Response to vjtorley’s post: Why there’s no such thing as a CSI Scanner, or: Reasonable and Unreasonable Demands Relating to Complex Specified Information

This would have been a comment in response to this post – but it got far too long. The original post is here. 

vj

I agree that  (i) to (iii) are reasonable and (iv) to (vii) not required.  I suspect that Mathgrrl would also agree on consideration. However, I disagree that (i) has been done satisfactorily for real-life systems.

To clarify – it is always possible to dream up a calculation and come up with a number.  The question is does it mean anything? And most importantly can it be used as evidence for design without knowing anything about how the object arose.

For me the main beneift of Mathgrrl’s challenge  was to clarify what is meant by CSI.  I don’t think any reasonable person can dispute there is some confusion, even amongst the ID community, about what it means.  Just look at the number of conflicting comments following her challenge.  And indeed your attempts to estimate the CSI for real situations have shown up a number of ways where the concept is unclear.

When you did the calculation for bacterial flagellum on the previous thread I made five objections which you kindly recognised as substantial.  Some of have become immaterial, others you have addressed (I believe wrongly) and others you have not addressed.  I also have some others to add. So I going to present a revised list:

1) The formula Chi=-log2[10^120.Phi_s(T).P(T|H)] contains a rather basic error.  10^120.Phi_s(T).P(T|H) is clearly meant to represent the probability of something meeting the specification at least once.  He does this by multiplying the probability of the observed pattern P(T|H) by the potential opportunities for it or a simpler pattern occurring:  10^120.Phi_s(T). This is wrong. If you have n independent events and the probability of single event having outcome x is p, then the probability of at least one event having outcome x is not np . It is (1 – (1-p)^n). So the calculation 10^120.Phi_s(T).P(T|H) is wrong – even if you could make sense of Phi_s(T ) and P(T|H).  I don’t think this a show-stopper.  The answer is still very small if p is small relative to n.  But it does illustrate the lack of attention to detail and general sloppiness in some of the work on CSI.

 

2) There is some confusion as to whether Phi_s(T) includes all patterns that are “at least as simple as the observed pattern” or whether it is only those patterns that are “at least as simple as the observed pattern AND are at least as improbable”.  As you say, on page 17  of Dembski’s paper Phi_s(T )is defined as the number of patterns that are “at least as simple”. On page 18 he introduces the additional requirement that the patterns should be at least as improbable, but then when we get to real examples such as the bacterial flagellum this additional requirement does not feature in the calculation.  He simply estimates the number of patterns that are as least as simple – and this is the estimate you use.  This is significant.  If we use the “at least as simple” criterion then some of the other patterns may be vastly more probable than the observed pattern.  So we really have to use the “at least as simple  AND are at least as improbable” criterion.  Dembski does not give any way to estimate the number of these patterns, but we do know that it is less than the number of patterns that are “at least as simple”.  So you could argue that he is estimating the minimum CSI. However, there is no adequate justification for only using the patterns that are less probable.  He simply says that this allows us to rule out “large targets”.  But this is simply to repeat the criterion – not to justify it.

 

3) When Dembski (and you) estimate Phi_s(T) you use a conceptual definition of T:  “bidirectional rotary motor-driven propeller”.  This is not necessarily (in fact is  almost certainly not) the same as the exact configuration of proteins.  We have no idea what other configurations of proteins might achieve this effect.  So when you insert P(T|H) into the formula it is a different T!  You do attempt to address this for the ATP case with a note in brackets. You say that any other configuration of proteins would be a lot more complex and therefore vastly more improbable.  I am not a biochemist (are you?) and cannot comment on the truth of this or whether it is also true of the bacterial flagellum.  I think you have to admit this is unproven.

 

4) The attempt to identify simple or simpler patterns through number of concepts is an absolute minefield.  First are we talking about the number of concepts that the agent has in their brain at the time or the number that they could potentially use?   A child has a lot less concepts than an educated adult so their value of Phi_s(T ) and their estimate of CSI is going to be a lot lower. So we lose the objectivity of CSI. So we might want to talk about the number of concepts in some idealised language which children and adults could acquire – but then there is no reason why, for example, every number could not be included as a separate concept and we have an infinite number of concepts.  Even if we draw some boundary round the language that can be used, it is not clear what counts one concept. A concept is not the same as a word. A “motor” can be broken down into many other concepts e.g. machine that converts other forms of energy into mechanical energy and so imparts motion.  We might even consider flagellum to be one concept.  The criterion of simplicity is fairly sensible when applied to noddy examples such as playing cards but it becomes meaningless when applied to real life.

 

5) The definition of H is very flaky.  You admit that you are not certain what Dembski means.  So you adopt your own definition – “a process which does not require the input of information”.  But as we are currently using this formula to clarify what we mean by information this is circular.  It also suggests that the level of CSI is relative to a specific chance hypothesis – not an absolute value for a given outcome.  A popular approach among the ID community is to estimate the probability relative to the hypothesis that all DNA pairs, amino acids or whatever are equally probable and independent of each other.  This makes extensive use of Bernouilli’s principle of indifference with all the associated problems and just isn’t true in real life. But I don’t think you support this approach because in the case of gene duplication you want to include the possibility of gene duplication in the chance hypothesis so you don’t end up with the awkward result that gene duplication doubles the CSI.  But once you admit that knowing about gene duplication radically affects the level of CSI you are open to the possibility that other unspecified or unknown events such as gene duplication can have enormous affects on the supposed CSI.  In other words we cannot even make a rough estimate of CSI without having a good account of all possible natural processes.

 

I haven’t finished with my objections – but I think this will do for the moment.

 

Yours Mark