Digital Irreducible Complexity - Author Thread

[WinstonEwert]

I recently had a paper published at bio-complexity: http://bio-complexity.org/ojs/index.php/main/article/view/BIO-C.2014.1

This is a thread to discuss the paper or ask questions of the author (me).

To participate in this thread, you must read the paper and come with an attitude of seeking to understand the paper's argument even if you don't accept its conclusions. Anyone who seeks to merely complain, mock, or disrupt discussion of the paper will be uninvited.

[johnspenn]

Winston,

Congrats on getting your work published. It all made perfect sense to me, although I am already convinced that life itself is irreducibly complex.

It seems to me like it's a lot of work to prove something that should be as obvious as the nose on one's face, but the work is necessary and thankfully there are smart guys like you to do it. Keep up the good work!

[DiEb]

Thanks for this opportunity to discuss the article! At Uncommon Descent, I wrote:

Winston Ewert, congratulations for your new paper. I just started to read it, but I already have a question about the last paragraph of your section on “Avida” (p. 3):

The parts in Avida are the individual steps in the process. If any of the steps in the process are missing, Avida will fail to calculate the EQU function. In this sense Pennock is correct, but we will discuss whether he is correct with respect to the other terms of Behe’s definition.

Isn’t the EQU function the irreducibly complex system, and Avida just the environment in which it dwells?

You answered:

Yes, Avida is the environment. However, I wouldn’t say that EQU is the system, rather it is the function that the system performs. I certainly didn’t mean to suggest that Avida itself was the allegedly irreducibly complex system.

For me, the system is the "digital organism", i.e., a "circular sequence of instructions". We have organisms which perform EQU, so this can be seen as the funciton of the system in the sense of Behe. Thus, system's components are the instructions in its circular sequence. If we answer the question whether all the components are required for the function (i.e., performing EQU) with yes, then this seems to be a good example for an irreducible complex system.

In this case, the function of the system can be identified easily. But generally, this is not the case. Take your example of the twin jet:

Commercial twin jet aircraft are required to be able to fly with only one functional engine. This means that the aircraft would continue to function if one of the engines were lost, albeit at reduced capacity. However, the airline does not fly aircraft with only one functional engine for reasons of safety; if the remaining engine cannot be repaired the aircraft will be dismantled. Does this mean that an aircraft with only one engine ceases to function? No, the aircraft with only one engine still works even if the eventual consequence is the dismantling of the aircraft.

You say that its function is flying. An equally valid description of the function of an aircraft is to "perform flights", and the jet has lost this ability, as it is generally not feasible to take off with just one engine.

Here, we touch a main difficulty of the "irreducible complexity": it is very hard to give a general definition. Like with the concept of "species", the term may be obvious in many instances, but very hard to describe for all. What is the function of a system? What are its parts? How many parts are there? These questions are not easily answered.

That's why I have a problem with your approach to Behe's definition: it sounds so exegetic ("Behe’s definition implies both a minimum of three parts and individual complexity of the parts.")

Especially for the genetic algorithms which you have presented, it seems to be difficult to get a consent: Look at the Steiner Tree. Its function is to connect the point with a minimum of road. Trivially removing an edge will destroy this function. But even if this is trivial, does this mean that it is not irreducibly complex? What degree of sophistication can we expect from a system whose genome exists just from a couple of numbers? What is interesting for me is that the seed population at the beginning of the experiment may have been exclusively 2-edge-connected graphs - or perhaps only disconnected graphs, and the trait of being irreducibly complex develops over time.

[WinstonEwert]

Dieb,

I think you do bring up a valid point with respect to parts and functions:

What is the function of a system? What are its parts? How many parts are there? These questions are not easily answered.

I would say that the decomposition and function of many systems is ambiguous, that is there are multiple equally valid decompositions/function descriptions. I did not discuss the point in the paper because it didn't naturally come up as part of the discussion of the computer models I was looking at.

The problem, I think, is easily avoided by saying that a system is irreducibly complex if it fits the definition of irreducible complexity under any valid decomposition/functional description. That is, if a system fits the definition under one decomposition, but not another, we consider the system to be irreducibly complex. Thus if it can be shown that a system is irreducible complex with respect to a given decomposition/function then it is considered to be irreducible complex. This avoids the intractable problem of trying to define a correct canonical decomposition.

But even if this is trivial, does this mean that it is not irreducibly complex?

