Review of WA Dembski's
"The Design Inference"
The Design Inference
by William A Dembski.
Cambridge (UK): Cambridge University Press. 243
Reviewed by Wesley R. Elsberry
Department of Wildlife & Fisheries
Sciences, Texas A&M University.
Posted May 6, 2002
In an article appearing in the October 1998 First Things, William A Dembski
announced the existence of rigorous and reliable means of detecting the action
of an intelligent agent. Its description and justification, said Dembski, would
be found in the pages of his new book, The Design Inference (TDI).
Dembski made a special point of applying this criterion, which he called
complexity-specification, to biological phenomena, with the claim
that biologists must now admit design into their science. Dembski's TDI
is a slim and scholarly volume, as one expects from a distinguished academic
press. Dembski employs clear writing, illustrative examples, and cogent
argumentation. The work, though, is motivated and informed by an
anti-evolutionary impulse, and its flaws appear to follow from the need to
achieve an anti-evolutionary aim. The anti-evolutionary bent is not as
overt here, though, as it is in other works by Dembski and fellow Discovery
Institute "Center for the Renewal of Science and Culture" colleagues Phillip
Johnson, Michael Behe, Paul Nelson, and Stephen Meyer. The closest that
Dembski comes within the pages of TDI to explicitly staking out a
position on evolutionary issues is in Section 2.3, where a "case study" is made
of "the creation-evolution controversy". In it, Dembski accuses
evolutionary biologists of rejecting one or more premises of his design
inference in order to avoid coming to a conclusion of design for
The "design inference" of the book's title is an argument to establish that
certain events are due to and must be explained with reference to design.
Dembski crafts his argument as a process of elimination. From the set of all
possible explanations, he first eliminates the explanatory categories of
regularity and chance; then whatever is left is by definition
design. Since all three categories complete the set, design is the
set-theoretical complement of regularity and chance.
Dembski's book and major concept share a name, The Design Inference.
The Design Inference is an argument which leads to a conclusion of
design for an event. Dembski deploys a large number of terms and
phrases in making his argument that design must be recognized as a
necessary mode of explanation in science. Fortunately, Dembski is also
scrupulous in making clear what each term means, even when it has a common or
casual usage. Design is one of those terms, and it becomes a
category defined by the elimination of events that can be attributed to
regularity or to chance.
Complexity-specification is a term used by Dembski in other works
to describe the diagnostic attribute of design in an event. It derives
from Dembski's earlier use of the phrase, Complex Specified
Information. The idea behind complexity-specification is
that the jointly-held attributes of complexity, as small probability, and
specification, as conforming to an independently-given pattern, reveal
the presence of design in an event. Complexity excludes high- and
intermediate-probability events, and specification excludes chance events.
Since regularity comprises events marked by high probability,
complexity-specification then yields those events that fall into
the exclusionary category of design as Dembksi uses the term. Since
these three categories (regularity, chance, and design)
embrace all events, and design is established by elimination of the other two
categories, design is thus the set-theoretical complement of
regularity and chance.
The Design Inference is a deductive argument which can lead to the
recognition of complexity-specification, and thus design,
for a particular event.
Dembski often talks about unpacking terms, and the "unpacked" forms of the 3
categories at issue need to be kept clear. Regularity is simply any event
with high probability. Chance is any event with intermediate or small
probability, but for which no specification exists. And design is any
event with both a small probability and a specification. These are the defining
criteria of Dembski's categories. A conclusion of design for an event means that
the event is not of high probability, intermediate probability, or small
probability without a specification. Dembski's usage of the phrase "due to" also
is somewhat different from standard (contrast with footnote on p 48). Unpacking
"E is due to design" results in: "The proper mode of explanation for E is the
negation of currently known regularity and chance."
The elimination of chance as a category of explanation comprises most of the
book. Chance is acceptable to Dembski as an explanation for all events of
intermediate probability and also for certain events of small probability.
However chance is excluded for events which both have small probability and
conform to a pattern that can be given independently of the event. Such an
independently stated pattern is called a specification by Dembski.
Dembski illustrates the meaning of specifications or "good patterns"
which indicate we can reject chance explanations and fabrications or "bad
patterns" which do not distinguish chance events from those due to other
explanations. If an archer fires an arrow at a wall and plants it in a
previously-painted bull's eye, the bull's eye represents a specification, and
the event of the arrow's hitting the target tells us that the archer has a high
level of skill. If another archer fires an arrow at the wall, then takes a
bucket of paint and draws a bull's eye around his already-implanted arrow, that
bull's eye represents a fabrication, an ad-hoc and after-the-fact pattern
that gives no information about the event with which it is associated.
