|
Author
|
Topic: Does anything at all exhibit specified complexity?
|
Rex Kerr
Member
Member # 632
|
posted 17. February 2004 15:28
Stephen:
(a) Humans can, through design and organization, create increases or decreases in Kolmogorov complexity. Dembski uses "complexity" but he doesn't mean it in the Kolmogorov-Chaitin sense. My claim is that humans cannot be shown to increase complexity in a Dembskian sense in any unique way.
(b) Other organisms can also use design and organization to alter KC complexity, but likewise cannot be shown to increase Dembskian complexity beyond the UPB.
(c) Due to the problems I've described here, Dembski's method will only detect faulty or missing assumptions without elucidating the nature of these missing assumptions. The Luria-Delbruck experiment isn't directly relevant; it just demonstrates what the default mode of change is, and doesn't rule out other mechanisms in special cases.
Before I go any further, I should ask what you mean by "SC"? Specifically, what is a specification and what is complexity? I would agree with a good deal of what you say if these terms are to be understood in a naive sense and not as Dembski defines them. But Dembski's filter operates on his definitions.
Erik: I suspect that you are right about mystifying intelligent agents. I haven't focused on that point here because I'm not sure ID proponents realize that Dembski's method requires it. To the extent that they do realize it, I agree that the assumption of the mysteriousness of intelligence is the appropriate target.
IP: Logged
|
|
The Deuce
Member
Member # 992
|
posted 17. February 2004 16:13
Have probabilities ever been attached to human invention in any sort of rigorous way? To my knowledge they have not, but that's the particular kind of activity we're interested in here, not the chances that a person will or won't respond positively to some particular suggestion (a simple binary decision that can be easily measured). Human intelligence, unfortunately, can't be made any less mysterious than human invention. And if by "mysterious" you mean "not defined in terms of mechanical physical processes", and if teleology is itself a real form of causation, then it may be in principle impossible to define it in a way that is not mysterious in those terms.
I agree, though, that Dembski could afford to de-mystify the concept of intelligence somewhat. Even if the definition can't be reduced to blind processes, I imagine that he could identify certain marking attributes that intelligence possesses which make it an appropriate hypothesis in cases of SC (perhaps this could also make the attribution of intelligence to things other than people less analogous).
As an aside, I don't consider mechanical causation, or chance causation, to be any less mysterious that intelligent or telic causation. In the cases of both mechanical and chance causation, we can't really define what they are except in their own terms. We simply accept that they exist because we observe them happening and we get used to it. But I have just as much evidence (perhaps more) of telic causation from observing myself making decisions and coming up with ideas, and I see other people doing the same thing.
IP: Logged
|
|
Stephen Wright
Member
Member # 195
|
posted 17. February 2004 17:39
quote:
Before I go any further, I should ask what you mean by "SC"? Specifically, what is a specification and what is complexity? I would agree with a good deal of what you say if these terms are to be understood in a naive sense and not as Dembski defines them. But Dembski's filter operates on his definitions.
Now we are tracking together. I put a little time into my first post on specifications and thought it was a reasonable attempt. K-C or in the ISCID listing of definitions; Connectivity and Descriptive/Interpretative Complexity are fine as naive views to express my perspective, but as I said before complexity is familiar but elusive as a concept.
Let me think about it a while and I can respond better. I would like to describe my outlook as pragmatic but with an out of the box bent. Information would be real and have structure when made complex to serve a targeted function, in my humble worldview. I would like to stay away from "magic" emerging both from intelligence and from matter self-organizing with impossible odds.
IP: Logged
|
|
Erik
Member
Member # 160
|
posted 17. February 2004 17:45
The Deuce asked: "Have probabilities ever been attached to human invention in any sort of rigorous way?"
1. My comments were in reply to John Bracht, who seemed to argue that there are conceptual problems with describing actions of intelligent agents in terms of probabilities. There may or may not be problems with arriving at non-arbitrary numerical values for such probabilities, but that's not what my comments were about. (This, of course, shouldn't prevent you from raising the point; I simply want to preempt potential imprecision and misunderstandings.)
