In the “Evolving complex features” thread, TristanM asks:
I’m curious what the group’s thoughts are on Jonathan Bartlett’s (AKA “JohnnyB”) argument about open-loops in the AVIDA program?
I haven’t studied Jon’s argument yet, but I think some of you have. What did you think?
Why I Love AVIDA – Detecting Design in Digital Organisms
Thoughts on Parameterized vs. Open-Ended Evolution and the Production of Variability
Irreducible Complexity and Relative Irreducible Complexity: Foundations and Applications
How does design solve this “problem”?
I don’t have any problem with the notion that there are conceivable biological structures which cannot be produced via (non-)Darwinian evolutionary processes. The question is, are there any biological structures which actually do exist now, or actually have existed in the past, which cannot be produced by evolutionary processes? Until some Creationist (and yes, IDists are Creationists) manages to pony up a genuine instance of such a thing, I’m really not interested in yet another Creationist argument of the general form there is a specific sort of whatchamacallit (which may or may not ever have actually existed) that must be a product of Design on account of it can’t have evolved.
AVIDA and other simulations are only attempting to model the possibilities created by variation and selection. Specifically, whether IC structures can evolve.
You can’t take a limited model and use it to prove limitations in the system being modeled. That’s just silly. AVIDA has one job and one job only, to test whether IC structures can evolve. It is a purely abstract exploration of Behe’s and Dembski’s math. It isn’t biology.
You appear to be asking AVIDA to explain the origin of itself, an argument analogous to asking evolution to explain the origin of life.
I would say that ID is bad science.
Any science that consistently and repeatedly reifies metaphors and mistakes the map for the territory is incompetent science.
ID does this by taking the metaphor of information and of computer programming and arguing that the limitations of programming apply to biochemistry. That’s science, but it is bad science.
The converse is not true.
If you make an abstract mathematical argument about what random mutation and selection cannot do, then a model can demonstrate that it can so do it. It is both reasonable and possible to prove a positive statement about what is within the bounds of probability.
But a limited or incompetent program cannot demonstrate that something is impossible. It is fairly easy to write a program that doesn’t work.
Well, the real question isn’t “how does design solve…?” The real question is the one which gives IDists the heebie-jeebies “When design is the solution, how did the design get incorporated into the first cell’s protoplasm? Did the designer’s physical agent have the equivalent of tiny fingers and tiny microbiological instruments with which it inserted a pipette full of the t-tRNA it had assembled in some lab? I mean, obviously the designer wasn’t actually human — that would be a time-traveling paradox — but it had to have interacted physically at some instance with the physical material of the (proto) cell which needed the t-RNA “infusion”. What is an IDists conception of how that, physically, happened? (Or happens now, if they imagine the designer’s physical agent is still mucking about with mutations in present day …)
The real question for Sal and people like him is “What do the designer’s ‘fingers’ do?”
Never once have any of them even begun to answer that.
petrushka:
I’ll vouch for that. It happens even when the programmer isn’t trying. Often.
The real problems for front loaders and programmed mutation folks is genetic load.
Apparently the Deeper Genome (to borrow John Parrington’s phrase) is indifferent to actual sequence, because it chugs along eon after eon paying no attention to mutations.
This is contrast to coding and regulatory sequences, which are conserved. Purifying selection pretty much limits mutations to synonymous substitutions.
It occurs to me that God must love evolution, because He made the rules of logic such that it is not possible to prove a negative. One can demonstrate that evolution is possible, but one cannot prove that it is impossible.
But the deck is not stacked. the same holds for Design. One cannot prove that Design did not happen or that Design cannot happen.
So it boils down to whether one thinks it is reasonable to extrapolate observed processes to explain historical tendencies. As in geological processes or solar system formation, or the creation of heavier elements.
Trying to parse this, and the best I can do is:
If a complex feature requires, as intermediates, variants that are less fit than their precursors, it will evolve less readily.
Which is obviously true. However, the point of AVIDA was to discover whether that meant that complex functions that could only be “reached” via deleterious intermediations, i.e. by “IC pathways”, including pathways that not only had many neutral steps but some quite deleterious ones. And the answer is “no”.
