“The intellectual equivalent of spray painting graffiti”
| August 10, 2009 | Posted by William Dembski under Education, Intelligent Design, Science |
The folks at RichardDawkins.net have their panties in a bunch over some of my class assignments (go here):
Thanks to Baron Scarpia for alerting us to this website, which outlines the rigorous academic standards [Quote-miners, please note: this is sarcasm] which William Dembski’s students have to achieve in his courses on Intelligent Design and Christian Apologetics at Southwestern Baptist Theological Seminary:
http://www.designinference.com/teaching/teaching.htm
If you follow the links, you will see that it is full of gems: we won’t spoil them for you by flagging them all up, but – just to whet your appetite – you will notice that, at both undergrad and masters level, there are courses for which 20% of the final marks come from having made 10 posts defending ID on ‘hostile’ websites! This could explain a lot.
You may be less amused at some of the questions in the final exam of the Christian Faith and Science module: http://www.designinference.com/teaching/2008_fall_sci-faith_mdiv/final_exam_10dec08.pdf. In particular, this one:
Trace the connections between Darwinian evolution, eugenics, abortion, infanticide, and euthanasia. Why are materialists so ready to embrace these as a package deal? What view of humanity and reality is required to resist them?
And this one:
You are the Templeton Foundation’s new program director and are charged with overseeing its programs and directing its funds. Sketch out a 20-year plan for defeating scientific materialism and the evolutionary worldview it has fostered if you had $50,000,000 per year in current value to do so. What sorts of programs would you institute? How would you spend the money?
It seems that sending my students to post on “hostile” websites, however, sticks especially in their craw. Slashdot has since picked up on it (go here — the keyword tags are precious).
Want to know how Darwinists really think? Go to the websites listed here and find out. Thus, when I require students to go to these websites and defend ID, it is sound pedagogy. Darwinists reflexively call this trolling (a projection of their own propensity to troll). One individual even emailed me that this is “requiring your students to participate in the intellectual equivalent of spray painting graffiti.” Nonsense. These sites provide a forum and, ostensibly, encourage discussion. My students go to these sites not to pretend to be something they are not but to defend their views — with civility.
In any case, I’ll make you a deal: let Darwinist, atheist, skeptic, freethinking, and infidel websites state prominently on their homepage the following warning — “Intelligent Design Supporters Strictly Prohibited” — and I’ll make sure my students don’t post on your sites.
93 Responses to “The intellectual equivalent of spray painting graffiti”
Leave a Reply
You must be logged in to post a comment.
To jerry,
Its not about getting you own way, its a simple principle of honesty. You cant claim (on an ID blog) that ID is science, then just ban comments you dont agree with. Its the hypocrisy that grates.
Yes Graham,
It’s akin to saying that all that matters in the material evidence, then ignoring to the point of pathological denial.
“Its not about getting you own way, its a simple principle of honesty. You cant claim (on an ID blog) that ID is science, then just ban comments you dont agree with. Its the hypocrisy that grates.”
This is the type of joke comment that appears here all the time from an anti ID person. No where will this person back up this comment. The only hypocrisy is in this type of comment.
“Dembski just proved my point for me.”
No he didn’t. He does not want to waste his time on the nonsense that appears. He answered the inanity that immediately showed up and decided not to answer any more. Who could blame him. It is his thread and he can do whatever he wants to do. You are commenting here and I am sure you could comment as much as you want and no one will shut you down here. So have at it.
So pick a non Dembski thread to discuss your insincere indignation and you and the other trolls can discuss it all you want.
jerry:
It’s certainly true that Dembski is under no obligation to hold a discussion with his challengers. It’s also a fact that he has good reason to avoid such a discussion, namely that his points are false as a simple matter of fact.
It is true that Dawkins’ WEASEL illustrates a very simple Darwinian principle, but Dembski agrees with rather than disputes that principle. What he disputes is a claim that is made by neither Dawkins nor evolutionary theory, namely that evolutionary processes do not create information, i.e. the active info of a search never exceeds the endogenous info of its parent search.
