|
Author
|
Topic: Evolution of complexity
|
Frances
Member
Member # 169
|
posted 22. October 2002 00:45
Recently increased interest into the evolution of complexity has resulted in a variety of very interesting papers. In this thread I propose to explore the contributions of these findings to Intelligent Design. The reason I started looking into complexity and entropy is the idea to explore the simularity and differences between Shannon entropy and thermodynamical entropy. For instance, thermodynamic entropy can only increase in a closed system but what about an open system. Adami has used computer simulations to show how a 'Maxwell demon' can extract information from its surroundings and pass it on to a system.
The amount of information in a certain population A stored about the environment is:
H(max)(A) - H(A|E) and in case of sequences this can be shown to be equivalent to
L - H[p_i]
L: sequence length
p_i: probability of finding genome i (given environment E)
H(max) seems to be very similar to the CSI measure of Dembski which uses a uniform random distribution function to calculation the amount of CSI in a sequence.
So the question really becomes, how is this information inserted into the genome, and what prevents it from escaping.
Adami proposes an interesting concept, the Maxwell Demon. According to Adami during evolutionary adaptations which increase fitness wrt the environment, information about the environment is gained. So just like Intelligent Design can affect the probabilities of occurrences of a sequence in a certain environment, this Maxwell demon transfers information from the environment into the population. In fact the Maxwell Demon seems to be 'natural selection', which creates low entropy populations by selecting information (accepting neutral and beneficial mutations) and rejecting detrimental ones. Thus the conclusion is that the complexity of adapting populations can only increase. But that is not the whole story since the assumption is that the environment remains the same. In fact a time varying environment can lead to decrease in information when the adaptation loses its function or becomes detrimental by itself. Other exceptions include co-evolution and sexual reproduction.
Adami continues to show the effects of this Maxwell Demon in various simulations.
Adami's results can be found at the Santa Fe Institute
Biocomplexity and digital life I
Biocomplexity and digital life II
Biocomplexity and digital life III
I believe the Maxwell Demon may very well be the solution to why algorithms, just like intelligent designers can inject specified complexity into a population. It's the information equivalent of the thermodynamicl 'heat pump'.
Glenn Morton raised a similar issue on ASA
This may very well explain the findings by Tom Schneider who showed how evolutionary mechanisms seem to be able to increase information in the genome.
Some very interesting work is being done in this area
So Dembski may be right 'information is being smuggled in'. But not by the researchers but by the environment.
[url=John Collier, Research Associate,
Department of Philosophy
]John Collier[/url] has some very interesting papers on this topic.
For instance in Information increase in biological systems: How does adaptation fit? Collier argues that the information of adaptation is the mutual information between the biota (Adami's organisms) and the environment.
quote:
The standard account of fitness and adaptation can be rephrased in terms
of information theory. From this, an information of adaptation can be defined in terms of a
fitness function. The information of adaptation is a measure of the mutual information between
biota and their environment. If the actual state of adaptation lags behind the state of optimal
adaptation, then it is possible for the information of adaptation to increase indefinitely. Since adaptations are functional, this suggests the possibility of progressive evolution in the sense of increasing adaptation.
The suggestions make a lot of sense to me when realizing that if one can argue that intelligent designers can affect the probability distribution within the organism's genetic sequence why variation and selection cannot achieve similar results. Of course there is a main difference that the intelligent designer(s) have the ability to bypass certain restrictions in evolution.
Which leads me to the inevitable conclusion that in order to differentiate intelligent design from natural design we need to formulate the hypothetical pathways and determine if we can find evidence in favor of one or the other pathway. It's interesting to notice that intelligent design is not much different from the mix of 'regularity' hypotheses and chance. Like regularity, intelligent design can be non-contingent. But like the world of nature around us, intelligent design has also some factors of unpredictability and chance in them.
The main question is, how do we differentiate between two mechanisms that could be very similar. That is how do we separate natural design from intelligent design? I argue that the only way to do this is to formulate hypotheses, find supporting evidence and/or falsification so that in the end we have what we consider the most likely hypothesis explaining a certain system. We certainly cannot distinguish between the two forms of design from merely looking at the end product.
[ 25. October 2002, 15:53: Message edited by: Frances ]
IP: Logged
|
|
warren_bergerson
Member
Member # 262
|
posted 22. October 2002 11:00
Frances,
I wasn’t able to access the site on "biocomplexity and digital life’ so I don’t know the performance characteristics of the Adami’s simulations. I do know, however, that you can define deterministic logic machines capable of simulating the generation of huge volumes of complexity/information. [I will be happy to either post (something like 8 pages of code) or share privately the pseudo code for such a logic machine.]
The issue, when you look at processes or systems which produce information, is performance. It takes an extremely powerful, efficient system/process to produce/maintain the volume of information/complexity associated with even simple life forms.
If you are suggesting that Maxwell demon’s or RM&NS processes can produce the bio-complexity observed in biological systems, then you need to demonstrate that these processes have the capacity to produce the observed level/volume of complexity. Demonstrating the ability to generate information or complexity is not sufficient. You also need to demonstrate that the process can generate the appropriate level of complexity(under appropriate constraints).
Quote Adami: The standard account of fitness and adaptation can be rephrased in terms
of information theory. From this, an information of adaptation can be defined in terms of a
fitness function. The information of adaptation is a measure of the mutual information between
biota and their environment. If the actual state of adaptation lags behind the state of optimal
adaptation, then it is possible for the information of adaptation to increase indefinitely. Since adaptations are functional, this suggests the possibility of progressive evolution in the sense of increasing adaptation.