So Behe’s IC pathway version of IC, as well has his IC function version of IC turn out not to be bars to evolution.
Alternatively, your premise is false.
It’s not “an OoL problem”. OoL theories must provide a mechanism for the first self-replicators to arise from non-self-replicators. Nobody suggests that a system with tRNA arose spontaneously from non-self-replicators. It’s an evolutionary problem.
Quite. Which is why Behe made his IC argument. But which is refuted by AVIDA.
But that is not what anybody is claiming (apart from William J Murray apparently). If you read the AVIDA paper you will see that in the absence of selectable simpler functions, EQU did not evolve.
With selectable simpler functions it evolved in about 50% of runs.
johnnyb,
Indeed, but it is evidence that the points are connected. It would require something to reject the hypothesis that this connection is traversible stepwise. You could easily declare the points unconnectable unless someone shows them being connected by a series of viable organisms with no fitness constraint, but you know that isn’t going to happen, don’t you?
So we have what we have. If you think traversal needs to be ‘information-guided’, and the environment could not provide that information but a Designer could, show a step where that was necessary. Which requires that you know the unknowable too – what the selective advantages actually were.
All of these discussions of guidance or lack of guidance seem to assume that evolution has goals, targets or direction.
This is painting bullseyes after the arrow has landed.
I don’t know if AVIDA has EQU as a target. If it does, it is not a good model of evolution. Perhaps EQU is the only complex function it has reached in trials, in which case it is a rather limited simulation.
But regardless, its only usefulness — to my knowledge — is demonstrating that IC can be reached by evolution. That there is no logical or probabilistic barrier to evolving IC.
Depends what you mean by target.
The virtual organisms can gain extra “energy” by performing specific logic functions, one of which is ==.
If it can do it correctly, i.e. return correct output for input, it gets “food”.
Just as in a natural environment, an organism that can, say, reach a higher branch, or crack a bigger nut, or resist a given pathogen, gets additional resources.
But there isn’t a single set of instructions that will produce each function. The function can be performed by many different sequences of instructions.
So it’s quite strongly analogous to a biological function that can be performed by a number of different proteins, but for which many mutations will produce a useless protein.
The sequence of instructions that will produce each function are not unique, and many may not be known in advance. The whole point of using EAs practically is to find virtual “organisms” that will perform specific functions, but for which the code is unknown.
I think that’s consistent with my understanding. AVIDA has a limited universe of rewardable ICs. Biochemistry does not. So AVIDA is a limited simulation. It can demonstrate that random mutation and selection can result in ICs. There is nothing special about irreducibility.
But one cannot ascribe the limitations of AVIDA to biology. Biology and biochemistry are not constrained by the limitations of AVIDA.
Yes. AVIDA doesn’t make it easy for things to evolve. Only a narrow set of functions are “rewarded”, all are IC, and none is a “simpler” version of any other.
Often, a second function is acquired at the cost of one already acquired by that lineage.
And it’s all asexual.
It turned out that “EQU” was particularly difficult to evolve, because it seems that the most probable pathways to it involve a large number of neutral and deleterious steps. and also requires that some of the other functions appear in the lineage first, even if they are later broken (may have to be broken).
Yet it still evolves in about half the runs.
I understand chaotic systems, but it is unclear to me what the fact that the behavior of chaotic systems is “unpredictable” has do with evolution. There is no “prediction” in evolutionary adaptation, so unpredictability is irrelevant.
In your 2010 paper you stated,
That’s what I was responding to.
stcordova,
Simple. Make primitive tRNA synthetases out of something other than protein. RNA for instance, which has all the catalytic repertoire necessary. Then, when protein synthesis starts to take off, they take over (in two apparent separate events – there are two families of tRNA synthetase with substantial homology within the two groups, indicating that all 20 of them were not present at the start, even though all 20 acids are now used in them all). Evolution, IOW.
