|
Author
|
Topic: Nature Refutes ID?: The Evolutionary Origin of Complex Features
|
Micah Sparacio
Member
Member # 6
|
posted 10. May 2003 20:43
As I start my second reading of this paper, here's one critical problem:
Consider the bacterial flagellum and the mousetrap. Removal of any one component of the system removes function in its current state. Of course, you could try to modify the mousetrap minus one part to achieve the same function. However, the key point is that given the structure of the system at time t1, if you remove one part of its IC core, it ceases function at time t2.
Now consider the fact that their entire system consists of only NAND primitives. Indeed, the EQU function consists of five NAND primitives. Remove one, and you've still got a logic function performed and a merit value of 16 (versus 32 in the EQU function). If the system were IC, you'd be left with a merit value of ZERO. There would be no function.
Notice that they admit, in the discussion section of the paper, that if the intermediates have no merit value, then the system doesn't find the EQU. But, if it was really IC that they were interested in getting, then at the very least we should expect that the removal of one NAND primitive rendered the function value 0.
IP: Logged
|
|
Pim van Meurs
Member
Member # 541
|
posted 10. May 2003 21:05
Micah: Notice that they admit, in the discussion section of the paper, that if the intermediates have no merit value, then the system doesn't find the EQU. But, if it was really IC that they were interested in getting, then at the very least we should expect that the removal of one NAND primitive rendered the function value 0.
But is that not begging the question? Why would removing of one of the primitives render the function value zero? Dembski's revision of Behe's IC definition may require such but perhaps such an assumption is not really reasonable? While original function may drop to zero, overall function does not because it fulfilss another essential role. This suggests why IC's assumption that removal of a component destroys (original) function seems meaningless if evolution works through co-option and hitchhiking rather than a tornado in a junkyard pathway.
Unless one wants to accept the ID premise, which needs to be shown to be true, one cannot require the function value to be zero. That seems to be exactly the problem with the IC argument. And certainly there need not be a forward-reverse symmetry in evolution, that is evolution need not be reversible. While the forward step may have high probability in reaching a beneficial, selectable step, the reverse need not be necessarily true. In fact the same seems to apply to protein networks, while network A may be close to B, B need not be close to A.
Since such occurrences seem to be quite common, IC ness may merely point to the irreversibility of evolutionary steps rather than to an evolutionary impossibility.
IP: Logged
|
|
Josh
Member
Member # 405
|
posted 11. May 2003 01:40
This has generated conversation on the ASA website. I really am only listening because I don't understand enough about these algorithms to critically analyze them. I would like to hear anyone's comments concerning the following exerpt:
Michael Roberts wrote, in part:
> This seems to me to give no more than Dawkins' computer models in the Blind Watchmaker.
It seems we have a rare point of agreement here. In the "Discussion"
section, I read the following:
------------------------ begin quote -------------------
Some readers might suggest that we 'stacked the deck' by studying the
evolution of a complex feature that could be built on simpler functions that
were also useful. However, that is precisely what evolutionary theory
requires, and indeed, our experiments showed that the complex feature never
evolved when simpler functions were not rewarded.
------------------------ end quote ---------------------
In other words, you have to stack the deck because Evolution won't work
unless the deck is stacked. But this is a circular argument and it proves
nothing about whether in fact the deck is stacked in this way in nature; all
they have done is make a simulation that takes that prior assumption for
granted.
Examination of the paper shows that they attributed higher degrees of
fitness to organisms that could perform more complex logic operations, with
the "reward" being 2^n where n was the number of logic operations combined.
The EQU function required 5 operations, so was rewarded with 32 points; but
intermediate rewards of 2,4,8,& 16 were also allowed for simple functions.
The fact that the complex feature could not evolve if the simpler functions
were not rewarded is a tacit admission that irreducibly complex systems
cannot evolve. With the intermediate rewards, the system is _not_
irreducibly complex. Knock out one of the 5 logic operations and there are
still 4 left, with a "reward" of 16 points, which is better than nothing and
gives the organism "energy" in order to reproduce. However, if you knock
out any of the components of Behe's mousetrap, then you have a
non-functional device.
