Eric Holloway: Honest Questions on Information Theory

Could you explain what is non-scientific about these arguments? @swamidass argues ASC cannot be used empirically, claiming any estimation will always overestimate the true ASC. I show this is false, so there is no fundamental barrier to use ASC empirically.

Nice try @EricMH. You show no such thing.

To recap, we agree a valid P_{est.} overestimates P_{true}. You claim this is only possible if P_{est.}=P_{true}, since P_{est.} must be normalized across all Xs. I point out if we only look at a portion of Xs, then we don’t need P_{est.}=P_{true} to meet the validity requirement. This shows your refutation is incorrect. You never respond to this point.

Perhaps we reopen that thread to clear it up?

Sure. I’m very interested if you’ve managed to show ASC is inherently useless. You cannot show it is in error, since the primary result is mathematically proven.

Okay when I have some time to engage I’ll open it again. To be clear, ASC is not inherently worthless. You are misapplying it because of equivocations in your proof.

Once again, do you have any experience applying IT in real world problems?

@EricMH Let me be clear, your work on ASC isn’t worthless. But you work in applying ASC to ID is and there are much more important problems to solve that can better mankind. Use your skills and knowledge that can have usefulness to people. I mentioned the DNA compression problem. There are many other problems that can use your insights and expertise.


That is kind of you @Patrick to rightly acknowledge @EricMH has insights and expertise. Thank you.


The subject matter that @EricMH has mastered is hard but very important and useful. But the PhD is just the beginning. You have to apply your skills to accomplish something new and novel. I feel that trying to apply these skills to psuedoscience ID is not going to add value to anyone’s lifes. But to do something that does improves quality of life is truly a useful purpose and meaningful way to use one’s PHd in Information Theory.


If there is no barrier, then why don’t you actually use it for a real life system? I am not trained in information theory, so I lack a nuanced understanding of your mathematical proofs. I have not spent the time to scrutinize them with my rusty math skills. But I am an experimental scientist. And in our field, even if you can calculate a prediction for say the expected decay rate for certain particles, these don’t mean much until someone actually carries out the experiment. In this case, you have not even calculated a prediction for an experiment - you are merely claiming that in principle, one could do so. Maybe that’s enough for some mathematicians, but ID is supposed to be a scientific, not mathematical framework. At this point, your version of ID is still not science - it doesn’t predict anything in a clearly testable manner. The danger is that these difficulties that you simply subsume under the vague umbrella term “practical implementation issues” - turn out to be fundamentally intractable due to some deeper truth about nature (not math) that we haven’t discovered yet.


Yes, I’m all about this. Most of my free time I work on this kind of thing. I come here because @swamidass claims what I’m trying to do is impossible. If he can demonstrate this, that would save me a lot of effort.

As a side note, I’d also argue all the great engineering computational biologists already do applied ID, just not recognized as such. It all deals with deriving the incredible structure in DNA, and structure is a form of algorithmic mutual information. In which case, ID is already well established as a scientific discipline, but credit has been unjustly taken by naturalism.



How can you control variables if one of the variables is miraculous activity?

How does one PROVE that to your satisfaction?

@EricMH you seem to be making an honest effort to bridge the gap of understanding between us. I’ve been engaging with you for a while now. I’m going to be thinking carefully about how to most efficiently and clearly make the point to you, given how you specifically are thinking about this.

I’m not sure when I’ll get time and inspiration. I don’t want you to think I’m blowing you off either. I think its possible we could turn the corner together on this.


3 posts were split to a new topic: The Great Computational Biologists Are All ID?

Great. I’ll continue working on the experiments you suggested. But, I think the main takeaway from this exchange is there is empirical work that can be done on this question, which is the minimal ID position. At the very least, ID can be empirically shown to be wrong and falsified, which would be great. And that makes ID a scientific hypothesis, though a potentially wrong headed hypothesis.

1 Like

The only way I can agree with this is if we are clear that ID, in this case, cannot refer to divine intelligence. With that restriction, it is a valid and worthwhile effort. Modifying it thusly, I could also agree:

At the very least, ID can be empirically shown to be wrong and falsified, which would be great. And that makes ID (of divine intelligence) a non-scientific but important hypothesis, though a potentially wrong headed hypothesis.

Detecting divine-design I think is a valid effort that could be valid, evidential, and science-engaged. I’d encourage such an effort, but only think it could work with sustained and explicit engagement with theologians and philosophers. They have to be part of the conversation too.

A simple proof would be the difference between the probability distribution generated by the system dynamics and a significant enough observed divergence from this distribution. Essentially the same thing as CSI.

Why is this restriction necessary to make it science? If halting oracles are necessary to explain what we see, and physical processes cannot be halting oracles, then the necessary explanation for our world is a divine intelligence. That is just deductive logic.

1 Like

You don’t have to agree. I holds this view of theological, philosophical, and scientific reasons. I’m just telling you what I can agree with, and I already have explained why on the forums several times.

Yeah, I know. Except we’ve shown that it is, in practice, not possible to compute this probability distribution (Predictability Problems in Physics). Admittedly this is for physics rather than biology, but biology is more complex. Eventually, I’ll get around to showing you what ASC is actually telling us, and how we are applying things like it to understand biology. When I have the time, it will clarify how it doing something very different than “design detection”.

Moreover, even when you are computing FI, you are not computing it correctly (Swamidass: Computing the Functional Information in Cancer). This thread on cancer was just for you @EricMH. I make immediate sense to the physicists, and they can follow my math. It also clarifies the error being made in Durston’s work, and the rest of the ID literature with an example. I hope you can look at that closely and understand it.