(Good luck with finals, BTW)

If we look at your coin toss example, both h1 and h2 are predictive. But suppose you have h3 = "the next coin flip will be heads or tails with unknown probability". h3 is consistent, but not in any way predictive. In fact, it's guaranteed. It also has the property of looking like the original question, i.e., we ask "is the next flip more likely to be heads or tails?" and h3 comes back with "the next flip will be either heads or tails, but I can't say with what probability it will be heads."

There will be some theories that are not tested by certain experiments because those theories are consistent with any outcome in those particular experiments. However, we should demand that those theories are inconsistent with some possible experimental outcomes somewhere.

I don't know what "generic ID" means. What I did was show one , out of many, valid and successful prediction of an ID hypothesis.

By generic ID, I mean an ID theory that says nothing about the abilities or motivations of the designer, nor the utility of artifacts to the designer.

I think that it's selective value needs to be taken into account (i.e., why would a designer incorporate a design that would be weeded out?), but when selective value was the only thing that was taken into account, or rather when the specificity/apparent want to be as accurate as possible and employing checkpoints was not taken into account, (which natural processes do not really care about, but engineers do) then it lead to a false prediction from Darwinian theory.

Can you elaborate on this further?

If there is a selective advantage to proofreading, then non-directed evolution "cares" about it. If an engineer cares about something that natural processes do not, then I would expect the feature to have a purpose beyond mere survival. Is that what you are suggesting here?

I don't mean to slight "postdiction" at all. I think postdiction is a very valid way to check an inference under certain conditions. In particular, I think you have to be able to subdivide your existing data, fit your theory to that subset, then predict (or postdict) the rest of the set. This process is like simulating initial observations and confirmatory experiments using subsets of known observations. I think this may be related to the time independent nature of Bayes' theorem you alluded to.

Sorry for the delay , I'm cramming for finals. I am sympathetic to the importance of distinguishing between prediction and accommodation. I think we agree on that. The main issue is whether confirmation is understood entirely as a logical (or quasi-logical) relation among propositions or whether it is understood more pragmatically. Those who favor a logical relation typically argue for the equivalence of prediction and accommodation. Ironically, your reply to me about Einstein and Mercury seems to make out a logical distinction. I am more pragmatic about confirmation, so I don't accept the equilvalence.

Because a Bayesian tends to make confirmation relations logical ones, Bayesians generally advocate the equivalence. There is a huge literature on this topic. For a good overview of the issues, see: Prediction versus Accommodation and the Risk of Overfitting. Hitchcock, Christopher; Sober, Elliott. British Journal for the Philosophy of Science, vol. 55, no. 1, pp. 1-34, March 2004

Drlogic wrote:

I think it is completely consistent with it, but irrelevant to it.

No, irrelevant theories are not potential explanations of the data, nor are they favored because they were compatible with the evidence. Maybe this example would illustrate what I mean, suppose you show me what looks like a fair coin and begin tossing it. The first three tosses are heads. One potential explanation is that you substituted a two-headed coin without my being aware of it. Call that hypothesis h1. As I interpret it, h1 implies heads on every toss. I think that h1 deserves some small, non-zero probability. Suppose the fourth toss is tails. Then the probability of h1 goes to zero, but I would expect the probability of the following hypothesis to increase ever so slightly: h2 = you substituted a two-headed coin for the first three tosses and a two-tailed coin for the next three.

It seems to me that, no matter what the actual sequence of tosses might turn out to be, there will some hypotheses of these kinds that earn some slight increase in probability on the basis of the data. It was a mistake for me to suggest that all hypotheses compatible with the data have their probability increased. I think it would be a mistake even to claim that all theories that would explain the data have their probability increased by it. But my example is meant to make it plausible that some potential explanations of the data do have their probability increased by the data, largely (if not entirely) because they are compatible with the data.

Drlogic wrote:

Consider your genetic microarray example. If each unique genome had a radically different response to the drug, then there would be no point in doing the exercise (everyone's reaction would be effectively unique).

However, nature was not constructed like this. A finite number of genes control such important functions as apoptosis. Even in such a complicated and heterogeneous disease as cancer, there are commonalities within subsets of patients. A person's cancer may be unique, but it has similarities with some other cancers.

Drlogic:

The theory works because it implicitly says that there are far fewer important factors in drug response than there are possible genomes (or dots on the microarray). Without this assumption of the theory, it would fail to be predictive.

Yes, but without order in nature there could be no science at all.

Drlogic:

I think we basically agree on the requirements of a good theory. What I am trying to do is come to a more formal criteria for deciding when a theory is either irrelevant or a tautological restatement of the data. I think that criterion is the failure of the theory to make any predictions. If a theory does make predictions, then either it has the same or fewer number of parameters as the data set, or else the theory also states that the effects of the unfixed parameters are small.