2. I would agree that it can be very difficult to arrive at non-arbitrary numerical values for probabilities of events involving human invention. The problems do not arise because of any phenomenon specific to humans or invention, though. The problems arise because we have a poor idea of what the set of possible outcomes looks like and of the constraints on, and biases inherent in, human thinking. It would be comparatively easy (for an adult!) to estimate the probability that a nine-year-old solves a problem that requires inventiveness of a nine-year-old, but whose space of adequate solutions is well-known to an adult. The task would also be made easier if we knew something about the nine-year-old's individual competence profile. It is comparatively difficult to estimate what kind of technological inventions that will be invented in the next 50 years, because we do not generally have a good idea of what kind of technology that will be feasible and desirable.
The same difficulties arise in all probability-estimation problems where we have a poor understanding of what the set of possible outcomes looks like and of the underlying outcome-generating process. It is, for instance, difficult to assign sharp, non-arbitrary probabilities on how the details of the climate will change in the 50 years because of global warming.
3. Dembski's method has never been applied in any sort of rigorous way, except possibly in less than a handful of toy problems. See "Design Inference Game II" for a notable lack of rigorous applications. Will you teach the participants in that thread how to make rigorous probability estimates for hypotheses not involving intelligent agents?
As for causation being mystical, I would rather say that our common notion of "cause" is actually a complex of many different definitions, some of which are reasonable, while others are incoherent or mutually inconsistent. Therefore I usually avoid that concept except in informal everyday situations where imprecision is affordable. I think there is much to gain by banishing the notion of "cause" from science. But that's probably a topic for another thread.
Erik
IP: Logged
|
|
John Bracht
Member
Member # 5
|
posted 17. February 2004 23:09
Erik,
I guess I'm not so much arguing against statistical trends in intelligent agents' actions (I acknowledged such and suggested voting as an example) but rather as a way of understanding how individual intelligent agents make decisions. I get most of this intuition from watching myself. As I said in my post, I don't think about probabilities when I make decisions; I consider various rationales, motivations, past events, desires, needs, etc, into my decisions. As such, to try to describe the process of decision making as a probabilistic outcome seems like a category mistake. As I argued before: how do you account for the fact that the founding fathers might have chosen a virtual infinitude of alternatives to writing the declaration of independence--how do you account for the possibility that they just got drunk instead--or comitted suicide? In my case--what is the probability that I would sit down and write a post analyzing these ideas? Well, it has to do with a fairly large number of other circumstances in my life, like having the time, desire to do so, having already made one post, having someone respond to my original post, etc.
As I said before, once an agent makes a decision to carry out an action, that action gets carried out with a high degree of probability. But the process of making the decsion seems like a non-probabilistic process in its essence. As I said before, how do you attach probabilities to reasons, desires, emotions, and needs that will all factor into a decsion? What's the probability of a "reason"?
So what is probability? I guess I'm thinking here of probability as applying to some stochastic (chance-driven) process, whereupon varying outcomes occur with characteristic frequencies but in undetermined ways (ie, they're not determined by laws). So I guess this is where I think the breakdown comes when applying to intelligent decision making. Decision making is highly rational, intentional, thoughtful (hopefully!) and there really isn't any element of chance or randomness to it. But it's not really deterministic either--my choices aren't fully determined by my circumstances, since I can choose to do the completely illogical thing or the less-rational of many choices.
From the outside you can talk of the probability of a person performing action "X"--but that doesn't really capture how that person decided to perform that action.
The final point (and one that Rex seems to have completely missed) is that Dembski's explanatory filter is completely fine with intelligently-caused events having a high probability. As I said before, the only way an intelligently caused event can happen is by an intelligent cause. And once an intelligent cause decides to act, it does so with high probability. From the outside, our probability function must more or less acurately map onto the motivations, needs, wants, desires, etc, of the intelligent agent before it will accurately predict what they will do, but if we can make it match up it shouldn't surprise us if intelligently caused events occur with a high probability--they are, after all, intelligently, deliberately caused. And Dembski's filter is all about separating the intelligently caused events from all the rest, so it only worries about whether an event has a low probability relative to non-intelligent causes. To include intelligence as one of the causes of the explanatory filter really voids it of meaning, since you've included all possible causes in the filter--you'll just find that all possible events map onto that set of causes. Definitely true, but uninteresting.