The debate hinges around whether there really are such things as irreducibly
complex objects in nature & that is still open to question, one which is not
addressed by the simulation of the evolution of a non-irreducably complex
system.
In short, the simulation shows that a GA can do hill-climbing, provided the
gradients aren't too precipitous, but we knew that from the "Weasel"
simulation ages ago.
Iain.
IP: Logged
|
|
Micah Sparacio
Member
Member # 6
|
posted 11. May 2003 08:49
Pim,
If you look closely at what I said, and then what you said, you will see that this paper and this experiment did absolutely NOTHING to move the discussion about IC forward.
All the experiment accomplished was an assertion that IC systems don't exist. That is all. They accomplished this by assuming that all proposed IC systems have functional intermediates, but THIS is begging the question.
Look. There is NO question that if you remove a part of the IC core of the standard mousetrap that you buy at the grocery store, it will cease functioning. It will not catch mice. Sure, if your intention was to find an intermediate and to manipulate the remaining parts to get a functional intermediate, you could get something that worked, at least in the imagination. But that is not the point. The point is that if our only action is to remove one part, the system ceases to function. This is an IC system, with an IC core, which is necessary for the function.
Behe's definition of Irreducible Complexity:
A single system composed of several well-matched, interacting parts that contribute to the basic function of the system, wherein the removal of any one of the parts causes the system to effectively cease functioning.
To sum up: The EQU function does not meet even Behe's original defintion!!! When you remove one NAND primitive, you are left with four others that perform some other logic operation and which has a functional value of 16 (as opposed to 32 in the EQU).
This is the smooth path I was talking about. Any combination of NAND primitives gives you a logic function that can be evaluated to a metric. This is not the case in many engineered systems nor in many biological systems.
And notice, I'm not even assuming that biological systems never had intermediates. I'm only pointing out the very basic fact that many biological systems are experimented on via knockouts which leave the system non-functional.
The EQU function in this system doesn't even meet that very basic requirement.
[ 11. May 2003, 08:51: Message edited by: Micah Sparacio ]
IP: Logged
|
|
Erik
Member
Member # 160
|
posted 11. May 2003 10:06
Micah Sparacio wrote: "To sum up: The EQU function does not meet even Behe's original defintion!!! When you remove one NAND primitive, you are left with four others that perform some other logic operation and which has a functional value of 16 (as opposed to 32 in the EQU)."
Fine. Just remember that by your interpretation of Behe's original definition, a flagellum is not IC either. Flagella can perform other functions than propelling bacteria forward in a fluid. They are involved in protein secretion, virulence, and adhesion. Perhaps the motility function ceases if any one of a set of components is removed, but other functions can remain. Another thing that isn't IC, by your interpretation, is a mouse trap, since you can remove components and still have a functional paper clip.
Regardless of which IC definition is prefered this week, the simulation proves that a function (EQU) can evolve by cooption of different functions. The authors reported:
quote:
"Of the 35 instructions required for EQU, 22 were needed for simpler functions. Three instructions required for EQU were also essential for replication; these were conserved from the ancestral sequence, as were five others. However, 27 of the 35 instructions required to perform EQU were evolutionarily derived, and all but one of them had appeared in the line of descent before this function was ever performed. Thus, although more than two dozen mutations were used to build EQU, undoing any one of them destroyed this function."
The responses to suggestions that flagella evolved by cooption of other functions have to this day been dismissed by ID advocates as unlikely. The simulation proves that there is no a priori why cooption is unlikely. It doesn't matter if you respond to this fact by interpreting the definition of IC so strictly that flagella and other known biological features are excluded or by admitting that IC can evolve. The testable part of the IC argument is now dead and cremated. But who knows, maybe its untestable soul will live on forever.
Erik
IP: Logged
|
|
charlie d.
