font face=arial size=-1 color=black>Yahoo! is not affiliated with the authors of this page or responsible for its content.
Specification: The Pattern That Signifies Intelligence
1
Specification: The Pattern That
Signifies Intelligence

By William A. Dembski

August 15, 2005, version 1.22

1 Specification as a Form of Warrant ..... 1
2 Fisherian Significance Testing .. 2
3 Specifications via Probability Densities .. 5
4 Specifications via Compressibility . 9
5 Prespecifications vs. Specifications . 12
6 Specificity 15
7 Specified Complexity 20
8 Design Detection 25
Acknowledgment ...
31
Addendum 1: Note to Readers of TDI & NFL ... 32
Addendum 2: Bayesian Methods 35
Endnotes ......38


ABSTRACT: Specification denotes the type of pattern that highly improbable
events must exhibit before one is entitled to attribute them to intelligence. This
paper analyzes the concept of specification and shows how it applies to design
detection (i.e., the detection of intelligence on the basis of circumstantial
evidence). Always in the background throughout this discussion is the
fundamental question of Intelligent Design (ID): Can objects, even if nothing is
known about how they arose, exhibit features that reliably signal the action of an
intelligent cause? This paper reviews, clarifies, and extends previous work on
specification in my books The Design Inference and No Free Lunch.


1 Specification as a Form of Warrant

In the 1990s, Alvin Plantinga focused much of his work on the concept of warrant. During that
time, he published three substantial volumes on that topic.
1
For Plantinga, warrant is what turns
true belief into knowledge. To see what is at stake with warrant, consider the following remark
by Bertrand Russell on the nature of error:

Error is not only the absolute error of believing what is false, but also the quantitative error of believing
more or less strongly than is warranted by the degree of credibility properly attaching to the proposition
believed in relation to the believers knowledge. A man who is quite convinced that a certain horse will win
the Derby is in error even if the horse does win.
2


It is not enough merely to believe that something is true; there also has to be something backing
up that belief. Accordingly, belief and persuasion are intimately related concepts: we believe that
2
about which we are persuaded. This fact is reflected in languages for which the same word
denotes both belief and persuasion. For instance, in ancient Greek, the verb peitho denoted both
to persuade and to believe.
3


Over the last decade, much of my work has focused on design detection, that is, sifting the
effects of intelligence from material causes. Within the method of design detection that I have
developed, specification functions like Plantingas notion of warrant: just as for Plantinga
warrant is what must be added to true belief before one is entitled to call it knowledge, so within
my framework specification is what must be added to highly improbable events before one is
entitled to attribute them to design. The connection between specification and warrant is more
than a loose analogy: specification constitutes a probabilistic form of warrant, transforming the
suspicion of design into a warranted belief in design.

My method of design detection, first laid out in The Design Inference and then extended in No
Free Lunch,
4
provides a systematic procedure for sorting through circumstantial evidence for
design. This method properly applies to circumstantial evidence, not to smoking-gun evidence.
Indeed, such methods are redundant in the case of a smoking gun where a designer is caught red-
handed. To see the difference, consider the following two cases:

(1) Your daughter pulls out her markers and a sheet of paper at the kitchen table while
you are eating breakfast. As youre eating, you see her use those markers to draw a
picture. When youve finished eating, she hands you the picture and says, This is for
you, Daddy.

(2) You are walking outside and find an unusual chunk of rock. You suspect it might be
an arrowhead. Is it truly an arrowhead, and thus the result of design, or is it just a
random chunk of rock, and thus the result of material forces that, for all you can tell,
were not intelligently guided?

In the first instance, the design and designer couldnt be clearer. In the second, we are dealing
with purely circumstantial evidence: we have no direct knowledge of any putative designer and
no direct knowledge of how such a designer, if actual, fashioned the item in question. All we see
is the pattern exhibited by the item. Is it the sort of pattern that reliably points us to an intelligent
source? That is the question my method of design detection asks and that the concept of
specification answers.


2 Fisherian Significance Testing

A variant of specification, sometimes called prespecification, has been known for at least a
century. In the late 1800s, the philosopher Charles Peirce referred to such specifications as
predesignations.
5
Such specifications are also the essential ingredient in Ronald Fishers theory
of statistical significance testing, which he devised in the first half of the twentieth century. In
3
Fishers approach to testing the statistical significance of hypotheses, one is justified in rejecting
(or eliminating) a chance hypothesis provided that a sample falls within a prespecified rejection
region (also known as a critical region).
6
For example, suppose ones chance hypothesis is that a
coin is fair. To test whether the coin is biased in favor of heads, and thus not fair, one can set a
rejection region of ten heads in a row and then flip the coin ten times. In Fishers approach, if the
coin lands ten heads in a row, then one is justified rejecting the chance hypothesis.

Fishers approach to hypothesis testing is the one most widely used in the applied statistics
literature and the first one taught in introductory statistics courses. Nevertheless, in its original
formulation, Fishers approach is problematic: for a rejection region to warrant rejecting a
chance hypothesis, the rejection region must have sufficiently small probability. But how small
is small enough? Given a chance hypothesis and a rejection region, how small does the
probability of the rejection region have to be so that if a sample falls within it, then the chance
hypothesis can legitimately be rejected? Fisher never answered this question. The problem here
is to justify what is called a significance level such that whenever the sample falls within the
rejection region and the probability of the rejection region given the chance hypothesis is less
than the significance level, then the chance hypothesis can be legitimately rejected.

More formally, the problem is to justify a significance level (always a positive real number
less than one) such that whenever the sample (an event we will call E) falls within the rejection
region (call it T) and the probability of the rejection region given the chance hypothesis (call it
H) is less than (i.e., P(T|H) < ), then the chance hypothesis H can be rejected as the
explanation of the sample. In the applied statistics literature, it is common to see significance
levels of .05 and .01. The problem to date has been that any such proposed significance levels
have seemed arbitrary, lacking a rational foundation.
7


In The Design Inference, I show that significance levels cannot be set in isolation but must
always be set in relation to the probabilistic resources relevant to an events occurrence.
8
In the
context of Fisherian significance testing, probabilistic resources refer to the number opportunities
for an event to occur. The more opportunities for an event to occur, the more possibilities for it to
land in the rejection region and thus the greater the likelihood that the chance hypothesis in
question will be rejected. It follows that a seemingly improbable event can become quite
probable once enough probabilistic resources are factored in. Yet if the event is sufficiently
improbable, it will remain improbable even after all the available probabilistic resources have
been factored in.

Critics of Fishers approach to hypothesis testing are therefore correct in claiming that
significance levels of .05, .01, and the like that regularly appear in the applied statistics literature
are arbitrary. A significance level is supposed to provide an upper bound on the probability of a
rejection region and thereby ensure that the rejection region is sufficiently small to justify the
elimination of a chance hypothesis (essentially, the idea is to make a target so small that an
archer is highly unlikely to hit it by chance). Rejection regions eliminate chance hypotheses
when events that supposedly happened in accord with those hypotheses fall within the rejection
4
regions. The problem with significance levels like .05 and .01 is that typically they are instituted
without reference to the probabilistic resources relevant to controlling for false positives.
9
Once
those probabilistic resources are factored in, however, setting appropriate significance levels
becomes straightforward.
10


To illustrate the use of probabilistic resources here, imagine the following revis