That claim is trivially true if the parent search space is defined in a restricted way, in which case it’s true for both non-intelligent and intelligent processes. Intelligent entities cannot pick a small target out of an informational void any better than other entities. The LCI follows from the trivial mathematical fact:
If P(A) = P(B) then P(A) >= P(B&C)
and that fact is true regardless of whether you’re intelligent or not.
(BTW, if the definition of the parent search space is unrestricted, then the LCI is false for both non-intelligent and intelligent processes. Pick your poison.)
So it is a simple fact that the LCI neither disputes evolutionary theory nor supports ID. This is math we’re talking about here. The ID community would be better off disputing matters of opinion.
R0b,
You have set yourself up as an expert on this so I have a suggestion. Namely, that you take your insight to the journals that have published his articles and publish a letter or article telling the world what you know and that you claim Dembski is lying. Lying would be the appropriate term if “his points are false as a simple matter of fact.”
So step up and back your accusations. Otherwise there is a religious expression that applies.
Hey R0b,
I’ve been away from UD for a little while, only stopping by occasionally to lurk or post something quickly, so I haven’t had the opportunity to see your development on the LCI.
From your last point to jerry, it seems that you now feel that the LCI would apply equally well to humans and other intelligent agents as it would to self-contained, mechanical processes. Is this in fact your new argument?
Atom
jerry:
Other than the WEASEL misinterpretation, I see nothing in his journal-published article that isn’t true. It’s his statements on this blog that are false. And the term “lying” is appropriate only if he knows them to be false, which I am in no position to judge.
Atom, welcome back. I’ve been in and out myself.
If the LCI is defined with the condition that you specified, then yes, I’m arguing that it is true for intelligent agents. It’s a very simple mathematical fact, and I see no evidence that intelligent agents can violate mathematics. I’ve pointed out many times on this blog that intelligent agents cannot find a target in an informational void any better than anything else, so it’s actually not a new argument.
R0b,
Intelligent agents don’t design informational voids for their designing purposes, only evolution has this incontrovertible problem.
R0b,
Sorry if the word “new” came across as disparaging, I meant new to me. I do recall you mentioning concerns about that issue but hadn’t seen you present it as an actual argument until now.
I think you bring up a good point that needs to be addressed (namely, if the LCI holds, which it does, how can intelligent agents possibly circumvent it?) and I’ll do my best to address your question after I give it some more thought. Maybe we’ll come up with some more interesting results.
Thanks for the warm welcome back.
Atom
jerry:
Can you tell me which of my accusations I have not backed up on this blog?
“Can you tell me which of my accusations I have not backed up on this blog?”
Take your arguments against Dembski to the journal in question. They will decide if you can back them up on not. I do not know that much about Dembski’s work to make a comment but I have never found much insight from what you write so I will let the editors at the appropriate journals have at it. You are wrong on your WEASEL comments so what else are you wrong on. You show no understanding of the information content in biology and its relevance so am I to take your other technical comments as accurate.
So if you are right go where you can make an impact instead of sniping here. Go and publish or at least gather a following for your ideas.
R0b,
Thinking over your objection it seems that we run into problems if we imagine that intelligent agents choose to construct a system in order to solve a problem, rather than just solving the problem directly. The meat of the LCI is that it takes more (or the same) amount of functional information to construct a system to solve a problem than it takes to solve the problem directly.
The set up for the LCI takes as a starting point a collection of systems/strategies/probability density distributions/etc which are systems we could potentially use to solve a problem. The functional information we must input in order to narrow that set down from a set with average probability of success equal to blind search to a set that performs at least as good as q is equal to or greater than the amount of functional information we gain by using that strategy/fitness function/algorithm/etc.
So, choosing one system from a group of systems requires functional information, and we get no more functional information out of that strategy than the functional information we input by our choice.
However, humans don’t usually solve problems in this manner; we don’t say “how can I construct a machine to write my term papers for me?” since we know that writing our term papers directly requires less functional information than creating a machine that can then write term papers for us. The original problem is the easier one.
So it seems we need to posit a source of functional information, that can create it directly. From experience, we know that humans regularly generate large amounts of functional information, even if we cannot yet explain how they do so. But the empirical fact remains that functional information is consistently associated with intelligent agency.