This is where you paste some Creationist tract on the impossibility of RNA catalysis.
johnnyb,
Further to this: whatever ‘x-guided’ process was involved shows a remarkable transversion-transition bias and a preference for silent sites and chemical property conservation for non-silent substitutions. Just like a mutational process, it seems.
Petrushka –
I’ve said this several times on this board. If you don’t like Avida, fine. Find me *any* evolutionary system that matches my requirements above that generates anything but the most trivial open-ended loop. Or, if you think I’m cheating by saying “anything but the most trivial”, then find me one that is even trivial. Let’s look at it and let everyone decide. People have been doing evolutionary experiments for over a half century. Find an example *anywhere* that meets the requirements. But don’t act like this is something unique to the AVIDA model, that if only they had done the right model we would have the right stuff.
Chaitin is the only one who has taken this problem seriously, and in his work he only got the most minimal results in a purely abstract (unimplementable) model. (NOTE – I actually think his results overstate the case, and may actually be directly wrong, but I have not had the time to investigate sufficiently to verify my thoughts, so I will take his results as being factual for the time being). Basically, his maximum evolutionary rate was (if I remember correctly) O(n^3), which is ridiculously bad for a toy, pure mathematical model whose only goal was to generate novelty and leave the hard parts of the world or biology out.
Also, I should point out that Chaitin could not get a model to work with basic, selective mutations, despite working on the problem for years. He had to include macromutations, including mutations which rewrote the whole genome in one generation. Without that, organisms never grew in complexity – they always quickly reached a threshold and petered out. Which is exactly my claim.
Could you clarify what you mean here, Jonathan?
Is your claim that evolutionary processes, as proposed by Darwin, can’t generate open-ended loops?
And yet biological organisms display them? And that open-ended loops may be the finger-print of design?
Apologies if I am missing your point.
I could write a program that doesn’t work in less time.
Seriously, if the program doesn’t emulate chemistry, what is demonstrated by negative results?
People have been working on OOL for a long time without any definitive results. What is demonstrated by lack of success?
I would think if life is designed, someone in the ID community could provide a proof of concept. Design something that is not a copy or pastiche of what has evolved.
Behe presented a continuum version of IC, but the AVIDA organisms still evolved by deeply IC pathways, i.e. ones involving many neutral and even deleterious steps.
You seem to be saying that only one person, a mathematician with AFAIK no qualifications in biology, is the only person who has taken the approach you take seriously.
Has anyone in the scientific community taken up his proposed research program?
I think Reciprocating Bill is asking some questions which probe this issue more specifically.
ETA: I admire your grit in taking on all comers in this forum.
I second that. Remember there are lurkers that may not comment but are taking it all in
peace
PS don’t let them bully you into using block quotes. The way you do it is much better
To amplify this point:
To this I would add that the unpredictability of chaotic systems works against the design hypothesis. Human design entails envisioning in advance the impact of particular changes in a system, small or large. That is not possible with chaotic systems – as the only way to determine the path a truly chaotic system will take is to run the system itself. Envisioning that result in advance, and arranging small changes in initial states to attain that result, isn’t possible.
Since we do, in fact, correctly envision the behavior of open-ended loops and employ them in our programs, I take that as an indication that changes in program behavior that result from variations in code are not truly chaotic. Were they, the ONLY way to arrive at functioning code that accomplishes what we desire would be through trial and error – e.g. variation and selection. Rather, you are simply breaking a fragile mechanism upon making random changes. No need to resort to a characterization of such as system as “chaotic” to understand that result.
FFS.
johnnyb,
Were they multicellular sexual organisms?
This is the problem with Chaitin’s analysis. He looks at evolution as a logic problem. But, properly understood, it is an “interaction with reality” problem.
Can you say more?
BruceS –
Thanks for pointing out Reciprocating Bill’s comment. I had missed it.
Reciprocating Bill –
“it is unclear to me what the fact that the behavior of chaotic systems is “unpredictable” has do with evolution”
It is based on the *reason* that it is unpredictable – it is chaotic. That is, there is not a relationship between the direction of selection and the needed adaptations. That is, getting near a local maxima is more than likely driving you away from a global one, so one cannot develop a selective system that directs evolution properly, because small gains do not bring you nearer to the optimal results.