John
IP: Logged
|
|
Rex Kerr
Member
Member # 632
|
posted 18. February 2004 01:26
John wrote:
quote:
Dembski's filter is all about separating the intelligently caused events from all the rest, so it only worries about whether an event has a low probability relative to non-intelligent causes. To include intelligence as one of the causes of the explanatory filter really voids it of meaning, since you've included all possible causes in the filter.
Exactly! This is why the filter doesn't filter anything!
Because we have to ask: why separate out intelligently-caused events?
Suppose we apply the filter, but leave out different classes of causes: leave out gravity; leave out Newton's Third Law; leave out quantum chromodynamics; leave out intelligence; leave out proteins; leave out evolutionary algorithms.
The filter will have a tendency to tell you that something is missing in every one of these cases, depending on the examples you pick. My question is: what privileges intelligence to be the output of the filter, and neglected from the possible causes?
(Note: it may be difficult to use statistical trends to understand how individual intelligent agents make decisions, but you don't need to know that to calculate a probability, just as you don't need to know anything about the rigid-body mechanics of collisions between plastic cubes and wooden planes in order to calculate the probability of rolling dice--you just make observations and characterize them. We do this all the time with human behavior--test-making, puzzle-creation, selecting passwords, etc..)
Deuce: I'm not sure we have to assign probabilities to human invention that rigorously. We seem to do a reasonable job at estimating probabilities of meeting specifications in a semi-rigorous way, and this can be used to at least set bounds on the probability of an event. (In what context? As I just mentioned: test-making, puzzle-creation, selecting passwords, etc..)
Stephen: keep in mind that my claim about a lack of specified complexity is only made for Dembski's usage. I have not considered all the other possible usages because we only have the claim of a rigorous method from Dembski's EF.
[ 18. February 2004, 01:27: Message edited by: Rex Kerr ]
IP: Logged
|
|
Erik
Member
Member # 160
|
posted 20. February 2004 17:21
quote:
John Bracht: As I said in my post, I don't think about probabilities when I make decisions; I consider various rationales, motivations, past events, desires, needs, etc, into my decisions. As such, to try to describe the process of decision making as a probabilistic outcome seems like a category mistake.
The ball in a roulette wheel also does not think about probabilities. It reacts to the friction from the surface under it, to the air resistance, to the normal forces from the slots, etc. To model the outcome of a roulette spin by saying that the probability of the ball landing in any one of the slots is 1/36 is an extreme simplification compared to performing an exact calculation based on classical mechanics. In fact, we use probabilities precisely because we want a model that does not describe the details of the process. (The only exception being quantum mechanics, some interpretations of which hold that the probabilities calculated from the theory constitute a complete description of all there is to describe.)
Why are probabilities appropriate for roulette, but not for your decisions?
quote:
John Bracht: As I argued before: how do you account for the fact that the founding fathers might have chosen a virtual infinitude of alternatives to writing the declaration of independence--how do you account for the possibility that they just got drunk instead--or comitted suicide? In my case--what is the probability that I would sit down and write a post analyzing these ideas? Well, it has to do with a fairly large number of other circumstances in my life, like having the time, desire to do so, having already made one post, having someone respond to my original post, etc.
The first step in setting up a probabilistic model is to think about which outcomes can occur (the set of possible outcomes is called the sample space). When we study the game of roulette, we will probably find it reasonable to include the ball landing each of the slots of the roulette wheel in the sample space. But we may also want to include the possibility that the ball bounces off the roulette wheel and falls down on the floor. We may also want to include the possibility that the ball stops so that it balances perfectly on one of the walls that separate the slots. The less we know about the underlying physics and the initial conditions, the more difficult it will be to non-arbitrarily choose a sample space.