You state that the above is based on the assumption of a stable environment. If you replace that assumption with a more realistic dynamic environment, you get some interesting results. First, if a system is to remain adaptive in a dynamic environment, it requires a greatly increased capacity to generate new information(because the information in the system is continually becoming obsolete.) As you point out, a dynamic environment will produce both devolution and evolution.
IP: Logged
|
|
Frances
Member
Member # 169
|
posted 22. October 2002 13:32
Warren,
Thank you for your response which suggests that at least in principle deterministic machines can generate information. Such a finding is very important for the intelligent design discussion since Dembski argues that CSI can only be generated by Intelligence.
Of course the next step will be formulating and testing various hypotheses. But you have to be careful with confusing terms like increase/decrease in information with evolution/devolution.
As far as time dependence of the environment, indeed this adds a complication to the whole issue in that if the environment changes too fast, the organism cannot 'adapt' thus there is a filter which removes high frequency fluctuations, retaining the low frequency components.
A time varying environment also has in principle far more information available to be extracted.
Added to clarify Warren's questions below
I did respond to your comment on the need to demonstrate performance capability. But that is not the issue here, that's a what science will do when it formulates and test hypotheses. What I am trying to establish is if CSI is a reliable indicator of design and if that can exclude natural design.
I am confused, I propose formulating hypotheses and testing them. That does involve determining if information increases. As far as whether or not a model can generate information, see Adami for instance "Evolution of Biological Complexity;
Proc. Nat. Acad. Sci. 97 (2000) 4463-4468 (with C. Ofria and T. C. Collier). "
You suggest without much foundation that evolution has different meanings. First of all what is the relevance of such a claim? The claim is whether or not non-intelligent design can generate CSI. Second of all, evolution is the change in frequence in alleles. That evolutions as the processes producing changes in species is different is hardly self evident.
It may be helpful to look into the following What is evolution
Macro evolution
When you state that 'life routinely adapts to very high frequency changes' you are using adaptation in a sense not directly relevant to the discussion since adaptation requires in a Darwinian sense genetic changes. Unless the information gets transfered to the gene or is inheritable, such information is essentially lost.
It would be helpful not to confuse the meaning of evolution as used in genetics here since it would allow for exactly the confusion you refer to in your posting about confusion in the meaning of the term evolution.
[ 22. October 2002, 15:02: Message edited by: Frances ]
IP: Logged
|
|
warren_bergerson
Member
Member # 262
|
posted 22. October 2002 14:33
Frances,
I note you did not respond to the comment on the ‘need’ to demonstrate performance capability. The issue of ‘substantiated’ and ‘unsubstantiated’ performance claims is certainly central to the issue of ‘what can be explained by known processes’.
Quote: Thank you for your response which suggests that at least in principle deterministic machines can generate information. Such a finding is very important for the intelligent design discussion since Dembski argues that CSI can only be generated by Intelligence.
The deterministic machines would, it seems, be far more relevant to information theory.
Quote: Of course the next step will be formulating and testing various hypotheses. But you have to be careful with confusing terms like increase/decrease in information with evolution/devolution.
Of course the next step is NOT hypothesize testing. The next step would be for someone with appropriate mathematical knowledge to evaluate the model to determine if it does in fact generate information as claimed. Using the deterministic model as the basis for a scientific theory is a separate issue.
The term ‘evolution’ has several definitions. Evolution in the genetic sense refers to changes in the frequencies of alleles in population. Evolution as the ‘process producing changes in species’ is a very different phenomena. The term evolution is used to refer different types of adaptive change processes in biological systems. Finally, the terms evolution/devolution are used to refer to the development/deterioration of information in an abstract system. Imprecise/ambiguous use of the term evolution is pretty much SOP.
Quote: As far as time dependence of the environment, indeed this adds a complication to the whole issue in that if the environment changes too fast, the organism cannot 'adapt' thus there is a filter which removes high frequency fluctuations, retaining the low frequency components.
Life forms routinely ‘adapt’ to very high frequency changes in conditions(organisms can and do often adapt to changes with frequencies in the millisecond range). It is not organisms, but ‘theories’ that have trouble with high frequency changes.
IP: Logged
|
|
Frances
Member
Member # 169
|
posted 22. October 2002 16:22
I would like to explore the following statement of mine in somewhat more detail
quote:
That is how do we separate natural design from intelligent design? I argue that the only way to do this is to formulate hypotheses, find supporting evidence and/or falsification so that in the end we have what we consider the most likely hypothesis explaining a certain system. We certainly cannot distinguish between the two forms of design from merely looking at the end product.
I just ran across an excellent paper by Sober
The essential argument is that:
quote:
The idea that evidence is essentially a comparative concept is often associated with the Law of Likelihood:9
Evidence E favors hypothesis H1 over hypothesis H2 if and only if
Pr(E * H1) > Pr(E * H2).
A good example is given just above this quote. Lets assume that you own two urns with white balls. You now randomly pick an urn and want to test if the urn contains 2% white balls. YOu pick a ball which happens to be white. Is this evidence in favor or against the hypothesis? As Sober shows, the original argument by Richard Royall shows that if the other urn held 0.00001% white balls that drawing a white ball would be in favor of the first hypothesis rather than against it.
Thus it seems essential in hypotheses testing that we can determine the likelihood of the event wrt the hypotheses.