So if humans are sources of functional information, rather than mere re-shufflers, this would seem to answer the question in a way that doesn’t require any additional adjustments to ID. ID has always posited that the answer to the information problem is intelligent agency and this would be one more case in which the problem arises and is answered by the same hypothesis. I find it no coincidence that the argument usually returns to the information problem and that intelligent agency is once again the most causally adequate explanation, based on what we know about the “cause and effect structure of the universe,” to borrow a phrase from Stephen Meyer. Intelligence is the only cause we know of capable of generating large amounts of functional information.
How exactly this occurs is an exciting open question at this point and is an example of research questions an ID perspective raises. The answer can’t be: “The choosing of a strategy to solve my original information problem generates functional information” since the LCI shows conclusively that this act does not generate information in and of itself; rather it must be that by generating functional information, I can then choose a strategy (if need be) or solve the problem directly (as is the usual, and easier, case.)
Atom
A couple clarifications on my previous post:
I accidentally wrote that we get no more functional information out of a system than we put in, but this isn’t 100% accurate; to be precise, the active information output is not greater than the functional information input. (The active information itself is what leads to success in the search, and therefore functional information on the lower level search, but the way that I wrote it wasn’t technically precise.)
Sorry for the imprecision of my phrasing, I trust you got the actual intent.
Atom
jerry:
I have already told you that I’m disputing statements that Dembski made on this blog, not in his published article. Why would the journal care about statements that Dembski has made on this blog?
Please quote one of these inaccurate comments.
This conversation started with you telling Anthony09:
So pick a non Dembski thread to discuss your insincere indignation and you and the other trolls can discuss it all you want.
Now you’re saying that I shouldn’t use this blog as a forum to dispute what Dembski has said on this blog. Sorry, but I’m going to continue to do so.
Atom, thanks for your response. You make excellent points that should be discussed, which I can hopefully do later today.
Take your time R0b. I know you’ll give a thoughtful response; hopefully we can clarify where exactly we disagree, even if we can’t agree in the end.
Atom
Atom,
I’m not sure if I’m reading your argument correctly, so I do a little bit of restating to see if I’m understanding you.
The LCI is a mathematically inviolable fact, but it only describes the relationship between a child search and a parent search. Intelligent agents find functional targets directly rather than searching for a search, so the LCI isn’t applicable. Am I on track so far?
But how do agents find functional targets? I submit that under both definitions of “active information” given toward the end of section II in the recently published paper, agents use active information to do so. Under the first definition, we know empirically that agents can’t find targets in an informational void. Under the second definition, an agent’s ability to find a target with higher probability than blind search constitutes active information, by definition.
So intelligent agents, like everything else, use active information to find targets with better-than-blind efficiency. It would seem, then, that Marks and Dembski’s concepts can’t be used to distinguish intelligent causes from unintelligent causes.
You, however, are talking about a certain kind of target, namely functional information. If we note that only intelligent agents are observed to produce functional information, then we can infer by induction that functional information of unknown origin was produced by an intelligent agent. That seems a very different argument than the one made by the EIL, which doesn’t involve or quantify functional information.
Am I way off track here?
Hey R0b,
The first part of your restatement sounds close to correct, though I might ask for the context of “find functional targets”, since I can think of two different cases that may obtain. In the first case, producing English text or functional, complex machinery, humans can seem to find the targets that convey meaning or accomplish their task without much trouble. For example, if I wanted to create a new piece of software to calculate the first million digits of pi, I can do so without trouble, though the number of programs that calculate that result are extremely small compared to space all programs. Somehow I know on how to zoom onto that small island of functionality, and find a small target that accomplishes my purpose. Intentionality, purpose, design. The same is true with finding vastly improbable functional strings of letters that convey the exact meaning I want.
However, in the second case, I can imagine trying to find the combination to a lock that I have no information about. That seems like a case where I wouldn’t do any better than blind search.
However, there is an interesting middle ground as well. Let’s say we’re trying to find a particular word in a book. If we know nothing about the book, I will have to resort to brute search, looking word by word. But, as I search I will notice that the words are ordered in such a way that the book is a list in alphabetical order. Then suddenly I can find the word much quicker, since my intelligence perceives this fact and applies this knowledge to my search.