Member
Member # 159
|
posted 11. May 2003 10:16
Micah:
actually, this is very instructive, because it illustrates the hopeless confusion due to the arbitrary definition of "function" in IC.
This is what Demsbki just recently said about it: quote:
A system is irreducibly complex in Behe's sense if all its parts are indispensable to preserving the system's basic function. That an irreducibly complex system may have subsystems that have functions of their own (functions distinct from that of the original system) is therefore allowed in the definition.
Thus, to go back to the AVIDA simulation, it is perfectly fine that some of EQU's subcomponents have functions, and that those functions provide a selective advantage. What matters is that the EQU logical function ceases to work in the absence of a number of those subcomponents (the "core") - which it does. Similarly, the flagellum, functionally defined as an "outboard motor" also is IC, despite that some of its subcomponents may very arguably function as secretion systems, pili, adhesion structures, membrane channels etc.
On the other hand, you are right that if one interprets the function of EQU to be more simply "to provide a selective advantage in the simulation environment", then of course even non-EQU subcomponents would do so, although to a lower extent, and EQU wouldn't be IC anymore. But if you adopt this definition, you also have to apply it to the flagellum, and admit that its function is "to provide a selective advantage to bacteria", which then its subcomponents (channels, secretions systems, etc) would also be able to do, to some extent, and the flagellum would not be IC either.
[EDIT: Erik just made the exact same point - oh well, repetita iuvant]
[ 11. May 2003, 10:18: Message edited by: charlie d. ]
IP: Logged
|
|
Micah Sparacio
Member
Member # 6
|
posted 11. May 2003 13:34
Hey guys, good points. I'll have to think a bit about them. For the time being, could you help me out with the following (not being a biologist, I just don't know...)
Is it true that in reality (in a lab with real kicking and screaming bacteria) if a knockout is performed for one of the IC core parts of a flagella, that some form of the system still develops and still functions in some way?
If I remember correctly, Scott Minnich indicated that flagella development ceased when errors occurred: in other words, that the bacterium shuts down production to save resources if it receives feedback indicating that the flagellum system has errors.
Of course I could be misunderstanding Minnich here. For the time being, I'm just trying to understand how it works biologically.
IP: Logged
|
|
Nel
Member
Member # 614
|
posted 11. May 2003 13:51
Charlie,
How does that statement by Dembski show that the function of IC is arbitrary? Of course subcomponents of a machine can have other functions, the fact that we can even talk of sub-functions shows that it is not arbitrary. Biological function may not be precise, but it is useful.
To anyone,
I havn't read the paper, just the various posts, I plan to get a bit more involved in this thread this week, but I was wondering what exactly is the point of saying that the NAND and EQU functions are IC, when it was written by the authors and not developed by the virtual organisms? (AFAIK it contains instructions that enable the organisms to replicate, and the organisms subsequently shuffled this around). Now it seems that EQU is IC, but not by much, 3 instructions were essential but the rest seems to be redundant, unless I'm reading that quote wrong.
[ 11. May 2003, 14:02: Message edited by: Nelson_Alonso ]
IP: Logged
|
|
Micah Sparacio
Member
Member # 6
|
posted 11. May 2003 14:02
Well, Nelson, I'm still up in the air as to whether to count the EQU function as IC.
But, I will give you some reasons for considering it at least relevant to ID.
1. The EQU function meets Dembski's independence criterion. It is independently specifiable.
2. The EQU function does something that its subcomponents are incapble of doing
3. Even man-made objects can be considered as IC systems (the mousetrap)
My concerns though not well articulated, are something like this:
The ontology and heuristic of the system lend themselves to the acheivement of the target set
The ontology consists of NAND primitives
The heuristic is looking at functional combinations of NAND primitives
When you combine NAND primitives together, you get logic functions which evaluate positively
I'm in the process of developing an analysis paper of the Nature paper. I hope to have it up over the next three weeks and as I develop my ideas, I'll throw them out on this forum, hopefully to get ripped apart by those who are more intelligent than me;-)
IP: Logged
|
|
Roger R
Member
Member # 200
|
posted 11. May 2003 14:06
Erik,
quote:
What the experiment proves is that Irreducible Complexity (in the sense of the 1996 definition) can evolve by cooption of different, simpler functions.