So we get to the next step in my argument, what can we do to fix the explanatory filter? I believe thatWilkins and Elsberry have provided a way in their The advantages of theft over toil: the design inference and arguing from ignorance
Their first improvement to the filter is to allow "we don't kmow' as an explanation.
quote:
If we can say of a problem that it is currently intractable or there is insufficient information to give a regularity or chance explanation now, then the Filter tells us we must ascribe it to design if it is specifiable. But it can be specifiable without the knowledge required to rule out regularity or chance explanations. This is clearly a god of the gaps stance, and it can have only one purpose: to block further investigation into these problems.
The next improvement is to incorporate design hypotheses in determining the probability of an event
quote:
That knowledge makes E a HP event, and so the filter short-circuits at the next branch and gives a design inference relative to a background knowledge set Bi available at time t. So now there appears to be two kinds of design - the ordinary kind based on a knowledge of the behavior of designers, and a "rarefied" design, based on an inference from ignorance, both of the possible causes of regularities and of the nature of the designer.6
I would argue that this is how archaeology and forensic sciences work.
After all as I argued both regularity and design share some very important features namely that the likelihood of an event E based on the hypothesis (regularity/design) is high. It's this high probability that lets us infer regularity/design.
So what is rarefied design? Rarefied design refers to design for which we do not have any hypotheses, but rather is based on an inference from ignorance.
quote:
Rarefied design inferences tell us nothing that can be inductively generalized. Consequently, analogies between artifacts of ordinary design, which are the result of causal regularities of (known) designers, and the "artifacts" of rarefied design do not h
This leads to the following improved filter
Now a philosophical view of the
argument
Michael J. Murray argues very similarly to Wilkins et al namely that
quote:
[quote]
Suppose that there is a certain non-nomically determined fact that has no known
explanation; suppose that one can think of a possible explanation of that fact, an explanation
that (if only it were true) is a very good explanation; then it is wrong to say that that event
stands in no more need of an explanation than an otherwise similar event for which no such
explanation is available.4
[quote]
On this principle, we look at events which exhibit patterns not explained by law and see if a
plausible alternative (read “design”) explanation can be given. If so, we should regard the event as
likely designed. If not, we should regard it as a result of chance.
Murray adds another interesting argument that may help us understand the limitations of the EF as far as detection of intelligent design is concerned. Murray distinguishes between modalities of design, intervention and 'deck stacking'. Other terms have been used to describe deck stacking such as 'front loading'.
Murray however argues that 'deck stacking' is not different from 'purely nomically regular processes'.
Thus leading to the conclusion that
quote:
If all we have access to is apparently designed outcomes, we cannot distinguish between those that result via “law” (deck-stacking) and those that result from “design” (intervention).
[ 22. October 2002, 18:57: Message edited by: Frances ]
IP: Logged
|
|
Janitor@MIT
Member
Member # 125
|
posted 23. October 2002 11:04
“The ultimate analysis of Maxwell’s demon must involve a definition of intelligence, a characteristic which has been all to consistently banished from discussions of demons carried out by physicists. On the other hand, intelligence has been—since Turing and his famous test—often invoked in the discussions of computer scientists. To convince ourselves (and the intelligent demon) of the limits imposed by the second law we shall, following [Zurek, W.H. 1989. Phys. Rev. A 40: 4731-47512 & Zurek, W.H. 1989. Nature 347: 199-124.], adopt an operational definition of intelligence which arose in the context of computation. It is based on the so-called Church-Turing thesis [References omitted.]… Using the Church-Turing thesis as a point of departure, the present author has demonstrated that even this intelligent threat to the second law can be eliminated—the original ‘smart’ demon can be exorcised (Zurek, W.H., “Algorithmic Randomness, Physical Entropy, Measurements, and the Demon of Choice,” http://arxiv.org/pdf/quant-ph/9807007, ©by Wojciech Hubert Zurek, March 12, 1998.).”
In the “Demonology and Biology” topic I noticed that intelligence was not necessarily an advantage in the never-ending quest to repeal the SLT. (This “quest” is almost a sort of biological imperative, isn’t it?) Since Adami et al, Schneider, etc. all qualify as “intelligent demons,” Zurek gives me reason to suspect that they have failed also in their attempts. (Their research needs to be more closely examined. As far as I’m concerned, they might as well be claiming they’ve repealed the law of gravity or contacted the shade of Darwin.) Now, I’m never one to discount ingenuity. I’ve seen some pretty amazing things. So I’ve come to suspect that the second law can be cheated, rather than defeated. But don’t ask me how. (‘Cuz even if I knew, I wouldn’t tell. LOL)
The problem seems obvious to me: evolution must proceed in exactly the way it proceeds in these experiments (but which the experimenters take little notice of). Evolution proceeds by “wasting” an enormous amount of information, by discarding it, averaging, approximating, etc. It’s a high throughput process, with small outputs. For every bit of information incorporated in the genome, a squillion bits are discarded and lost forever. Incalculable and truly staggering amounts of information are lost in this process.
This is exactly how Darwin naively conceived of it. But he did not understand the probabilistic implications. We’re not going to change that. These experiments are poorly conceived, but at least they recognize, if even only implicitly, the problem that Dr. Dembski (among others) has pointed out. (One might even say that these scientists are conducting an ID research program!)
Some questions: When we say that the genome encodes information about its environment, what exactly do we mean? I mean, can we isolate a gene that records the ambient temperature/pressure, or the acceleration of gravity at the Earth’s surface? No one thinks this, do they? So it must be a “kind” of information. Information comes in “kinds” like everything else. What kind of information?
