random process by the technical definition of the term. Given any stochastic process, you can almost always apply a filter to its random output that yields non-random results.

(See my previous response in this thread which touches upon this issue.)

sol is applying a filter to the output of the "petri dish evolution" stochastic process. Specifically, he's applying the "Does every petri dish develop the ability to metabolize citrate" filter. This may very well yield a non-random result ("Yes"). The fact that this particular filter yields a predictable outcome in no way implies that the original, unfiltered process is non-random. One could just as easily apply the "On which generation does the first dish develop the ability to metabolize citrate?" filter. Is there any doubt that the output of such a filter is a profoundly random variable?

There is more to the definition of what constitutes a random process than just what the long term trend or outcome is, even if that particular final state happens to be non-random.