In order to "account for" the different possible decisions of the founding fathers we should first decide on exactly which question is to be studied. Which kinds of details do we care about and which details can be lumped together? (E.g. in the case of roulette we usually don't care about how the ball has rotated relative to its original orientation. There is also an infinite number of ways in which the ball might hop off the wheel and roll over the floor, but it could be reasonable to lump all of these together and represent them by the single outcome "the ball hopped off the wheel".) If we have a poor understanding of the founding fathers' motivations, goals, means, psychological quirks and the constraints imposed on them via interactions with their environment, we will probably have make more or less idealized assumptions. It can be quite difficult to make a non-arbitrary choice of a sample space. But two things should be noted about this difficulty:
(1) The difficulty is not specific to modelling of intelligent agents. The same difficulty occurs in modelling of any complicated process with a large set of conceivable outcomes and biases that are difficult to describe.
(2) You seem to prefer an objective, non-epistemological interpretation of "probability". This means that our difficulties in justifying statements about probabilities in no way implies that there are no true probabilities. Even if we can't account for the possible outcomes the founding fathers' decision-making process, this does not mean that isn't a true choice of sample space.
quote:
John Bracht: So what is probability? I guess I'm thinking here of probability as applying to some stochastic (chance-driven) process, whereupon varying outcomes occur with characteristic frequencies but in undetermined ways (ie, they're not determined by laws). So I guess this is where I think the breakdown comes when applying to intelligent decision making. Decision making is highly rational, intentional, thoughtful (hopefully!) and there really isn't any element of chance or randomness to it. But it's not really deterministic either--my choices aren't fully determined by my circumstances, since I can choose to do the completely illogical thing or the less-rational of many choices.
I fail to see why there is a breakdown for intelligent decision making, but not for balls spinning in roulette wheels. Can you clarify? You have tried (but not succeeded, AFAICT) to explain why probability descriptions of intelligent agents breaks down, but you have not explained how the alleged breakdown is avoided for unintelligent processes like coin flips, roulette wheels, etc.
Erik
IP: Logged
|
|
Stephen Wright
Member
Member # 195
|
posted 24. February 2004 17:55
quote:
Why are probabilities appropriate for roulette, but not for your decisions?
Erik, I agree that probability computations, modeling potential connectivity to specific circumstances, absolutely apply to human decision making. The survey is the measurement tool of choice, documenting behavioral data. However, the nature of the decisions made by individual intelligent agents, is much more complex than purely material transformations. The critical issue becomes defining the specific environmental parameters with potential for physically logical interaction. Earlier in this thread there was opinion about what makes a good working spec for an object, event or process.
Can we examine sources of additional information that come into consideration if we see the roulette ball as an intelligent agent?
1. New source of information that is the smart ball’s detection of its local experience in real time.
2. New source of information that is the smart ball’s evaluation of potential responses realizable as interaction. It is assumed that smart ball has capability of computation and mental analysis.
3. New source of information created when the smart ball’s evaluation is turned into actions. Cybernetic feedback both positive and negative is available for further orientation and timing adjustments.
The sample space now includes a much larger set of vectors. The probability for selection of slots is now dynamically changing with feedback from the environment. It would be assumed that smart ball could make determinations that can be directed back into the environment, which institute willful changes; such as using the air conditioning currents, slight surface imperfections and rotation to effect the rebounding from the slot walls, ever so slightly, so that a pre-determined preference for patterns of red or black can be achieved. Also, as an intelligent agent, smart ball could communicate or mutually encode these patterns with a partner. Coordinate these pre-arranged sequences with betting and a new functional purpose is created.