Another claim of Dembski's is also problematic, and that is the claim that
his Explanatory Filter encapsulates the process of how humans ordinarily
detect design, whether that design is attributed to humans, other animals, or
extra-terrestrial intelligences. Complete with flowchart (p 37), the explanatory
filter has 3 decision nodes. At the first, if an event is deemed to have high
probability, it is classified as due to a regularity, or rather that the
proper explanatory mode for the event in question is regularity or
law-like physical processes. An as-yet unclassified event then moves on the
second decision node. If it has intermediate probability, it is classified as
due to chance. Events that are still not classified then move on to the
third decision node. If the event both has a small probability and also conforms
to a specification, it is classified as due to design; otherwise it is
classified as due to chance.
Dembski makes various forms of the same argument, showing that deduction
leads ineluctably and conclusively to certain events' being due to
design. The catch is that Dembski is using his own definition of
design, where design is simply the explanation that remains after
chance and regularity are eliminated. This is touted by Dembski as
an advantage for the purposes of his argumentation, since he avoids attributing
either causal stories or the intervention of intelligent agency a priori.
In no fewer than 3 separate passages in TDI (p 8, 36, 226-7), Dembski assures
the reader that the design of TDI does not imply agency.
Design and Designer
One may wonder what TDI was supposed to accomplish, if design
no longer means what Paley meant by it and the attribution of agency
no longer follows from finding design. But Dembski believes that
finding design does imply agency, even though he has identified
that implication as being unnecessary. In his view, because we can often find
that design is found where an intelligent agent has acted, we can
reliably infer that when we find design, we have also found evidence of
the action of an intelligent agent. Section 2.4 gives Dembski's take on how we
go from design to agency. Dembski invokes his explanatory filter
as a critical piece of this justification.
Dembski believes that not only design but also agency is found
by his argument. This is the message being spread by various and sundry of the
"intelligent design" proponents and by Dembski himself in other writings. But is
it a secure inference? In his First Things article, and to a lesser
extent in his section 2.3 of TDI, Dembski takes biologists to task for avoiding
the conclusion of design for biological phenomena. Dembski says that to
avoid a design conclusion, biologists uniformly reject one or more of the
premises of his argument. But Dembski does not exclude natural selection as a
possible cause for events which can be classified as being due to design.
The apparent, but unstated, logic behind the move from design to
agency can be given as follows:
an inductive argument. Notice that by the second step, one must eliminate from
consideration precisely those biological phenomena which Dembski wishes to
categorize. In order to conclude intelligent agency for biological examples, the
possibility that intelligent agency is not operative is excluded a
priori. One large problem is that directed contingency or choice
is not solely an attribute of events due to the intervention of an
intelligent agent. The "actualization-exclusion-specification" triad mentioned
above also fits natural selection rather precisely. One might thus conclude that
Dembski's argument establishes that natural selection can be recognized as an
- There exists an attribute in common of some subset of objects known to be
designed by an intelligent agent.
- This attribute is never found in objects known not to be designed by an
- The attribute encapsulates the property of directed contingency or choice.
- For all objects, if this attribute is found in an object, then we may
conclude that the object was designed by an intelligent agent.
It is an error to argue from the casual meanings of regularity, chance, and
design when discussing causes for events classified by Dembski's explanatory
filter or by TDI. Someone might seek to exclude natural selection from
consideration as a source of events that meet the criteria of design by
claiming that it is either a regularity or chance. But TDI classifies
events, not causes. Dembski points this out himself when saying
that the explanatory filter may not always conclude design for an event that we
know is due to the action of an intelligent agent, for agents can mimic the
results of regularity or chance.
The point is broader than Dembski admits. A causal class cannot be lumped
into regularity or chance in advance without begging the question. Specifically,
one cannot state that natural selection is either regularity or
chance. The events which are due to natural selection must be evaluated
by their own properties to establish which category best describes those events.
Just as intelligent agents can sometimes produce events which pass for
regularity or chance rather than design, so too can natural selection be
responsible for events in all 3 categories. It is insufficient to show that some
examples of natural selection fall into either regularity or chance explanation
categories. One arguing that design never has a physical process as an agent
producing an event must show that natural selection is incapable in principle of
producing events with the attribute of design. Such a demonstration would have
to address both the application of natural selection in biology and also in
computer science, where use of the principle of natural selection has been
employed in solving very difficult optimization problems.