That would seem to be going out on a limb this early in the analysis. It also seems to be beyond what the authors claim, since I didn't see any discussion about IC in the article, merely complexity itself.
As for cooption, it is not a priori unlikely, but maybe in the real biological world it is. From the article:
quote:
Some readers might suggest that we ‘stacked the deck’ by studying the evolution of a complex feature that could be built on simpler functions that were also useful. However, that is precisely what evolutionary theory requires, and indeed, our experiments showed that the complex feature never evolved when simpler functions were not rewarded.
I think many IDers might agree with this: evolutionary theory requires that the simple path exists, without too many large gaps in said path. The question is, does that path exist in the biological world? This program doesn't elucidate that question.
[ 11. May 2003, 14:09: Message edited by: Roger R ]
IP: Logged
|
|
Nel
Member
Member # 614
|
posted 11. May 2003 14:07
Micah,
I don't mind calling it IC, I realize that even man-made objects can be IC, I would just like to know how much of this is actually IC and how much of this ICness can be gleaned from the end result, that is, if EQU is IC, and it was written by intelligent agents, it seems to be moot for ID critics to argue it is IC, since they want to show that an IC system can likely evolve. Was this ability to evolve the end result front-loaded by intelligent design? Or is EQU not IC but just an example of CC (cumalitive complexity -- I doubt it, however, the evolutionary derived versions may be nothing but CC built upon the originally designed IC)? Or is EQU IC but not by much (IS irreducibly simple) and can this be gleaned from the end result. (I"m assuming that the authors started the organisms with functional replicators, and the organisms subsequently evolved their own version of EQU with parts of the original EQU. The question is, if the function of replication itself is simple, then it's possible to glean that from the evolved version, it is not possible to do this with bacterial flagella because there is no such thing as a simpler bi-directional rotary motor-driven propeller)
In other words, something like this is irrelevant to both ID and Darwinism: Start with two logic function with no evolutionary history at all, (from the organism's perspective that is, the programmers are the Intelligent Designers), subsequent evolution only improved speed of these intelligently designed functions, then it's irrelevant to the ICness of the bacterial flagellum (well it's irrelevant anyway since they are not genetic sequences but it's fun).
For example, I find it hard to believe that the 3 essential instructions were conserved, but that the evolutionary derived ones make up , what? Another IC system? Seems as though the 3 instruction set is IC, and the rest is just CC (redundancy). What they probably mean by essential is that the organism might become less fit (approach it's ancestral state) if some of this evolutionary derived state was lost, but it did not kill them, indeed, I think the author found it very difficult to kill them, even when he wanted to. And in fact, I think the paper mentions several points where the organisms simply became less fit. Again, the true IC system can probably be gleaned from the end result, as design theorists are doing for the flagellum.
But with IC, essential doesn't mean it works poorly, essential means it does not work at all. If the 3 instructions were absolutely essential and the rest was just perks that added degrees of fitness, then the only free lunch here was eaten by intelligent design.
[ 11. May 2003, 14:43: Message edited by: Nelson_Alonso ]
IP: Logged
|
|
|
|
RBH
Member
Member # 380
|
posted 11. May 2003 14:56
yersinia wrote quote:
The source code and executable for Avida can be downloaded by following the links on their page, and the parameter files they used as input are also available. I presume one could replicate the whole thing on a PC if you used the same random number seeds (which they specify on the page also).
Be aware that they ran their work on a 64-Pentium Beowulf machine. I run Avida on a 2.4 GHz PC, and I'd hate to try to replicate the whole experiment. One could, but it would be a time- and cycles-intensive project. However, when I have time and the spare cycles I plan to retest some of the key aspects of their work.