As to the continuum statement, go back to my technical note above. The minimum size of a selectable system with an open-ended loop gives you at least a starting point for estimating the probability of its occurrence. It isn’t just a yes/no, but quantifiable. It is usable as a yes/no (like Behe does), because after the very most simplistic systems, it quickly becomes ridiculously improbable. Think about digital electronics. Technically, they are still continuous signals, but they break fast enough that they are counted as discrete.
BruceS –
“You seem to be saying that only one person, a mathematician with AFAIK no qualifications in biology, is the only person who has taken the approach you take seriously.”
It’s not quite that, but rather that he is one of the few who takes seriously the computational issue of biological novelty. The reason for this is that he is one of the prime moves of information theory over the last century, focused on the question of novelty in mathematics. Therefore, even if you disagree with him, I think his ideas of what constitutes novelty should be taken seriously. I also agree with him in this regard.
But what is *really* interesting, is that no one been able to tackle this just in regular evolutionary computation. Why do you think that is? Open-ended loops (or their equivalents, i.e. recursive functions) are required for open-ended functionality. Why are the availability of evolved loops so rare among researchers? And this is not even considering my specifications above (Turing-complete). Even with non-Turing complete systems and close-ended loops, usable loops are astoundingly rare. Check out this review paper on generated loops in evolutionary computing literature:
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.207.9680&rep=rep1&type=pdf
I found this long after my paper, but it makes a lot of the same points I do (NOTE – the application to biological evolution it presents is pretty superficial in my opinion – I am talking about what it says about computation).
The only one that comes close is this one: “Learning Recursive Sequences via Evolution of Machine-Language Programs”. It is interesting, but disqualifies from the running for the following reasons: (a) it only has 12 memory storage locations (think of a computer with 12 bytes) – not even approximating Turing-completeness. (b) Fixed program size of 12 instructions – not even approximating Turing-completeness. (c) It limited the number of evaluation steps. This is an arguable point, and the paper doesn’t describe well the consequences, so I’m not going to push it. (d) the instruction set was highly geared towards looping. I’m not going to push that one either, but I think it worth pointing out.
Also, if you look at the results, where the data exists, you will find that most of the information came from the random search. A friend of mine pointed out that only 3-5 bits of information (measured by Dembski Active Information) were generated by the hill-climbing, with most of the bits (~18) being generated by random search. This means that, as I said, for open-ended loops, the probability of evolving a random solution is given by the minimum selectable open-ended loop.
Given that this is, as far as I can tell, the best that evolutionary computing has done to counter my thesis, I consider it well-evidenced by the literature.
And yet, in AVIDA, even where a deleterious mutation was necessary to get to a new function, that new function evolved.
So even in a very low-dimensioned environment (only a handful of ways of increasing fitness, and some of those not mutually compatible apparently) a complex function reliably evolved by means of pathways with many neutral and deleterious steps.
So Darwin was too conservative. It turns out not the case that his mechanism only produces X if there is an advantage to every step “towards” X . This is because of drift.
But in addition to that, in real life, the fitness “landscape” is very high dimensioned – so “local maxima” are not only escapable via drift, but are also not nearly as common as simplistic models (like AVIDA) present – because there is so often SOME dimension along which a population can “travel” away from what would otherwise be a “maximum”.
Reciprocating Bill –
“To this I would add that the unpredictability of chaotic systems works against the design hypothesis. Human design entails envisioning in advance the impact of particular changes in a system, small or large. That is not possible with chaotic systems – as the only way to determine the path a truly chaotic system will take is to run the system itself. Envisioning that result in advance, and arranging small changes in initial states to attain that result, isn’t possible.”
You are incorrect on two counts. First of all, the underlying system *must* be chaotic to allow open-ended design. This is basic Computer Science. Turing-complete systems are chaotic, and without the system being Turing-complete, it is not open-ended as a programming system.