I've been thinking about this some since I wrote the review, and natural
selection is but one of a general class of processes which would produce events
meeting the requirements of Dembski's triad. In general, any iterative feedback
system with error-correction will be capable of producing events with the
attributes of Dembski's triad. Natural selection shows that the
"error-correction" part of the process need not have an end solution state for
comparison (see Dawkins' discussion of "distant ideal targets" in "The Blind
Watchmaker"); comparison of merit of currently-existing instantiations is a
sufficient basis for making progress in the parameter-space for many problems.
If the parameters of the system are open to change, then at the end of the
process those parameters will be closer to the optimal set of parameters than
they were at the start. The problem constraints provide the specification of the
optimal parameter settings. During the process, different parameter settings are
actualized. The trajectory of the system through parameter-space will show that
exclusion of other possibilities occurred. This indicates that a much broader
class of processes meets Dembski's triad of criteria for the recognition of
action of "intelligent agents".
It is time to look more closely at Dembski's design inference, to find out if
it does allow us to detect design by the elimination of alternative mechanisms.
The design inference is a deductive argument based on the elimination of
alternatives. Such arguments only work if the conclusion is the result of
exhausting the available alternatives. Dembski assures that this is the case by
defining design to be what is left after regularity and chance
have been eliminated. Thus, what "design" means depends upon how
regularity and chance get eliminated.
Process of Elimination
Dembski offers two somewhat different mechanisms for eliminating regularity.
In the first, regularity is recognized if an event has a high probability of
occurrence. This is part of his discussion of the explanatory filter. The second
method asserts that an event conforms to relevant natural laws, but is not
constrained by them, and thus is not attributable to those laws. This method is
discussed in relation to Dembski's design inference discussion (p 53). It is not
clear that these two methods classify the same set of events as not being due to
regularity. This ambiguity increases our uncertainty concerning the residue that
is left over to be split between chance and design.
There is a difficulty in discussing these concepts in that the meanings of
the terms regularity, chance, and design can become confused with newer meanings
which arise from the argument of The Design Inference. It is important to
keep the casual meanings separate. Unfortunately, it is not clear that even
Dembski manages to keep track of what the terms really mean. For example, even
though Dembski clearly explains that design does not imply agency,
Dembski offers as the 3 possible categories of explanation in his first example
"Regularity", "Chance", and "Agency" (p 11).
According to Dembski, because humans identify human agency using the
explanatory filter, the explanatory filter encapsulates our general method for
detecting agency. Because TDI is equivalent to the explanatory filter,
the conclusion of design in TDI is equivalent to concluding agency.
Dembski specifies a triad of criteria —
actualization-exclusion-specification — as sufficient for
establishing that an intelligent agent has been at work, and finds that
design as he uses it is congruent with these criteria.
However, Dembski's triad of criteria for recognition of intelligent agents is
also satisfied quite adequately by natural selection. "Actualization" occurs as
heritable variation arises. "Exclusion" results as some heritable variations
lead to differential reproductive success. "Specification" occurs as
environmental conditions specify which variations are preferred. By my reading,
biologists can embrace a conclusion of design for an event of biological
origin and still attribute that event to the agency of natural selection.
For comparison, I will propose an alternative explanatory filter and discuss
various points of difference with Dembski's. My alternative explanatory filter
works as follows. An event that cannot be statistically distinguished from a
random event is classified as due to chance. An event that conforms to
properties of known law-like physical processes is classified as being due to
regularity. An event that conforms to known properties of similar events that
are due to intelligent agents are classified as due to design. Any event which
has not yet been classified is now classified as being due to an unknown cause.
My alternative explanatory filter differs in several critical ways. First,
the ordering of decisions is different. Dembski justifies his choice of order
with an explication of explanatory priority (p 38-40). But I find
Dembski's arguments for arranging to eliminate regularity before eliminating
chance to be unconvincing and not reflective of how people ordinarily proceed in
finding explanations. Random events conform well to the null hypothesis (that
is, that the event is due to chance and not to design or regularity) and should
be eliminated first in consideration of causation. Dembski's own example of how
regularity has explanatory priority over chance illustrates the fact that his
filter has the order reversed.