Everything would remain under the guidance of physical laws, except for our smart ball’s imperceptible ability to nudge its inertial values. (this is a thought experiment and almost all intelligent agents have some ability for causal output)
The key moment in this flow of physical and informational events is the instant when an intentional decision to nudge is enforced and smart ball realizes an event that conforms to the specified pattern for red and black, utilizing the fixed complexity of apparent parameters to mask its purposeful actions. While all other events can be mapped to deterministic factors, the enforced decision binds the needs of the intelligent agent’s local reality with information from its circumstances.
Maybe this behavior is an achieving of specified complexity (naïve version) due to the fact that there is little probability for the pattern to emerge without directed purposeful actions. There is guidance from situational analysis, which can be aimed at personal reward on physical, emotional or informational levels. It requires recognition of a self-referential benefit that is satisfied by targeted behavior. There is no analogue for this in strictly physical processes, as no sense of self-identity exists.
I agree with John Bracht that there is a difference in category. One could use the worldview of Gerald Schroeder in the 2001 publication titled The Hidden Face of God and say that there is “wisdom” in the heart of all matter, but this is not a pragmatic stance. There is simply - no manifest strategy in strictly physical processes like roulette, only organization that generates patterns. When intelligent agents make decisions, there are multiple strategies available, among them those based on timeframe expressed by short, medium and far-reaching goals. Daniel Goleman author of Emotional Intelligence would point to long-term goal setting, with its associated principle of gratification postponement, as a characteristic of this type of intelligence.
The pathways of independent decision-making are at a different level from the linear and deterministic probability of an unfettered roulette ball. Intelligent agents recognize and copy the performance of a new successful strategy. While we can predict with statistics from surveys and studies, the performance expectations for behavior exhibited by living organisms, any given intelligent choice can set a new standard for sweeping change if it is creative enough and leave the current norm behind.
[ 25. February 2004, 09:25: Message edited by: Stephen Wright ]
IP: Logged
|
|
Erik
Member
Member # 160
|
posted 26. February 2004 17:40
Stephen Wright, I agree that it is more complicated to formulate a statistical model for an intelligent agent than for a ball in a roulette wheel. The reason can be stated simply and plainly: The ball in a roulette wheel typically has a very simple internal state, whereas an intelligent agent has a much more complicated internal state. If the internal state of the intelligent agent is known, then our model can be conditioned on this fact. It is this conditioning that is complicated.
As usual, it should be noted that this extra complication is not unique to intelligent agents -- it arises whenever we are studying a system with a complicated internal state.
Erik
IP: Logged
|
|
Stephen Wright
Member
Member # 195
|
posted 27. February 2004 12:15
quote:
As usual, it should be noted that this extra complication is not unique to intelligent agents -- it arises whenever we are studying a system with a complicated internal state.
Erik, We differ on this view, not in an absolute way but as a matter of category definitions. There are pathways that are unique to intelligent agents including the ones my gambling Smart Ball character could take. The set of potential events deriving from deterministic internal states have specific kinds of events excluded, which are found in the set for intelligent agents. There appears to this writer the fact that all complex (complicated) states without a self-referential computation capability cannot participate in directed causality like that coming from an intelligent agent. The continual evolution of adapted behavior from cybernetic feedback is likewise excluded.
On the other hand - novel pathways are always being generated by intelligent agents, due to the merging of their local information state with the environment’s information state, into new information that is referential to both sources. I think this logical fact lies at the core of conceptions regarding complexity resulting from conscious or unconscious decisions made by living things. When we study systems that are physical only, without the skill-sets of organisms, they will never exhibit without intelligent input from the outside, the self-referential and willfully enforced guidance of cybernetic feedback.
[ 27. February 2004, 13:56: Message edited by: Stephen Wright ]
IP: Logged
|
|
kyle7
Member
Member # 191
|
posted 06. March 2004 22:06
Rex Kerr says:
quote:
Exactly! This is why the filter doesn't filter anything!
Because we have to ask: why separate out intelligently-caused events?
Suppose we apply the filter, but leave out different classes of causes: leave out gravity; leave out Newton's Third Law; leave out quantum chromodynamics; leave out intelligence; leave out proteins; leave out evolutionary algorithms.
