What was Dembski thinking? (CSI math)

Starting this as a header for a new topic.

This started with a side discussion between @EricMH and myself, sharing what we think about about Dembski’s CSI and the criticisms of that work. This is a discussion of the meaning of CSI and the criticisms of it, not an attempt to put Eric in the position of defending it. If Eric wants to do that, OK, but let’s try to fully understand first.

Good point. CSI is essentially a log likelihood test.

1 Like

Aside to @EricMH: I have a hard time understanding what Dembski must have been thinking. He must know about LRT’s, and he has previously discussed “rejection regions” as part of interpreting CSI. Surely he knew the obvious critiques that would be made.
I’m not asking you to defend him, just offering my thoughts on CSI.

CSI is a bit more defined than merely an LRT. It makes certain requirements, such as the specification being independent of the event, which result in guaranteeing true positives (at least in a probabilistic sense). Do you know of any mathematical problems with his definition of CSI?

The only thing I’ve been able to find is that the specification can be a harmonic series and diverge, but this is more a technical flaw than a fundamental flaw, and is addressed in a variety of ways, such as through ASC’s use of prefix free Kolmogorov complexity or by normalizing.

I can name several from Dembski’s 2005 paper, which is the one I know best. One or two of those are minor bugs.
Elsberry and Shallit have extensive critiques (2011, 2014, IIRC), and there is another by Devine (2014).

If you don’t mind, I’m not going down that rabbit-hole today. We could have a rousing good time arguing about it, but I’ve got to get some work done. :slight_smile:

PS: I agree that ASC is a better formulation of what Dembski was trying to do. Also, the claim made for ASC are not so strong as for CSI.

I’ve read through Elsberry, Shallit, and Devine. Would you be able to name the non minor bugs you know of? I’m not looking for a debate here, just am interested to know.

I wrote a blog post about it.


  1. he estimates the probability of just one event, but should be estimating the probability of one or more events. That puts him off by a factor of \pi, I think. I worked it out once. This is minor.
  2. Related to the above, he conflates probability and expected value.
  3. He allows probabilities to be greater than 1.0, and even uses this in one of his examples that shows negative CSI. This means CSI is undefined for anything not already known to have CSI. (What was he thinking?)

My biggest criticism is essentially that he only has half of a likelihood ratio test. He needs a likelihood of design to go along with it.

I would be willing to accept “this is not Dembski’s best work.” It’s an odd paper.

OK, going offline now, for realz!


Then where is the independent specification for biological life, and how did you determine it is independent of what was actually observed in the genome?

Well, unless all of our inventions, like the outboard motor are directly inspired by biological organisms, then describing the flagellum as an outboard motor is an independent specification.

Thank you. I’d be interested in going through this with you in the future, since it appears to make an argument by misattribution.

FI is a lower bound on MI, and MI cannot be produced by natural causes. Therefore, FI is indeed a valid indication of ID.

Sorry but analogies are not specifications. Where is the independent specification for biological life, and how did you determine it is independent of what was actually observed in the genome? If you don’t have an answer just say so, don’t try to tap dance.

Another empty assertion with no relevance whatsoever to biological life.

Does a rain cloud contain FI? Yes or no, and why?

It is a specification. My example is exactly the example Dembski uses in one of his papers.

Regarding rainclouds and FI, I don’t know the biological definition of function, but in general yes the raincloud has FI if we stick with the CSI version.

No, an analogy is not a specification. Dembski’s ID argument failed miserably too in case you haven’t noticed.

So according to you apparently naturally occurring rain clouds must be Intelligently Designed. Please describe this design process for us - how was the manufacturing done, how were the raw materials gathered, who or what did the designing. Be as specific as possible.

This is a pretty interesting claim. Can you elaborate? How exactly does it fail? Don’t worry, I won’t debate you, I am just interested in any specific reasons people can give to disprove ID, as I’m interested in doing so myself.

As for your other questions, I don’t have an answer for you. CSI doesn’t say much about the design process itself, although that is a pretty interesting question.

How many professional scientists has Dembski’s arguments convinced of the Intelligent Design of biological life? How much follow-on research has it spawned? It’s been two decades since Dembski’s popular press book The Design Inference came out and was promptly laughed out of the scientific room, metaphorically speaking. If not for the DI keeping ID on life support it would have faded from existence a decade ago. That’s pretty much the textbook definition of failure

@Timothy_Horton I pulled your comments into this new thread too, but you are sort of racing ahead to the criticism before we have identified the source of the problem. You might slow down for a while until the rest of us catch up.

Specifically, we should be sure we are all on the same page about that Specification means before we argue if it is right or wrong.

I have long wondered about that. His Ph.D. at U.of Chicago was on probability, so he should have known enough to see the problems with his ID arguments.


I would be happy to discuss it. I keep meaning to revise parts of it, but haven’t gotten around to it yet. Feel free to tear me up! :slight_smile:

For a bit of background, it was this fairly minor error that sparked the original “What was he thinking!” thought, and inspired me to dive deeper in to the paper, including reading-up on Kolmogorov Information theory to try to understand what Dembski actually meant.

[Note: I have edited to add some appropriate links to my older comments.]

1 Like