Micah deplored the speed with which people have embraced the outcome of the reported work. That's not really amazing, given that it is wholly consistent with what biologists have been saying for some time: possible evolutionary pathways to given structures are not singular or unique; populations are not necessarily direct climbers of a single monotonic hill (a la MESA) but juke and deke around on complex landscapes; and the possibility of an evolutionary pathway cannot be rejected by examining only the end product.
In my opinion, that last point is among the most interesting aspects ot the study. The files preserved during the runs allowed tracing the exact lineages of the outcome EQU programs (not merely an instruction, as YZ2 suggests, but sequences of primitive instructions comprising assembly language programs). All of the lineages in the 23 runs that produced EQU were different - there is no single evolutionary pathway to that function. Moreover, each outcome EQU genotype was different from the others. That has substantial implications for the probability estimates used in attributing "complexity" in the Dembskian sense, by inflating the numerator, the size of the target set.
The research also shows that sweeping the field clean of 'regularity' hypotheses is nowhere near as easy as Dembski would have us believe. The requirement to eliminate (deny the existence of) regularity hypotheses in order to treat the alternative to design as chance-assembled discrete combinatorial objects is much more difficult to defend now. In fact, it cannot be defended on formal grounds now.
That one cannot reject the possibility of an evolutionary pathway based on looking just at the end product is a critical point, and makes the notion of "irreducible complexity" much harder to sustain. Those who would would disregard or discount the study must somehow refute the following:
1. Start with a population of simple replicators that perform no logical functions, but merely replicate themselves.
2. Make available raw materials (a set of primitive assembly language instructions) that if arranged appropriately could form an irreducibly complex object.
2. Add random mutations of biologically plausible kinds; a fitness function; and selection based on competition for a scarce resource.
3. Run for 100,000 or so cycles.
Using only the mechanisms invoked by neoDarwinian evolutionary theory, a non-trivial proportion of the populations will evolve an instruction sequence that is irreducibly complex in Behe's 1996 definition. That is the central finding, and it is a fatal blow to the too-glib claim that those kinds of structures cannot evolve by Darwinian mechanisms. They plainly can.
An avenue worth exploring is tracing the implications for specifed complexity, via its connection with irreducible complexity. If irreducibly complex structures can be evolved by Darwinian mechanisms, what do we make of specified complexity? It's clear that "complexity" goes away - the kinds of structures that can be evolved have high probabilities and hence are not complex in the Dembksian sense. If irreducible complexity and specified complexity go away, what remains of the design detection apparatus?
As Micah and others think about how they might defend their discounting of the implications of study, let me call one or two other considerations to their attention.
First, while the externally imposed fitness function was fixed in the study, the fitness landscape on which a population in Avida evolves was not static. In that population, very large numbers of different gene strings arise through several sorts of mutations, and large numbers of species - clusters of different genotypes that are functionally identical - also arise. At any given moment, a particular genotype's likelihood of surviving and extending its lineage depends not only on its fitness with respect to the extrinsic fitness function, but also with respect to the (ever-changing) composition of its competitor species. A genotype that might be very (reproductively) successful in one competitive context will be less successful in another competitive context. Reproductive fitness is thus partly context-sensitive, and the context changes continuously as new competitiors emerge and evolve. One cannot legitimately invoke the image of a population wandering around on a static landscape climbing its fixed hills.
Second, the concept of cooption of 'simpler' functional sequences is very important. Critics of the study will have to show that the employment of cooption by the populations in the study is somehow biologically illegitimate. As the study shows, cooption of sequences evolved to perform simpler functions is a critical component of the evolutionary pathway of IC structures, just as biologists have argued. (Incidentally, that's why much of the commercial GA work (along with MESA) cannot be generalized to questions of biological evolution except with great caution. Those kinds of GAs seldom enable the possibility of cooption. This study has already inspired one of my engineers to begin to look at its utility in our commercial GAs.)