The filter will have a tendency to tell you that something is missing in every one of these cases, depending on the examples you pick. My question is: what privileges intelligence to be the output of the filter, and neglected from the possible causes?
Rex, what is the purpose of the filter? Is it not to distinguish intelligent design? You seem to be talking about the "side information". Certainly, if you leave out the relevant side information then your result may be faulty. But the utility of the filter is that it forces us to evaluate if something has CSI by resorting to known scientific laws and theories. Even if someone fails to include relevant "side information" then someone else can always question this and further analysis can take place. The notion of including intelligence as a cause is superfluous. The whole purpose is to distinguish natural causes from intelligent causes using scientific laws and theories. I find it ironic that so many find Dembski's filter threatening. Actually, the filter forces us to be scientific and not accept any theory on speculation and guestamations. Those who claim to uphold the scientific standards are afraid of Dembski's filter because it forces them to put away their biased ideology and remain in the realm of the scientific. How ironic!
IP: Logged
|
|
Rex Kerr
Member
Member # 632
|
posted 07. March 2004 10:06
Kyle, maybe you can explain why "including intelligence as a cause is superfluous"?
Half of my point is that the assumption that intelligence need not be included is wrong, or at the very least makes it impossible to construct any examples where we know the filter is working.
I am less interested in the intended purpose of the filter than in figuring out what it is actually doing.
Edited to add a missing word.
[ 08. March 2004, 01:05: Message edited by: Rex Kerr ]
IP: Logged
|
|
kyle7
Member
Member # 191
|
posted 09. March 2004 17:58
Rex,
Nature is limited in the complexity and specification that it can generate. For example, a mosaic found in the desert could be examined. It could possibly be due to a random configuration of rocks. If the complexity was great, a probabalistic explanation would be ruled out. Next, you could examine known physical laws and see if a reasonable explanation could be found. If none could be found, the only rational conclusion would be that an intelligent agent constructed the mosaic. There is no need to think initially about a cause due to intelligence. It is superfluous. I think this mosaic example would suffice to show that the filter is working.
IP: Logged
|
|
Rex Kerr
Member
Member # 632
|
posted 10. March 2004 06:48
Thanks for explaining. I think I understand your position now. Unfortunately, it's a little too simplistic to advance this discussion.
I'd encourage you to think about questions such as: do we, in fact, know all relevant physical laws? If we do, can we, in practice, predict the probability of any natural event? What are the key differences between a description of natural law and human behavior? How do we know when a human behavior is specified?
After considering such issues, perhaps the concerns I raised in the first post to this thread will make more sense. You may not agree with my conclusions, but at least we'll be talking about the same issues.
[ 10. March 2004, 06:49: Message edited by: Rex Kerr ]
IP: Logged
|
|
kyle7
Member
Member # 191
|
posted 10. March 2004 15:09
Rex,
My response may appear simplistic, but I am trying to understand why you think Dembski's
filter is lacking. Having perused your thread I don't see any profound thoughts concerning the issue.
Here are the issues you raised:
1) Do we know all the relevant physical laws?
Scientifically, we can only use the known physical laws. This is one benefit of Dembski's filter. It forces us to remain scientific. Where we don't know the relevant physical laws, Dembski's filter will force use to find out what we don't know. For example, someone may question that natural causes formed a complex mosaic. That person could then propose new laws of physics to explain the complex mosaic. Experiments could then be done to either prove or disprove the new theory. Usually, the known scientific laws sufice to explain natural phenomena -- at least today with the level of scientific advancement that we have attained.
2) Can we predict the probability of any natural event?
Yes, for example, if we wanted to determine the probability of sand configuration due to wind errosion, we could run fluid models coupled with statistical and probability methods. Any natural event could be examined.
3) What are the key differences between a description of natural law and human behavior?
What does this question have to do with the cost of tea in China?
4) How do we know when a human behavior is specified?
We aren't concerned about human behavior. We are concerned with the artifacts made by humans.
IP: Logged
|
|
|
Printer-friendly view of this topic