Lets call your criterion the "parameter counting criterion". I don't believe this is a very good criterion for distinguishing theories from non-theories. For one thing, it does not allow for probabilistic theories (you can never derive data points from probabilistic theories, but only probabilities of data points). Second, your parameter counting criterion is one way that philosophers have tried to work out an account of theoretical simplicity. But there is a lot of disagreement about how to understand theoretical simplicity.

I think there is something right about what you are saying, which is that, other things being equal, Bayes' Theorem should give more support to simpler theories than to more complex theories. Unfortunately, Bayes' Theorem has no implications one way or the other. It is the prior probabilities to which the theorem is applied that determine whether simpler theories are selectively confirmed over more complex theories or whether more complex theories are selectively favored over simpler ones. So your constraint is a constraint on prior probabilities, not a consequence of Bayes' Theorem.

Drlogic:

The question is, what are the predictions of generic ID, if any?

I don't know what "generic ID" means. What I did was show one , out of many, valid and successful prediction of an ID hypothesis.

Drlogic:

Do you think that generic ID is predicting that there will be proofreading in RNA transcription even if it isn't advantageous (i.e., that the designer overdesigns)? Or is the additional proofreading not a prediction of ID?

I think that it's selective value needs to be taken into account (i.e., why would a designer incorporate a design that would be weeded out?), but when selective value was the only thing that was taken into account, or rather when the specificity/apparent want to be as accurate as possible and employing checkpoints was not taken into account, (which natural processes do not really care about, but engineers do) then it lead to a false prediction from Darwinian theory.

1) High information content (or specified complexity) and irreducible complexity constitute strong indicators or hallmarks of (past) intelligent design.
2) Biological systems have a high information content (or specified complexity) and utilize subsystems that manifest irreducible complexity.
3) Naturalistic mechanisms or undirected causes do not suffice to explain the origin of information (specified complexity) or irreducible complexity.
4) Therefore, intelligent design constitutes the best explanations for the origin of information and irreducible complexity in biological systems.

As I said if you had read ID literature (the type written by IDists) your questions would have been answered. Then when you say the following:

"Go ahead and explain to me (in a short paragraph) how it's done without using a process of elimination of mechanistic theories."

It truly exposes your lack of understanding on things are done. Using a process of elimination is only way to come to a design inference without being biased towards that end. That is in the absence of direct observation or designer input.

Wm. Dembski pg 36 of The Design Inference:
"The principal advantage of characterizing design as a complement of regularity and chance is that it avoids committing itself to a doctrine of intelligent agency.
Defining design as the negation of regularity and chance avoids prejudicing the causal stories we associate with the design inference."

Why we call it the design inference:

pg. 91 of The Design Revolution:
"The prospect that further knowledge will upset a design inference poses a risk for the Explanatory Filter. But it is a risk endemic to all of scientific inquiry. Indeed, it merely restates the problem of induction, namely, that we may be wrong about the regularities (be they probabilistic or necessitarian) which operated in the past and apply in the present."

As for "bad form"- bad form is discussing a topic which you have no knowledge of or false knowledge of.

CSI and IC have been well defined. Not knowing about them is evidence you don't know ID.

Again your ignorance is not a refutation…

Look, I'm a reasonable guy. Just make a compelling case for your side. That's all I'm asking. If you have a point, I'll concede it. Calling me ignorant is just bad form.

Ya see we have strict criteria that has to be met BEFORE coming to a design inference.

If I observe a structure, what tests do I do to determine that it is IC or CSI? Go ahead and explain to me (in a short paragraph) how it's done without using a process of elimination of mechanistic theories.

I am sure you do. However evolution isn't being debated. This is all about the mechanism and TO's "predictions" could also fit in with any number of evolutionary models.

doctor:
You claim ID predicts IC and CSI, but it doesn't.

ID absolutely predicts IC and CSI. The only way to deny that fact is via ID ignorance.

doctor:
How is designed complexity researched? Someone points to something they don't understand and say "Hey! That looks too complex to be evolved!" Then, actual scientists figure out an explanation of how it works and how it could have evolved, and then it's not complex anymore. There's no formula for designed complexity that you can compute based only on observations. Designed complexity is determined by the absence (i.e., gaps in)of compelling NDE theories.

Yeah right. If we listen to you then there is nothing that can be said to be the product of intentional design unless we directly observed it. It is a good thing reality demonstrates otherwise. Ya see we have strict criteria that has to be met BEFORE coming to a design inference.

doctor:
The problem ID has is that NDE also predicts "designed" structures, but those structures have only one purpose: survival. And guess what? For almost all of the last 3.8 billion years, the only apparent purpose to life has been survival.

The NDE "predicts" any and everything. And how did/ do organisms know to survive?

doctor:
Maybe there is some designer out there who had a hand in creating life on Earth, but it's hard to make a compelling case for this when there's no apparent purpose to life here.