Let’s say that a stray peptide in the environment docks at a receptor in a cell’s surface membrane. What happens next is a very interesting cascade of events inside the cell. We might say that the cell/genome “interprets” or “translates” a signal. (It is performing computations.) Now we have a saying, “Something is lost in translation.” We might just as well say, what is just as true, that something is added in translation. But that’s exactly the problem with treating the genome and its evolution as bit strings filling buffers in a computer. Everything is lost in translation here, most importantly, this very process of translating or interpreting that is so fundamental to life. (For an interesting take, see e.g., the article by Hillol Kargupta in the ISCID Bibliography.)
Suggestion: maybe we could conceive of evolution not as a rudimentary (Irreducibly simple?) communication system (source-channel-destination), but supply what is missing in this framework—an interpreter-complier, a decoder. Rather than life as the virtually lifeless subject of evolution, we might reverse the logic here: evolution as the object of life? Too weird?
EF thoughtexperiment: You are a scientist conducting an experiment, perfectly routine experiment. All you are doing is refining a measurement, adding n+1 digit to a physical constant that is already known with high precision to the nth digit. But then your experiment returns gibberish, totally unexpected results. “These numbers are all wrong!” What is the first thing you think?
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 23. October 2002 12:09
Hi, Janitor. You wrote quote:
As far as I?m concerned, they [Adami and Schneider] might as well be claiming they?ve repealed the law of gravity or contacted the shade of Darwin.) Now, I?m never one to discount ingenuity. I?ve seen some pretty amazing things. So I?ve come to suspect that the second law can be cheated, rather than defeated. But don?t ask me how. (?Cuz even if I knew, I wouldn?t tell. LOL)
Here's a place to start learning some ways of (locally and temporarily) 'cheating' the SLT: Exploring Complexity. (Gregoire Nicolis & Ilya Prigogine, Freeman, 1989). They move Prigogine's analyses of far-from-equilibrium thermodynamic systems in physics 'up' to chemistry and biology.
You also wrote that evolution is an enormously wasteful process, discarding squilliions of bits of information, and quote:
This is exactly how Darwin naively conceived of it. But he did not understand the probabilistic implications. We're not going to change that. These experiments [Adami's and Schneider's] are poorly conceived, but at least they recognize, if even only implicitly, the problem that Dr. Dembski (among others) has pointed out. (One might even say that these scientists [Adami & Schneider] are conducting an ID research program!)
If one takes your remarks seriously, they imply that every single scientific experiment on any phenomenon whatsoever is part of "an ID research program." After all, every scientific experiment is 'designed' by an intelligent agent. If I do an experiment on some properties of water am I conducting an experiment on the intelligently designed properties of an IC (water molecule) system? Seems to me you've got some category confusions going on here.
I'd sure like to see some explication of "poorly conceived." In what respects are the experiments poorly conceived and how would you correct them? Generic and allusive (and elusive!) critical remarks are one thing; constructive criticism with concrete suggestions for improvement is another (and harder) thing to do. How about taking a shot at the latter? ![[Smile]](smile.gif)
RBH
IP: Logged
|
|
Janitor@MIT
Member
Member # 125
|
posted 23. October 2002 12:52
I am familiar with Nicolis & Prigogine. Here is a particularly "unfriendly" critique, Lavenda, Bernard H. 1978. Thermodynamics of Irreversible Processes. Dover Publications. NY.
I'll dig up my notes on Adami's experiment published in Science, I believe. But my question about what kind of information is contained in the genome applies. As does my question about the dynamics of the interaction between evolving populations and their environment--I'm not the only one who's questioned the scope and relevance to biological evolution of "simulations" performed in closed and highly simplified systems.
These simulations and systems don't support any sweeping claims such as are implied here. But that's a very big question now isn't it?
Schneider simply contradicts himself, as was noted in the "Demonology and Biology" topic, buried somewhere in back pages.
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 23. October 2002 13:36
Janitor,
That's a real interesting thread and will take some digesting. Here's the URL in case anyone else is interested.
RBH
Added in edit so this post is not completely content-free: I will note that I agree that many simulations of evolutionary processes in the form of various evolutionary algorithms are often oversimplifications to the point of near-caricature. An adequate EA that has a chance of telling us something useful about biological evolution will take into account at least the following (copied from my recent post on ARN): (1) knowing that a population evolves on several fitness landscapes simultaneously and those landscapes are not topographically equivalent; (2) the fitness landscapes are high-dimensioned; (3) the fact that populations themselves form part of the selective environment of their members; (4) the fact that organisms are inefficient at utilizing environmental resources; (5) the fact that in sufficiently complex and large-scale evolutionary algorithms one observes 'speciation'; (6) the fact that mutations are random with respect to current selective pressures; and (7) some kinds of mutations are more complicated than simple point (bit-flipping, base-substitution) mutations.
But then, the use of GAs to model biological processes, like another endeavour I could name, is heavily dependent on analogies from human engineering and design.
[ 23. October 2002, 13:45: Message edited by: RBH ]
IP: Logged
|
|
Frances
Member
Member # 169
|
posted 23. October 2002 14:56
Lets step back and take a pause at what I my argument is wrt the Maxwell Demon. Dembski argues the CSI cannot be created by chance or regularity.
Dembski comments on the Maxwell demon It is CSI that enables Maxwell's demon to outsmart a thermodynamic system tending towards thermal equilibrium (cf. Landauer, 1991, p. 26
It is natural selection that behaves like a Maxwell Demon, increasing the information in the genome:
From Adami
quote:
Christoph Adami, Charles Ofria, and Travis C. Collier's Evolution of biological complexity, Proc. Natl. Acad. Sci. USA, Vol. 97, Issue 9, 4463-4468, April 25, 2000.