Third, critics will have to contend with the observation that not every mutation in the lineages that lead to EQU was "beneficial" in the sense of immediately getting closer to some goal state or higher on some slope of a fitness landscape. Too often an evolving population is treated as a point on a fitness landscape, when in fact variability means that within a population there is an array of individuals of greater and lesser fitness, and the latter are as (or perhaps more) important than the former when fitness landscapes are dynamically deforming.
Finally, suppose that the extrinsic fitness function initially rewarded only simpler logical functions for the first 50,000 cycles, and then the experimenters suddenly decided to see if a more complex logical function could evolve and added a reward for EQU. Would artificial organisms capable of performing EQU have emerged from the cooption of previously adapted simpler logical functions? My bet is yes. The evolutionary process could have taken advantage of the previously evolved functions exactly as it did in the experiments reported. In other words, the final "target" did not have to be encoded in the fitness function from the beginning for it emerge in the end. The possibility of EQU did not have to be front-loaded in the original conditions of the experiment.
This last point might be taken to mean that nothing about the initial conditions was critical. Obviously that's wrong - the initial conditions had to include the raw materials out which genomes peforming EQU could be constructed if evolutionary processes could assemble them appropriately. But no information about EQU itself need be encoded in the primordial Ancestor or in the raw materials. That 'information' emerged because evolutionary mechanisms used the materials available, in a number of different ways, to perform EQU. A fascinating follow-on would be to do the same experiment with different instruction sets. One wonders about the range of initial conditions that would enable EQU to emerge.
RBH
One or two minor edits to compensate for eyes tired from being up until 4:00 a.m. on various emergency squad runs, semi-dozing through church, and then writing in too-small a font.
[ 11. May 2003, 15:05: Message edited by: RBH ]
IP: Logged
|
|
charlie d.
Member
Member # 159
|
posted 11. May 2003 15:11
Just quickly:
- The EQU program is not written by the operators (it's not a predetermined, fixed "target" - the logical EQU fucntion is, which can be obtained in many different ways -just like "bacterial motility"). IIRC, every single successful run of the simulation (not all were successful in reaching EQU) achieved EQU function in a different way, oftentimes using unpredictable (to the operators) computational strategies.
- More than just 3 instructions are necessary for EQU, and eliminate it if knocked out. The 3 mentioned in the quote just happened to be part of the replication instructions present originally in the starting program, and were coopted unchanged into the EQU function. As the quote says, however, most of the other instructions were evolutionarily derived, and provided independent advantages.
- Nelson, it was not Dembski who was talking about arbitrariness, of course. Critics of IC and CSI have pointed out the problem of arbitrary definitions of function from the very beginning. Erik and I just remarked how this was a perfect example of this fundamental problem: if you define the function in term of what the system does right now, both EQU and the flagellum are IC (and at least EQU can evolve by mutation/selection processes and cooption, apparently). However, the way Micah had defined the EQU function (providing a selective advantage to the evolving and competing programs), neither EQU nor the flagellum are, because intermediate systems providing advantageous functions exist for both. The fact that Micah, despite being a very smart guy and certainly an ID-savvy polemicist, had not noticed this contradiction himself shows precisely how arbitrary these functional definitions are.
IP: Logged
|
|
Roger R
Member
Member # 200
|
posted 11. May 2003 15:46
charlie d.,
quote:
The EQU program is not written by the operators (it's not a predetermined, fixed "target" - the logical EQU fucntion is, which can be obtained in many different ways -just like "bacterial motility").
Putting aside the issue of whether the EQU function is indeed a fixed target, I am more troubled by your assignment of the function EQU as being analogous to "bacterial motility". One could argue that "logical function" is analogous to "bacterial mobility", and the difference in the two is quite significant in terms of the IC aspects of the discussion.
If the latter is more appropriate, then clearly IC is not evolving, since logical operations, though not as heavily selected for, can occur 1 step back from the EQU function. We would have complexity, the only claim made by the authors, but not IC.
[ 11. May 2003, 15:47: Message edited by: Roger R ]
IP: Logged
|
|
|
Printer-friendly view of this topic