If a shape has four right angles, but unequal sides, is it a square? No. A system that only fulfills part of the definition or irreducible complexity is not irreducible complexity. You can't shave off part of the definition and claim that your system is close enough. The Steiner tree example is too simple to exhibit irreducible complexity.

However, it seems that you don't think that having non-trivial complexity is part of the definition of irreducible complexity. But I pointed to the definition, Behe's examples, and Behe's arguments to show that Behe has assumed that all irreducibly complex system are made up of individually complex components. Behe as the originator of irreducible complexity gets to define what irreducible complexity is, and if he assumes complex components, than only systems of complex components are irreducibly complex.

[Roy]

Two quick questions:
1) Why does your calculation for Tierra group the individual instructions into 'genes', but your calculation for Avida not do so?
2) Why do you state that Dave Thomas's Steiner tree model contains only binary state objects when the additional nodes have a million possible positions?

Roy

[stcordova]

Winston,

The forum is very honored that you would host this discussion here. Congrats on the paper and the very fine work at the Evolution Informatics Lab.

As you know I was hoping to work with Dr. Marks in 2007, but then a week before the semester began, the lab was shut down. I'm glad the lab reopened, and it is quite evident the lab made a very good choice in selecting you (you were a superior choice, and it is probably provident I didn't work there since I don't think I would have had as much impact as you).

I have mostly praise for your article, and a few side questions and comments. When Dr. Marks and Dr. Dembski approached me in 2007, and what little association I had with the EIL, the one thing that I remember was how unreadable the Avida paper of 2003 was. At the time Dr. Marks and I couldn't make heads or tails of the claims because the wording was ambiguous.

One big item that I did not have a chance to pursue, but suggested to Dr. Marks, was to determine if Avida 1.6 was the version to the 2003 paper. The reason that was of interest is that I found that if the radiation parameter is set to max in 1.6, rather than killing the population, it keeps procreating! They claimed to have fixed it in future versions especially in light of the fact it was identified.

This echoes the observation in your paper that "there is no attempt to show that the parts are necessary for the working of the system". When I posted this bug at the old ARN forum, Adami sent Evan Dorn to try to calm the fire that I found a bug in Avida 1.6. If the source forge records can reasonably show that the 2003 paper was written in Avida 1.6 , this would be a major embarrassment, imho. So the question is:

"what version of Avida was used for the 2003 paper, were you all ever to ascertain it?"

The other major development that has happened since those days in 2007 is the now widespread acknowledgment that natural selection destroys IC in the wild (eyes, stomachs, wings, legs, paralogous redundant DNA, hemoglobin, and who knows what other proteins, etc.). If NS destroys IC systems, then it at least shows NS is under no obligation to build them in the first place.

All Avida showed was as you said:

Out of all the possible features that could be studied, the developers of Avida chose features that would be evolvable. They have deliberately constructed a system where evolution proceeds easily. They justify this by stating that it is required by evolutionary theory. However, the question is whether this requirement will be met in realistic cases, and Avida has simply assumed an answer to that question.

It assumes as true the very questions IC poses.

Regarding the Dave Thomas, I challenged him to solve a password. The only feedback was success or failure, if even that. It shows then there are a larger class of searches that Darwinian evolution cannot solve than the ones it can. Finding sets of functional proteins that bind to each other in signaling cascades is like finding passwords. It would be instructive to model objects not resolvable by Darwinian evolution. Of course it would be a trivial result but that is what Avida is as well. All this to say, is to echo the point, Thomas like the Avida authors chose evolvable systems, whereas the question of IC in biology is how evolvable the systems are to begin with. I could easily challenge Thomas with systems that are not evolvable, so his simulation proves nothing.

A question of clarification. The phase or word "co-option" was not found in your paper. I suppose the notion was implicit. Would you say Avida used a lot of "co-option" to construct functioning precursors to EQU? One thing I should point out that is related, the supposed cooption of the flagellum are not legitimate cooptions.

If there was a functioning password "soup" and another "opus", it would be pretty much improper to say one password coopted the letters of the other, but this is the sort of erroneous claim of cooption put forward in the Dover trial for the flagellum. Real cooption would be like the passwords
"supercalifra" and "supercalifagilistic".

It's been a while since I worked with Avida, but would you say they use cooption to achieve EQU, and is it real cooption where the proto-EQU's can be described as fragments of a working EQU?

Thanks in advance, and congrats again!



PS
Thanks also for you DNA Skittle comments in the other thread.

[WinstonEwert]

Roy,

Great questions!