To make a case for or against a trend in the evolution of complexity in biological evolution, complexity needs to be both rigorously defined and measurable. A recent information-theoretic (but intuitively evident) definition identifies genomic complexity with the amount of information a sequence stores about its environment. We investigate the evolution of genomic complexity in populations of digital organisms and monitor in detail the evolutionary transitions that increase complexity. We show that, because natural selection forces genomes to behave as a natural "Maxwell Demon," within a fixed environment, genomic complexity is forced to increase.
Source
It is interesting to hear that the experiments by Adami and Schneider are 'poorly conceived'. I would certainly like to hear more about this. I am familiar with Dembski's 'arguments' about Schneider. It seems that Dembski admitted that CSI was generated but he claimed that it was generated by the 'fitness function'.
Of course the Maxwell Demon is 'expensive' and 'wasteful'. In Schneider's experiments half of the population dies in every generation or as Schneider shows 5500+ deaths per bit. Which leads him to conclude that
quote:
So the evolution is, not surprisingly, a rather inefficient information generating mechanism
Source
As far as Schneider and the Maxwell Demon
quote:
Equations (15) and (26) say that for every bit of decision Maxwell's demon makes, it must dissipate at least (1 bit) x eps_min joules of energy into the surrounding environment. It can dissipate more, but it cannot dissipate less, since to do so would violate both Shannon's channel capacity theorem and the Second Law.
In context the statement [url]The demon violates the Second Law by decreasing the entropy of a system without compensatory dissipation. Maxwell's demon is dead.[/url] is quit clear, unless the Demon has a way to dissipate energy/information, the demon would be violating the 2nd law of thermodynamics.
So the statement by Janitor that
quote:
So, Schneider is well aware that “Maxwell’s demon is dead.”
is with respect to a Demon that has no compensatory mechanisms.
In case of the program eV the loss of 5000 lives to flip one bit seems to provide you with the answer.
So lets get back to Adami and Schneider who have found how information in the genome can increase. One may very well claim that there is some fitness function which plays the role of the Maxwell Demon, sorting through the available information but then in principle natural selection and mutation seem to be sufficient mechanisms to generate CSI.
But lets move on and look at what objections were given for Murray and Sober's arguments.
quote:
only legitimate reason to limit all design to front-loaded design is if there could be no empirical grounds for preferring interactive design to front-loaded design. Michael Murray in his recent paper "Natural Providence" for the Wheaton Philosophy Conference (October 2000, www.wheaton.edu/philosophy/conference.html) attempts to make such an argument. Accordingly, he argues that for a non-natural designer front-loaded design and interactive design will be empirically equivalent.
Yet I do not believe that Murray is arguing for limiting design to front loaded. Still, it is an open question whether or not all miracles could be accounted for via deck-stacking.23
I believe that Murray is arguing the following
quote:
The relevance of this observation should be obvious. As we saw earlier, IDT advocates,
most notably Behe and Johnson, argue that design inferences are defeated if the processes that lead
to the designed outcome are nomically regular. This is a mistake. Even if the designed outcomes
can be explained by appeal to the regular operations of the laws of nature, inferences to design can still be warranted
[/url]
and
[quote]
The second implication, the flip-side of the first perhaps, is that if all we had access to was
the outcomes of the poker games, there would be no way in principle to discern whether or not the
cheating occurred via deck-staking or intervention.
Thus if Murray is correct that the EF cannot distinguish between 'deck stacking' and 'intervention' then it cannot eliminate nominally natural causes.
Thus Murray suggests that
quote:
This means, of course, that Dembski’s explanatory filter requires further adjusting. As we will see however, the required adjustment will force friends of IDT to abandon intelligent design as a paradigm of natural science inconsistent with methodological naturalism.
So what needs to happen? According to Murray
quote:
So perhaps disciplinary territorialism should not rule out Intelligent Design as a genuinely
scientific explanation. But we are not out of the woods yet. For even though countenancing design
as an explanation might in principle count as genuine science, it cannot if the design hypothesis is not empirically distinguishable from explanations which appeal only to the natural powers of natural substances. If such empirical distinguishability is not possible, then there is no scientifically respectable way, by IDT’s own lights, to defend intelligent design as an explanation distinct from law and chance.
If we assume that the outcome can be explained by either 'deck stacking' or 'intervention' but we have no access to the actual sequence of events that led the the outcome then there are no empirical grounds favoring explanations via law over explanations via design.
Dembski argues that the Cambrian explosion could be some evidence of 'intervention' but the problem still remains that we cannot study the sequence of events that led to the Cambrian explosion thus we cannot eliminate on empirical grounds explanations via law.
In fact while the Cambrian explosion may appear to be without any precusors recent findings suggest that the pre-cambrian fauna was hardly that absent. In fact both fossil evidence as well as phylogenetic evidence shows that the Cambrian was not that 'abrupt'. But lets assume for the moment that there were no precursors to be found, what would this mean for design? Since there may be very well a natural explanation for this such as a catastropic event which eliminated all fossil evidence before the Cambrian, how can we determine if this was natural or design that was at work?
Dembski's objections do not help us resolve the issue. So we get to Sober's suggestions about the need for competing hypotheses. As Sober has shown, in order to reject hypotheses one needs to know what other hypotheses are relevant. In his example, finding a white ball may very well be a confirmation rather than a rejection of the first hypothesis that the urn contains 2% white balls, when the alternative hypothesis were that the other urn contained 0.00001% white balls.