Again your ignorance is not a refutation. Science has given us an apparent purpose- scientific discovery. As in "The same narrow circumstances that allow for our existence also offer the best over all conditions for making scientific discoveries."

It would serve you better if you actually read some ID literature written by IDists. Arguing from ignorance is not a very good position.

I think the talkorigins page did a perfectly good job of explaining the predictions of evolution. What kinds of predictions were you hoping for?

You claim ID predicts IC and CSI, but it doesn't. How is designed complexity researched? Someone points to something they don't understand and say "Hey! That looks too complex to be evolved!" Then, actual scientists figure out an explanation of how it works and how it could have evolved, and then it's not complex anymore. There's no formula for designed complexity that you can compute based only on observations. Designed complexity is determined by the absence (i.e., gaps in)of compelling NDE theories.

However, this is not to say that a specific ID theory can't make predictions. ID theories predict designed utility, not designed complexity. In SETI, forensics, and archaeology, there's an assumed utility to the pattern of intelligent design, e.g., communication, weapons, tools, etc. All you have to do to show utility is to be specific about why the design is useful to the designer. It's not enough to show that it's useful to the survival of the device under study.

The problem ID has is that NDE also predicts "designed" structures, but those structures have only one purpose: survival. And guess what? For almost all of the last 3.8 billion years, the only apparent purpose to life has been survival.

Maybe there is some designer out there who had a hand in creating life on Earth, but it's hard to make a compelling case for this when there's no apparent purpose to life here. There's no obvious utility in life on Earth to any designer capable of designing it. Who plans ahead 4 billion years?

What predictions have NDE given us? Either you or someone else linked to a talk origins article (once) but not one "prediction" on that page was due to random variations culled by NS or any other NDE mechanism.

Dennett (and the PBS series Evolution) has already told us that there is no way to predict what will be selected for at any point in time.

With "evolution" we can "predict" minor variations, stasis, and great transformations. IOW anything can be accomodated by "evolution".

BTW ID predicts IC and CSI…

I don't mean to slight "postdiction" at all. I think postdiction is a very valid way to check an inference under certain conditions. In particular, I think you have to be able to subdivide your existing data, fit your theory to that subset, then predict (or postdict) the rest of the set. This process is like simulating initial observations and confirmatory experiments using subsets of known observations. I think this may be related to the time independent nature of Bayes' theorem you alluded to.

If P(B|A) = P(B) then this is still a perfectly good use of Bayes theorem, A could still be informative in an overall sense because the probability of some other hypothesis"“such as C"“could change. For example, P(B|A) = P(B) but P(C|A) <> P(C).

I agree with your statement, but you're only saying that experiments/experiences are useful in constraining theories.

My claim is that hypothesis B is never informative/useful until P(A|B) <> P(A), i.e., until B predicts something about observation. Otherwise, P(B|A) is never updated, and believing B never helps you predict the next data point. B is not an explanation of A unless P(A|B) <> P(A).

There is a more formal way to say this but I think it is simply a matter of the data itself constraining the theories. Greenday releasing a single tomorrow is not inconsistent or consistent with shark feeding habits, it is simply irrelevant to it.

I think it is completely consistent with it, but irrelevant to it.

Let's look again at what it means for the data to constrain a theory. A theory has some mathematical structure which relates possible observations. That structure may have a large number of parameters in it that can be fixed by observations.

But what can we say about a theory that makes no predictions, and can therefore accomodate any observation? I think it says that the theory has an infinite number of degrees of freedom, so that any observation can be accomodated by fixing a new parameter. In essence, the parameters of the theory become the observations themselves. There are extremes of prediction and accomodation, and this is one of them. So my claim is that a theory that has no predictions and infinite accomodation isn't a theory at all. It is just a tautological restatement of the data.

If a theory does make predictions, then either it has the same or fewer number of parameters as the data set, or else the theory also states that the effects of the unfixed parameters are small. (My previous statement about parameter counting was too strong.)

Consider your genetic microarray example. If each unique genome had a radically different response to the drug, then there would be no point in doing the exercise (everyone's reaction would be effectively unique). The theory works because it implicitly says that there are far fewer important factors in drug response than there are possible genomes (or dots on the microarray). Without this assumption of the theory, it would fail to be predictive.

I think we basically agree on the requirements of a good theory. What I am trying to do is come to a more formal criteria for deciding when a theory is either irrelevant or a tautological restatement of the data. I think that criterion is the failure of the theory to make any predictions.

The question is, what are the predictions of generic ID, if any?

Do you think that generic ID is predicting that there will be proofreading in RNA transcription even if it isn't advantageous (i.e., that the designer overdesigns)? Or is the additional proofreading not a prediction of ID?

"We can't remotely put ID on the same footing as GR."

Incidentally, I doubt we can put any proposition in historical biology on the same footing as general relativity.