So I agree that in some cases we do need structural information, even as intelligent agents, to find certain targets. But I also see how we can create functional information when finding the targets directly, as we do in outputting improbable lines of text.
To relate functional information to active information, we see that functional information in our higher level search can result in active information for our lower level search. (The functional information of the set Q, which is the set of “good” functions, depends on how small it is compared to O2, the higher level space.) So to achieve active information associated with an improvement of q, we need a measurable amount of functional information, measured to be at least as much as the active information. It is here that we ask “Where did this higher level functional information come from?” My argument is that it comes from the same place we usually see large amounts of functional information come from: intelligent agents. We are either the source of the functional information, or something else is. Experience tells us that we can produce large amounts of functional information, and we’ve yet to see another cause capable of producing it so readily. So intelligence is currently the most adequate explanation.
The problem I would like to see resolved is to measure exactly how much functional information humans are capable of generating (it seems like a lot…just look how long my post is!) and to investigate what exactly causes us to be capable of doing such things (if we can discover this). I wish I had answers to those questions, but more research needs to be done.
In the meantime, I think it is a good question you raise, even if it isn’t a knock-down argument against the LCI or ID. We find that a search for a search once again returns to the “Information Problem” we see popping up again and again. Where did all the functional information come from? ID says from intelligence, materialists say from the cosmos and the environment. However, since evolutionary searches require fitness functions specific to the forms they produce, and therefore require functional information to explain the narrowing of the fitness function set, they don’t appear to be the ultimate answer. I have a strong sense that the ultimate answer will involve intelligence, since that is the only causally adequate class I can think of.
Atom
PS, to guard myself from nitpickers:
When I said the number of programs that output the first million digits of pi’s decimals is small compared to the space of all programs, I obviously am implying an upper limit on the size of the programs. (In other words, all programs of a fixed size that output pi.) The space of all programs is infinite, as is the number of programs that output pi, since I can always add useless operations to any program.
I know you got my meaning, but I can imagine someone in webspace saying “what an ID-iot…the space of all programs infinite!”
Yes, yes it is.
Atom
Atom, thanks for your response.
It’s true that, given an alphabetically ordered book, an intelligent human is likely to notice the order and find the target much faster than blind search. Let’s suppose that the probability of the human succeeding within a certain number of queries is 90%, while the probability of a blind process succeeding is .000001%. This gives a ratio of 90 million, or about 26 bits of active info.
Note that 26 bits of active info is both necessary and sufficient in order for a non-blind search to have a 90% chance of success. So the success, and therefore the resulting functional information, is wholly explained by the active information. All that remains is to explain the active information.
In order show that at least some of the active information was actually created by the human, we would at least have to show that the endogenous info of the higher-level search is less than 26 bits. How many configurations in the higher-level space consist of an alphabetically ordered book combined with an agent that can detect the order? (Or some other configuration that yields at least 90% chance of success.) If it’s less than one 90 millionth, then Marks and Dembski’s accounting method says that no information has been created.
And given that the LCI is a mathematical fact, the endogenous info of the higher level search is guaranteed to be at least 26 bits. So no matter what, even if an intelligent agent is involved, we’re stuck with the mathematical fact that no information is created.
Sorry for the long-winded reply. The upshot is that I see no way for intelligence to get around the math.
R0b,
I think I see where we’re missing each other. You’re using functional information only to refer to the results of success on the lower level search, as in the Active Information creates functional information on the lower level. This is true in a sense.
But what you’re not taking into account is that functional information occurs on the higher level search as well, explaining the active information.
Namely, in the LCI proof we end up with as the final line:
I+ <= -log(|Q|/|O2|)
What I noticed is that this relation defines the relationship between functional information in the higher level (O2) search space and active information on the lower level search. What is on the right hand side is functional information as defined by Hazen et. al. in their PNAS paper "Functional information and the emergence of biocomplexity“, where q is the threshold defining acceptable function, |Q| is the number of states achieving that threshold, and |O2| is the total number of possible states. So the LCI says that for any amount of active information, it takes at least as much functional information (built into the system) to achieve it.