Without knowing what kind of design may be expected from non-natural designers, how can we formulate any such hypotheses?
Dembski argues that in archaeology no such knowledge is required but I would argue to the contrary. Through our knowledge that 'arrow heads' are commonly found among other evidence of 'civilization' such as pottery, fires, bones we can build a picture of the 'designer' of these arrow heads. Dembski argues that designers are unpredictable but does that mean that designers are totally unconstrained? I do not believe that such an argument can be made unless one presumes a godlike designer.
But we need to get back to Sober's original argument that unless we have a competing hypothesis it is impossible to determine if a certain observation is in favor or against a particular hypothesis.
The same is argued by Wilkins and Elsberry by distinguishing between rarefied design and 'design' for which we have background information.
And to show that I am not ignoring Warren, here is an interesting paper on the 'two faces of fitness'.
quote:
The concept of fitness began its career in biology long before evolutionary theory was
mathematized. Fitness was used to describe an organism’s vigor, or the degree to which organisms
“fit” into their environments. An organism’s success in avoiding predators and in building a nest
obviously contribute to its fitness and to the fitness of its offspring, but the peacock’s gaudy tail seemed
to be in an entirely different line of work. Fitness, as a term in ordinary language (as in “physical fitness”) and in its original biological meaning, applied to the survival of an organism and its offspring, not to sheer reproductive output (Paul Cronin 1991). Darwin’s separation of natural from sexual selection may sound odd from a modern perspective, but it made sense from this earlier point of view.
[ 23. October 2002, 19:11: Message edited by: Frances ]
IP: Logged
|
|
warren_bergerson
Member
Member # 262
|
posted 23. October 2002 16:37
Janitor,
Quote: The problem seems obvious to me: evolution must proceed in exactly the way it proceeds in these experiments (but which the experimenters take little notice of). Evolution proceeds by "wasting" an enormous amount of information, by discarding it, averaging, approximating, etc. It’s a high throughput process, with small outputs. For every bit of information incorporated in the genome, a squillion bits are discarded and lost forever. Incalculable and truly staggering amounts of information are lost in this process.
It may possibly be true that the earliest or most primitive biological design processes where highly inefficient and wasted vast amounts of information. This is clearly not true of the biological design processes that exist in life forms today. Humans routinely solve very complex adaptive problems with a minimum of effort. Living systems may have started out as very inefficient, ineffective random variation/direct selection, but they clearly developed or evolved into highly efficient information generating systems.
Quote: Some questions: When we say that the genome encodes information about its environment, what exactly do we mean? I mean, can we isolate a gene that records the ambient temperature/pressure, or the acceleration of gravity at the Earth’s surface? No one thinks this, do they? So it must be a "kind" of information. Information comes in "kinds" like everything else. What kind of information?
‘Information’ in biological systems would appear to be in the form of ‘teleological causal relationships’ of the general form "input(cause) produces output (effect) which increases likelihood of goal(survival)". The ‘input causes output’ part of information at time t can be denoted by Dxt where Dxt is a member of some possible set of causal relationships D and also a member of some functional or purposeful subset Dft of D. Genes or alleles represent only a portion of the information used by cells.
Quote: Suggestion: maybe we could conceive of evolution not as a rudimentary (Irreducibly simple?) communication system (source-channel-destination), but supply what is missing in this framework—an interpreter-complier, a decoder. Rather than life as the virtually lifeless subject of evolution, we might reverse the logic here: evolution as the object of life? Too weird?
Even better, we should view evolution as an artifact of a very powerful, effective, efficient information generating/utilizing processes that maintains cells in adaptive states in a rapidly changing environments.
If you are going to look at life forms in terms of design and information, then it is essential to look at design/information in terms of realistic time frames. The assumption/perspective that a life form is some type if information generating logic machine operating with a processing speed of one cycle per life time is absurd.
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 23. October 2002 16:41
I wrote above that quote:
An adequate EA that has a chance of telling us something useful about biological evolution will take into account at least the following (copied from my recent post on ARN) ...
followed by a list of 7 conditions. Beware of interpreting that posting as invalidating the conclusion that Schneider drew from his work with ev, to the effect that an evolutionary algorithm can generate information. That conclusion stands, as far as I can see. I've just re-read much of that material, and I see no good reason for rejecting it as "poorly conceived."
Another problem I'm having is trying to tell the difference between the occasions when Janitor is referring to local and temporary 'violations' of TSL, and when he is talking about global (universal?) violations. I'm failing. Yet I am certain Janitor knows that the difference is not trivial, and that keeping it straight is real important for clarity in discussion. Knowing that he knows this, I keep looking for something deeper in his remarks and the references he supplies, and I can't. What am I missing?
RBH
IP: Logged
|
|
Frances
Member
Member # 169
|
posted 23. October 2002 22:23
Somewhat of a side issue yet relevant if we want to evaluate whether or not CSI can be generated by natural mechanisms.
It was pointed out to me that my statement that H(max) were similar to Dembski's CSI needs some explanation.
It is not totally clear to me with respect to what assumption CSI needs to be calculated. Is it wrt the likelihood of the event given the hypothesis? But then regularity nor intelligent design seem to be able to CSI since the probability is 1. Or is it the probability of the event with respect to a uniform probability distribution as suggested in NFL. Thus a sequence of L tail-head sequences will have L bits of CSI? The latter one is suggested when looking at the prime number 0/1 sequence in NFL. Perhaps we could ask Dembski to take some of his undoubtfully very sparse moments of free time to help clarify the issue at hand.