Second, It is only impossible to do it *computationally*. But it is, at least apparently, possible to do it mentally. This is why we are able to program. If we weren’t able to do it mentally at all, then I contend we we simply could not program. But that is a more appropriate topic for the Materialism thread:
Elizabeth –
I disagree that what Avida comes up with are “complex adaptations”. In addition, this is one way in which Avida often behaves un-evolutionarily – dead organisms don’t normally reproduce, but they can in Avida.
As to your complex landscape idea, I think it makes the problem harder, not easier. Organisms have to solve multiples of these dimensions *simultaneously* to survive. A loss in any one of several will prevent that organism from *ever* reproducing. Additionally, for open-ended loops, even if it wasn’t an area that was being mortally selected for, the open-ended nature would drain resources from the organism and prevent them from functioning in all other aspects.
In addition, even if multi-dimensionality was a way to mitigate some of this (which I don’t think it is), it would only apply late in the evolutionary game, when organisms had more dimensions to work off of. In other words, it would require existing information to work.
So what is not complex about them?
Do you disagree that they are IC?
Can you explain what you mean by this?
Again, can you explain what you mean by this? What I am proposing does not entail that any organism has to “solve” anything. And even a population doesn’t have to “solve” any “dimension” “simultaneously.
Let’s take a toy example:
A fish is being predated by a larger fish. Camouflage helps. Those with the best camouflage survive. Eventually, the whole population has spots the colour and size of the background pebbles. The population (not “organism”) is on a “local maximum”. But some fish have slightly faster reflexes when they detect the shadow of the predator. So that population moves off the “local maximum” – we now have fast spotted fish.
Two dimensions, yet neither the increased reflexes nor the better camouflage have to happen “simultaneously”.
I’m not sure why you say this. Most loss-of-function mutations are non-lethal, even where the function is normally beneficial. In AVIDA, there are loss of function mutations that are not lethal, but seem to be necessary steps on the pathway to EQU.
Not sure what this means either. Can you give an example? You seem to be assuming a very brittle system, in which any loss-of-function is lethal. Why should we assume this?
As Darwin (and lots of us in this thread) have said – clearly Darwin’s mechanism only kicks in once you have self-replication.
But why need you wait long for multiple dimensions? In the early stages of a proto-biont there are likely to be easily as many ways of improving reproductive chances as there are later, surely? Easier to improve on a non-optimised system than an optimised one!
There’s an online book about it:
A mathematical theory of evolution and biological creativity
It’s a while since I last looked. My recollection is that it is mostly logic.
To me, the creativity of evolution has to do with the struggle to survive in a dynamically changing environment. And I don’t see Chaitin’s ideas as particularly relevant to that.
Elizabeth –
I don’t really want to get into an argument over Behe’s definitions – it seems kind of pointless since that’s not the topic, but I’ll give a post to it. It was probably not as well developed in my paper as I had thought. Basically, I think that open-ended loops *are* the proper computational mapping to Behe’s idea. If I remember correctly, Behe, in the larger development of his idea, tried to convey the fact that they were core processes, and not just basic external catabolic systems. I think the reason for this is because the more core processes are more likely to be subject to recursive control. Behe, not being a computer scientist, did not know how to express that, but I think that this is the meat of what he was trying to get at. Therefore, since it is non-recursive (aka not an open-ended loop), it is not IC.
As to Avida having dead organisms reproduce, I think (it has been a while since I dug into it) that if you don’t properly perform tasks, it just slows you down, it doesn’t necessarily kill you. So, if no one ever performed a task correctly, they would still function and reproduce, but at a reduced rate. I think they do die if they exceed a population threshold, but I don’t think that they do from lack of performing function. They just reproduce more slowly.
“Two dimensions, yet neither the increased reflexes nor the better camouflage have to happen “simultaneously”.”
I don’t disagree that there are also dimensions that work out in evolutionary favor, only that they are compounded by the number of dimensions that *do* have to happen and work simultaneously.
“Most loss-of-function mutations are non-lethal, even where the function is normally beneficial.”