The question then becomes “Where does this functional information come from?” This is the same as the question “Where does functional information, in general, come from?” From experience, we know the most likely answer: intelligent agents. You know what it is about yourself (namely your intelligence) that allows you to output functional information in the form of mathematical proofs and UD posts.
Now does the LCI hold in reverse, so that every amount of functional information requires a given amount of active information (leading to a circle)? I don’t think it does, but it could be that I’m not thinking about it hard enough. If you can show that it does, please let me know.
If not, then there is no problem positing a functional information generator…especially since a) we need one to account for the functional information we see and b) we have one very good candidate we see generating it all the time: intelligence.
Atom
Addendum:
To use your example from the last post, the human began with no information about the search space in question (namely, the book). Therefore, using one definition of active information (knowledge of a search space used to achieve results better than blind search) the agent began with none.
At t0, active info = 0 bits.
This is further confirmed, since the agent is initially using blind, brute search.
At t_f, a time period later, the agent is now using a binary search, and thus can find the word. This search method matches the search space well, given the ordering, and thus active information is now present in both senses of the word: 1) the method used reflects and incorporates problem specific information, and 2) the performance dramatically improves.
Thus, at t_f, active information increases to let’s say 26 bits.
Where did the active information come from between t0 and t_f?
Well, the agent used her intelligence to notice, perceive, understand and learn something about the search space. Furthermore, she used that information to choose a good search method based on the search space itself.
Functional information generated by the agent, based on intelligent activity, was used to design and implement a good search algorithm.
Active Information requires functional information at the O2 level, and functional information at the O2 level was generated by intelligence. This thought experiment illustrates my point well.
Atom
Atom, the connection you draw between -log(|Q|/|O2|) and functional information is insightful, and it ties Marks and Dembski’s concepts to Hazen, Durston, etc.
As for the functional info / active info loop, it seems to me that you’re equating functional info with higher-level endogenous info (-log(|Q|/|O2|)). Since active info, by definition, is always required in order to feasibly produce a sizable amount of endogenous info, I would say that there is such a loop.
But again, it seems that the argument for intelligence that you’re making is not the one that Marks and Dembski are making. You’re answering the question “Where does functional info come from?” by induction based on empirical observations. Marks and Dembski are answering the question “Where does active info come from?” via a purely mathematical argument. Your argument seems independent of the active info concepts.
WRT your addendum, I would argue, that the active info at t0 is not 0 bits. I have two different arguments, which I’ll spread over two comments.
First, the only quantitative definition that I know of for active info is -log(p/q), where p is the probability of success for blind search, and q is the probability of success for the non-blind search. So it seems that if the human is more likely to succeed than blind search, the amount of active information must be non-zero. If a different definition of active info yields a different number, then it seems that the framework is inconsistent, and Marks and Dembski have no coherent foundation on which to base their arguments.
You might argue that the active info exists at some point in time but not necessarily at t0. That would raise the bigger issue of what it means for active info to exist. Active info is a property of a search, but what is a search? In their latest two papers, Dembski and Marks characterize a search simply as a pair consisting of a probability distribution (the definition of which includes the definition of the sample space) and a target. As long as the probability distribution and the target exist, active info exists as a property of the pair.
In the case of the human and the book, the probability distribution is defined by that human’s abilities and the content of the book, both of which pre-exist the execution of the search. The target also presumably pre-exists the execution of the search, or it wouldn’t make much sense to call it a target. Since the active info is a property of the human’s abilities, the content of the book, and the target, I submit that the active info pre-exists the execution of the search.
Second, I submit that if we use the definition of active info that you stated (knowledge of a search space used to achieve results better than blind search), we still have non-zero active information at t0.
If the human truly had no idea what order the words are in the book, then he would have no reason to infer that the book is alphabetical unless he queried every word. No matter how many queries he made, all of them turning out to be consistent with a hypothesis of alphabetical order, there would always be more non-alphabetical configurations for the remaining words than alphabetical configurations. (Until he got down to the last word.)