On the one hand Dembski seems to suggest that
quote:
What this means is that even though with respect to the uniform probability on the phase space the target has exceedingly small probability, the probability for the evolutionary algorithm E to get into the target in m steps is no longer small. And since complexity and improbability are for the purposes of specified complexity parallel notions, this means that even though the target is complex and specified with respect to the uniform probability on the phase space, it remains specified but is no longer complex with respect to the probability induced by evolutionary algorithm E.
[End Quote - WA Dembski, "Specified Complexity", MetaViews 152]
But that calculation should find similarly that an event is not CSI anymore if an Intelligent Design hypothesis is applied. So when does one apply the uniform probability and when does one apply the actual probability and since P(E|Intelligent design) = 1 for intelligent design, does that not indicate that the event was not CSI after all?
It was also pointed out to me that my usage of the term "Maxwell Demon" is confusing since it may suggest that evolution requires the violation of the 2LOT. I would like to stress that neither Schneider nor Adami use the Maxwell Demon in the meaning of violating the 2LOT. In fact Schneider is quite clear that the Maxwell Demon requires the loss of enery elsewhere to compensate for the decrease in entropy. When Adami uses the term "Maxwell Demon" it is in the context of the mechanism of the Demon without suggesting a violation of the 2LOT.
A friendly voice also whispered the following link into my ear
Self-reproducing System can Behave as Maxwell's Demon: Theoretical Illustration under Prebiotic Conditions
quote:
Once the stability of a self-reproducing system is established in the above sense, the self-reproducing system can operate as Maxwell's demon to regulate the outside flow of matter by its catalytic function without any contradiction to the second law of thermodynamics.
The following paper addresses "Self-Organization of Template-Replicating Polymers and the Spontaneous Rise of Genetic Information"
[ 23. October 2002, 22:51: Message edited by: Frances ]
IP: Logged
|
|
Janitor@MIT
Member
Member # 125
|
posted 25. October 2002 11:26
Since I’m being called to account for my choice of words, “poorly conceived:”
The Adami et al paper is about how genomic information/complexity is correlated with environmental information/complexity. First, how “complex” is their model of the genome and the environment? On the face of it this is just a “space-filling” model. Bits fill buffers. They seem to recognize that their model is comparatively “simple,” but captures some “rich” dynamics (even given the representational limits and the arbitrarily bounded artificial environment). They appear to want to warrant some simplicity in the model (which is supposed to model biological complexity) with some equivocal observations about C-value paradox and neutrality. Unfortunately, in their own words the apparent complexity of genomes does not seem (impressionistically) to be directly correlated to the complexity of the organism—at least not for computer modeling purposes (LOL). Why then should I grant the premise that the complexity of the genome is correlated in some significant way with the environment, when I can’t even correlate the complexity of the genome with its primary environment, the organism in which it exists?! (The basic premise I’m willing to grant, but which they have now given me only reason to doubt.) Maybe, first things first?
Some drastic (cosmetic) surgery is being performed on genomes here. This is a very general criticism (repeated by many) of the whole field of EC as it is related to biological modeling. No doubt developments in one discipline are going to lag relevant developments in another. But ~two decades of developments in biological research and theory are being ignored here. E.g., research by the bundesfraktion has raised significant questions about the “canonical model” (adopted by EC from popgen) of discretely Darwinized genes diffusing over Sewall Wright’s landscape. Treating neutrality as a “never mind, makes no difference” is no longer acceptable in models of biological evolution. Neutrality has a significant effect on the search dynamics, i.e., how populations search their environment, so it obviously should enter the equation genome complexity = environmental complexity.
The authors say that the (natural) environment is extremely complex. No doubt. Lewontin, for one, has argued that for all practical purposes it might as well be infinitely complex! Any strong correlation between finitely complex genomes and “infinitely” complex environment is going to be difficult to establish convincingly (i.e., statistically). Especially as the basic tools of analysis here, information and computer theory, require finite fields of play. (Problems of computability, decidability, and satisfiability must be solved in the course of biological evolution. These are not just academic problems or complications for our analyses. They must be solved for life to survive and evolve.)
Modeling and simulation are limited in what they can accomplish, what they can tell us. Understood. It is therefore incumbent upon the modeler to make some explicit accounting for the factors that are discounted or ignored in the analysis that may have some bearing on the validity or applicability of the results. Complexity is, well, its complex, now isn;t it? But the very basic and widely shared premise stated at the very top of this article is that complexity is basically simplicity itself. Maybe we should consider more seriously how such a premise shapes our understanding of the world. But that's a big question and far beyond the scope of the paper. Science forges ahead even with some very big questions unanswered. What else can we do?!
This presents us with more than an analytically vexing problem. Taking an “epistemological” perspective (as the authors almost do following Deutsch), one might wonder if much of the complexity of the environment might be “selectively” (epistemically) eliminated and compressed by some “intelligent” adaptive strategies implemented by organisms? But, if that’s a possibility, then any strong, straightforward (statistical) correlation between genome and environment might not exist. But that’s a bit of misplaced literalism, because biologists don’t generally really believe that genomes “know” their environment. (Nonetheless, there is a general solution to the problem in engineering design theory. But not one in evolutionary theory that I am aware of.) Basically, the idea that a genome would simplify (and not complexify) its relation to its environment as an evolutionary-adaptive strategy, in the same way that these researchers simplify their models of organisms, environments, and evolution, is unacceptable theoretically and not even given much consideration, even as a possibility. (This is not to say that genomes become simpler vis a vis the environment, but that they become more complex in a very differnt way than is explored in this model.)