Why does your calculation for Tierra group the individual instructions into 'genes', but your calculation for Avida not do so?

Primarily, those who claim that Avida evolves irreducible complexity take each instruction to be an individual part. I simply did my calculations following the definition of parts they had adopted. I could have also looked at Avida if we group the instructions, but there did not seem to be much point in setting up that straw-man just to knock it down.

Secondly, Tierra has a natural decomposition into genes. That is, there is actual structure in the genome and process that corresponds to the divisions being made. Avida doesn't have any such decomposition. There is no obvious way to break an Avida programs into any sort of sub-program.

So basically, I was just following the parts as defined by those who claim irreducible complexity. If somebody were to claim irreducible complexity based on some group of those instructions, I'd evaluate that separately a do the calculation for a more complex part. However, I don't think there is any sort of coherent way of dividing Avida programs into genes.

Why do you state that Dave Thomas's Steiner tree model contains only binary state objects when the additional nodes have a million possible positions?

You are correct that nodes have position information in addition to the bits controlling which nodes are connected. However, a functioning system is defined to be one in which all the cities are connected. For this, it doesn't matter where the nodes are located, it only matters which nodes are connected. That's controlled solely by the bits and not by the position.

[WinstonEwert]

Sal,

As you know I was hoping to work with Dr. Marks in 2007, but then a week before the semester began, the lab was shut down. I'm glad the lab reopened, and it is quite evident the lab made a very good choice in selecting you (you were a superior choice, and it is probably provident I didn't work there since I don't think I would have had as much impact as you).

Thanks.

what version of Avida was used for the 2003 paper, were you all ever to ascertain it?

This is the first I've ever head of this incident. So I've never made any attempt to ascertain which exact version was used. A quick look through the paper reveals:

Experiments ran using Avida version 1.6 on the Linux operating system on a Beowulf cluster of 64 Pentium III processors.

However, I only think thats problematic if the bug were to significantly change the result of the paper, and I don't think it would.

It's been a while since I worked with Avida, but would you say they use cooption to achieve EQU, and is it real cooption where the proto-EQU's can be described as fragments of a working EQU?

I would say that Avida uses cooption to achieve EQU. The proto-EQU's become fragments of the working EQU.

[Roy]

Primarily, those who claim that Avida evolves irreducible complexity take each instruction to be an individual part. I simply did my calculations following the definition of parts they had adopted. I could have also looked at Avida if we group the instructions, but there did not seem to be much point in setting up that straw-man just to knock it down.

Secondly, Tierra has a natural decomposition into genes. That is, there is actual structure in the genome and process that corresponds to the divisions being made. Avida doesn't have any such decomposition. There is no obvious way to break an Avida programs into any sort of sub-program.

So basically, I was just following the parts as defined by those who claim irreducible complexity. If somebody were to claim irreducible complexity based on some group of those instructions, I'd evaluate that separately a do the calculation for a more complex part. However, I don't think there is any sort of coherent way of dividing Avida programs into genes.

I disagree. Both Avida and Tierra 'organisms' consist of a sequence of instructions that are executed in order. Neither simulation* contains anything that groups those instructions into divisions or genes of any kind. Any such divisions result only from external analysis of the organism's working, or from the design of an original 'seed' organism. Avida 'organisms' can be deconstructed into 'genes' in exactly the same way as Tierra ones can; for example, when Pennock refers to the reproduction loop, or the diagram here.

Roy

*Or at least the versions I've run

[WinstonEwert]

Both Avida and Tierra 'organisms' consist of a sequence of instructions that are executed in order. Neither simulation* contains anything that groups those instructions into divisions or genes of any kind. Any such divisions result only from external analysis of the organism's working, or from the design of an original 'seed' organism. Avida 'organisms' can be deconstructed into 'genes' in exactly the same way as Tierra ones can; for example, when Pennock refers to the reproduction loop, or the diagram here.

100% agreed. In both cases, the simulation only recognizes instructions, and notion of genes is only in the minds of external observers.

When I suggested that Avida programs couldn't be broken up into sub-programs, I was specifically thinking of the instructions that compute EQU. I misstated when I said that there wasn't any way to break up an Avida program. There certainly is. But when looking at the allegedly irreducibly complex system, the EQU, there isn't an obvious coherent way to group the instructions.

I'm not even saying that its impossible to group the instructions into genes. But until someone defines such a grouping, the only way to evaluate the system is with the instructions as individual parts.