He illustrates his arrangement of explanatory priority using the example of a
pair of loaded dice. Because the loaded dice yield high probabilities for
certain faces' coming up, Dembski explains that the explanation to be preferred
is regularity. However, Dembski ignores the fact that in order to determine that
regularity and not chance is at work with the loaded dice, we must compare the
rolls of the dice to the expectation of "fair" dice. Only when chance is
rejected can we then entertain the notion that the results for the particular
loaded dice in question are due to a regularity. In point of fact, with
sufficient testing and knowledge of the circumstances, the loaded dice example
resolves into an instance of design, not regularity. This does not mean that
design then has explanatory priority. Rather, it illustrates the superior
explanatory power of the alternative filter in which the other explanatory
classes of causation, chance and regularity, had to be considered and rejected
first before design could be concluded.
Second, my explanatory filter has one more alternative classification than
Dembski's, that of unknown causation. This alternative recognizes that
the set of knowledge used to make a classification can alter the classification.
By allowing an event to be classified as due to unknown causation, I
simultaneously reduce the number of false classifications that will later be
overturned due to the availability of additional information and also identify
those events whose circumstances require further study in order to resolve a
causative factor. The use of unknown causation as a category is common in those
day-to-day operations of humans looking for design in events, such as forensics.
Forcing final classification of events under limited knowledge ensures that
mistakes in classification will be made in Dembski's explanatory filter.
Third, my alternative explanatory filter retains the common meaning of design
as a reliable indicator of agency. We recognize design in our day-to-day life
because of prior experience with objects and events designed or caused by
intelligent agents. It is important to recognize that there is a difference
between a reliable classifier and an oracular design detector. Dembski utilizes
the Search for Extra-Terrestrial Intelligence (SETI) project as an example of
the detection of design in the absence of particular knowledge of a designer.
But SETI does not support the notion that novel design/designer
relationships can be detected. SETI is only capable of detecting signals that
conform to certain properties of signals known from prior experience of humans
communicating via radio wavelengths. SETI works to find events that conform to
our prior experience of how intelligent agents utilize radio wavelengths for the
purpose of communication. ETI that communicate in ways for which humans have no
experience will be completely invisible to, and undetected by, SETI.
In summary, the process of detecting design, as it is done by humans in
day-to-day activities, is not accurately captured by Dembski's Explanatory
Filter. The order in which classes of causes are eliminated makes a
difference. Humans attempting to explain phenomena can and often do find
insufficient evidence to make a final determination of either design or any
other explanatory category. And when humans use the word design, they
typically mean it to carry a real implication of being due to an agent, or
Dembski utilizes the Explanatory Filter and equivalent logical arguments in
order to place his criterion of design on a deductive footing. That
criterion, complexity-specification, does not help us to identify
a cause, or an agent, of an event. Its sole purpose is to detect design
as Dembski employs the term. The step from detection of design
to implication of an intelligent agent is made via an inductive
argument, and shares in the problems of all conclusions drawn from an inductive
basis. Dembski argues that a triad of criteria reliably diagnoses the
action of an intelligent agent, yet this same triad of criteria fails to exclude
natural selection as a possible cause of events that have the attribute of
complexity-specification. Somehow, I doubt that natural selection
is what Dembski had in mind for the agent of biological design.
The Design Inference is a work with great significance for the group
of anti-evolutionists who have embraced "intelligent design" as their organizing
principle. TDI is supposed to establish the theoretical foundation for
all the rest of the movement. My judgment is that it fails to lay a solid
foundation. There are flaws and cracks that can admit the entry of naturalistic
causes into the pool of "designed" events. It is unfortunate that
Dembski's focus is the establishment of "intelligent design" as an
anti-evolutionary alternative, for his insights into elimination of chance
hypotheses would appear to have legitimate application to various outstanding
research questions, such as resolving certain issues in animal cognition and
intelligence. Despite Dembski's commentary in his First Things article,
there appears to be no justification for the claim that biologists must now
admit design (in its old, agency-laden sense) into biological explanation to any
greater degree than it is already used.
Dembski WA. Science and design. First Things 1998 Oct; 86:21-2. Accessed April 8, 1999.
Dembski WA. The Explanatory Filter: A three-part filter for understanding how
to separate and identify cause from intelligent design. Accessed March 8, 1999.
Dembski WA. Intelligent design as a theory of information. Conference on
Naturalism, Theism, and the Scientific Enterprise (Austin, Texas). Accessed March 8, 1999.
[Thanks to Bob Schadewald and others who gave helpful commentary on drafts of
Wesley R. Elsberry is a Ph.D. student in the Department of Wildlife and
Fisheries Sciences at Texas A&M University. His research involves
computational analysis and modelling of cetacean biosonar and cognition.
See also The
Anti-Evolutionists: William A. Dembski: Online resources and commentary.
Wesley R. Elsberry's home page