But, basically the premise is dogged by the analysis, the limits of the analysis. That correlations exist is not seriously doubted here (i.e., by me). That correlations exist as “mere numbers” in a table the researcher compiles is seriously doubted. And that simply points to the limits of information and computer theory for biological modeling. Many interesting discussions about the limits of these tools were held here at ISCID (more earlier than lately). I’ve related it directly to the question I’ve asked about what “kind” of information is contained in genomes. There is a “kind” of information, with which we are all intimately familiar, but which has received virtually no information theoretic treatment. If genomes consist largely of this kind of information, then this model, which doesn’t consider it, except to effectively eliminate it as complicating the analysis, is just “poorly conceived.”
What kind of information is contained in the genome and is it subject to meaningful statistical analysis?
And then there is the “abuse” of the original and historical usage of the term “Maxwell’s demon,” which means basically the opposite of this “redefinition.”
To anticipate the inevitable objection that I’m expecting more than what the authors intended, let me just remind the participants what a solution to the demon paradox means: It would revolutionize our knowledge and understanding of statistical physics and thermodynamics, require a fundamental revision of information and computer theory, and basically supply a limitless supply of free energy to fuel the world’s economy!
(Not to mention explain the evolution of biological complexity.)
Poorly conceived? My opinion.
[Sorry, I’m ignoring some (many) questions. But I’ll get back to ya.]
IP: Logged
|
|
Frances
Member
Member # 169
|
posted 25. October 2002 15:21
Dear Janitor
The issue was how information was being added in this admittedly simplified simulation of evolutionary mechanisms. The answer was stunningly simple: A Maxwell Demon like mechanism.
I understand that you were confused by the use of the term Maxwell demon but the demon is the same as the original version but with the additional required loss of information/energy required. The Maxwell Demon may have been used in different context but the context of this Demon is quite well explained.
Your suggestion what the solution to the Maxwell Demon paradox would mean seems to be contradicting your expectations, no free energy after all.
You ask 'what kind of information is contained in genomes' again Adami and Schneider are quite clear what form of information they are discussing: Shannon information. Interestingly enough Dembski seems to consider Schneider's experiment to have generated CSI but 'blames' the selection method.
So lets not forget what the experiments did and did not do. They 'merely' showed that Shannon information in the genome can increase under mechanisms of mutation and selection and provide us with an explanation for how this happened.
How does Adami define complexity
quote:
To make a case for or against a trend in the evolution of complexity in biological evolution, complexity needs to be both rigorously defined and measurable. A recent information-theoretic (but intuitively evident) definition identifies genomic complexity with the amount of information a sequence stores about its environment.
Their equation (4) defines complexity in a tractable manner as the sum of all Shannon information of all sites.
In Schneider's experiment the information is more easily tractable than in Adami's experiment.
How complexity relates to the environment is explained Here
Amount of information in population A stored about environment E is
H_max - H(A|E) =
L -H[P_i]
L: Sequence length
P_i: probability to find genotyp i given environment E
So I still am confused about what you consider poorly conceived about the Shannon notion of information/complexity and its use by Schneider/Adami?
The model may be a simplification but it addresses some important issues such as how information may enter the genome. What kind of information is stored in the genome and is it open to statistical analysis: That's what Schneider and Adami are all about. Schneider has applied these concepts quite succesfully to real life examples.
The Maxwell Demon is a thought experiment in which the molecules are 'sorted' thus apparantly violating the 2LOT. The solution lies in realizing that information/energy needs to be expended by the Demon to do his job. Thus no violation of the 2LOT is required.
In Schneider's example, the balance is equalized through the large amounts of deaths per bit of information.
Complexity may be defined url=http://www.santafe.edu/~shalizi/notebooks/complexity-measures.html]in a large variety of ways[/url] but Adami and Schneider have selected a relatively easily tractable version namely Shannon's information
The following link provides some information about Shannon and information theory.
As far as Shannon information and CSI, Dembski gives us some insight:
quote:
What I do is define a type of information--specified complexity--that enriches Shannon information but at the same is not reducible to it.
Not the most helpful statement but we can work from it as Schneider has attempted:
The Following link may help clarify natural selection's role
Curious has raised some interesting questions which need to be incorporated in my analysis of the EF and my proposal for a better filter/approach.
IC is often used as an example of a system that shows evidence for design but I believe that the argument is flawed in the simple sense that although it is argued that direct darwinian routes are impossible it is also admitted that indirect routes may exist or that perhaps a non-Darwinian mechanism may be at work. Either way the existence of the possibility of a natural pathway for IC does not help us establish the likelihood of intelligent design. As Sober has shown, hypotheses can only be evaluated by comparing two or more hypotheses.
quote:
My critique of the intelligent design movement has been based on the comparative principle I
stated about evidence – to say whether an observation counts as evidence against evolutionary theory and in favor of the hypothesis of intelligent design, one must know what each predicts about the observation.
An eliminative approach may seem enticing to detect intelligent design but it seems to introduce more problems than it solves.
I would propose that 1) the meaning of certain terms as used in the EF are clarified: Is CSI calculated wrt the uniform chance hypothesis or wrt to the actual probability of the event happening given a particular hypothesis? If it is the former then CSI is merely the common probability argument so often used against evolution by assuming chance, if it is the latter then how can ID generate CSI when the probability for the ID hypothesis is quite high and thus the CSI quite low?
What am I missing?
[ 25. October 2002, 18:52: Message edited by: Frances ]
IP: Logged
|
|
|
Printer-friendly view of this topic