I agree, but they certainly aren’t selected for. Also, if natural selection requires you go down too many non-selectable highways, can evolution really be said to be operating “by natural selection”? If natural selection is too weak, it doesn’t do anything at all, as John Sanford has pointed out. This is a problem for evolution, which needs NS to build organisms, not design.
“Not sure what this means either. Can you give an example? You seem to be assuming a very brittle system, in which any loss-of-function is lethal. Why should we assume this?”
Here I’m not talking about loss of function. In the paper, the biochemical equivalent to open-ended loops is a negative feedback system, where a downstream product is required to suppress the generation of a protein. If the system is not completed yet, or the feedback is damaged, the open-ended loop is left wide open, draining organism resources. This is the same problem in both computer science and in organisms. I have run across several references to diseases where things are not turned off properly, and most of them are pretty bad, but I don’t have the references at the moment. This is why open-ended loops make selection for them so difficult – if the full system isn’t in place, it can easily spin out of control, not just basic loss-of-function.
“clearly Darwin’s mechanism only kicks in once you have self-replication.”
Right. My point is that self-replication is just one of the things that are needed. In order to successfully evolve across multiple dimensions (take your fish example) you need lots of existing code in multiple dimensions (pigment and motor genes, for instance). Therefore, even if multiple dimensions eases your problems, to enable multiple dimensions requires a *lot* of existing information, much more than just self-replication.
“But why need you wait long for multiple dimensions? In the early stages of a proto-biont there are likely to be easily as many ways of improving reproductive chances as there are later, surely?”
I think that’s wishful thinking. In orde to improve something, you need a parameterized system. It takes a lot of information to build systems sufficiently parameterized to evolve along different dimensions. And, at the start, the number of dimensions you are successful on is greatly outweighed by the number of dimensions you could die on.
Just as a general FYI, I am probably going to stop responding after tonight. I had some extra time today as I had a day off, and a taekwondo injury that prevented anyone from expecting anything from me today. Anyway, I’ll try to make one more set of responses this evening, but after that I’m going to call it a day. Thanks for the thread, I’ve enjoyed talking to you all!
See. Someone else who understands how Avida works.
Behe’s claims were about Darwinian evolution.
How do you figure living things are parameterized? Have you been reading Gary Gaulin?
I know what one of them is doing. 😉
Good. Maybe by the time you come back you’ll have figured out how the rest of us dummies communicate with blockquoting.
It’s just a 10 or so page summary; to me, it reads like someone having fun playing around with some mathematical simulations, not someone seriously interesting in understanding biology and evolution and what types of mathematical simulations are appropriate. The only biology references are Dawkins popularizations.
I hadn’t noticed that it was the short version.
Try this link
johnnyb,
I’d like to clarify what would meet your requirements before going to too much effort.
One possible example is Core Wars. There are several papers on evolving Redcode (an assembly language) programs that compete to take over a virtual memory. It wouldn’t surprise me to find that some of the resulting programs included iterative constructs. Redcode includes three instructions that could be used to implement iteration, JMZ (jump if zero), JMN (jump if not zero), and DJN (decrement and jump if not zero). Would you consider those too “designed” for evolution of a loop to be difficult?
Another possibility is genetic programming (GP). Unlike genetic algorithms, GPs operate on the abstract syntax tree directly. Much of the GP work is done in Lisp variants for that reason. As with Core Wars, it wouldn’t surprise me to find that GP has evolved iteration. Would a rich language like a Lisp be too “designed” for such results to refute your claim?
If your position is that an “open-ended loop”, which I take to mean an iteration control construct without a fixed number of steps, cannot evolve by known evolutionary mechanisms in a digital environment like one of these, I’ll happily spend the time doing some research.
DNA is not a programming language.
Just don’t forget the hacking you promised to do for me. You said it would not take more than a couple of weeks
peace
fifthmonarchyman,
I haven’t forgotten. I’m actually enjoying working on it, although real paying work is taking precedence. I’m generating the sample data from Yahoo! finance, the next step is setting up the model for training. So, about two weeks of work in total, spread over more calendar time than that.
There are still four random strings I provided to you waiting to be assessed, if I remember correctly….