But we humans are pattern-finding creatures. When we observe part of a space and notice a pattern, we assume that it continues throughout, unless we have a reason to believe otherwise. That is, we do not follow the philosophy on which Marks and Dembski’s work hinges, namely the Principle of Indifference. (Marks and Dembski don’t follow it consistently either, nor do they justify their choice of when to follow it and when not to. And yet their argument depends on this choice.)
Humans do not start with a uniform prior distribution. If we did, we could never draw inferences or inductions. Our typical prior seems more in line with what Kirchherr, Li, and Vitanyi call the universal distribution (see http://homepages.cwi.nl/~paulv/papers/mathint97.ps). We assign disproportionately high probabilities to configurations that preserve an observed pattern.
This stems from our knowledge that the world around us is filled with dependencies resulting in order, especially when it comes to human artifacts like books. Certain patterns, like alphabetical ordering, are particularly familiar to us.
So, following Marks and Dembski’s characterization of a search as a distribution/target pair, our distribution starts off very skewed, based on what we know about the world around us. The knowledge that we live in a universe that isn’t pure random noise, where extrapolations of observed patterns are often correct, constitutes problem-specific knowledge.
R0b,
Sorry I haven’t had a chance to respond yet.
For my addendum, you are correct. The agent could not be sure that the text was really in dictionary order, even if it appeared to be that way. It could be the case that the pages they looked at were ordered, but there were other pages that were not. This is a defeater of my example, so I’ll concede it isn’t a very strong one. (It does raise an additional question, however, about what the likelihood of having a book that isn’t dictionary ordered but that the random pages you looked at were all in dictionary order…I’m sure there is some interesting question dealing with subset and functional matching, but I don’t know if it would do anything to save my example.)
However, you did make one statement in your refutation that I wish to address. You wrote:
This is true, but you are assuming (without warrant) that the state of the human with regards to its true knowledge and ability to surpass blind search at t0 cannot change after t0. It is like having a computer system. You would argue that the computer is loaded with the ability (hardware + software) to execute a certain search at t0. This is only true if the computer has the correct software. What I am arguing, in a sense, is akin to a computer that can create its own software on the fly when it desires, after t0. If it does not desire to do so, the abilities remain limited, which would mean the active information at t0 was indeed 0 bits. Before it has decided to generate functional information or not, it does not have that functional information, so it cannot be said to have the software/hardware configuration it needs to solve the problem with greater than p probability of success. It can only be said to have the potential to gain that ability, which is a different thing. It may be the case that it never decides to do so, in which case the active info at t0 would be zero for that search.
Active Information changes in response (or rather as a consequence of) functional information. So if we have a unit generating functional information, this means our active information will change over time. You can plot this as average queries remaining until success as the search progresses; the addition of functional information during your search will alter this number from the expected value, leading your search to terminate faster than expected from any given point along your search.
It is a tricky concept, because now we’re getting into intentionality and intelligence, much less well defined areas than mere search algorithms. However, a resolution to the Information Problem will probably push us there in the future.
The D&M paper doesn’t address the ultimate resolution to the information problem, but it is implicit that intelligent agency is one solution to the problem, since the active information in the examples given traces back to functional information encoded by intelligent agents. Perhaps with more time we can flesh out a fuller, more mathematically formal model of this process, but I think this is still a few years off. For now, D&M have had an uphill battle getting others to even acknowledge that there is a problem that needs addressing, since many still feel that natural selection can somehow generate information surpassing the encoded (and quantifiable) information embedded in the fitness function. It cannot. The LCI prohibits this on strict grounds that unless the fitness function itself is logically necessary, then we need information to constrain it to a proper subset.
Again, humans and other intelligences usually do not generate searches for a search to solve problems. We usually just generate what functional information we need to solve a problem. (I don’t build machines to construct sentences for me, I usually just construct my own sentences.) How we do so is something of a mystery at this point, but I’m open to seeing a proof that there can be no information generating devices in history. If you feel the LCI proves that no device in the history of the cosmos (including intelligent agents) could have functioned as a generator of functional information, it would be an interesting result indeed.
Atom
PS I don’t know why my posts to you are so long! Forgive me, I’ll limit the subjects I discuss in my next posts.