By Ben Kissling
Introduction
The argument from information in contemporary terms is a novel teleological argument for the existence of God with its deepest roots in the mid-20th century. Most would describe its origins differently, including many proponents of this argument. They would begin their history of it with William Paley’s “watch in the heath” argument from 1802. Both arguments point to an object with unknown origin and reason from features of the object to the conclusion that it was designed by some intelligence rather than “naturally” occurring. Some modern design arguments are similar to Paley’s, such as Michael Behe’s irreducible complexity or Douglas Axe’s functional coherence. However, these types of arguments do not appeal to the concept of information as William Dembski’s specified complexity argument does. Ideas like irreducible complexity and Paley’s watch argument mostly appeal to a common-sense inference from the mechanical aspects of biology. But an argument from information is a different type of argument. Paley’s and others’ explication of what these features are simply could not have been understood the way modern information argument proponents understand it until the advent of information theory in the mid-20th century. Applying this argument to living things had to wait until the discovery of the structure of DNA in 1953, which showed that DNA is quite literally a digital message string exactly as information theory describes. Information theory deals with coded digital messages usually sent over long distances with modern technology. The contemporary argument from information as formulated by Dembski and others is intended to be the general logical form of many different types of design inferences, including all the ones mentioned above. In this article, I will explain what the contemporary argument from information is and show why it is among the most compelling arguments for the existence of God today.
What is Information?
First, we need to understand the difference between analog and digital information. Most people think “digital” just means “electronic.” However, in information theoretic terms, “digital” means “discrete” or “discontinuous.” Analog information is like the lines of a drawing or a geometric curve on a graph. In order to exactly copy this type of analog information, one would have to have a perfect replica of the drawing at every point of every line. Digital information is much different, because it only depends on the characters in a message string and the code in which those characters are used. If I changed the font of this document, that would change the analog information, but it would not change the digital information. Digital information is superior to analog in the sense that it is much harder to copy analog information or translate it to some other medium. Copying the Mona Lisa, for example, is making a copy of analog information. Copying this sentence is copying digital information, and it does not depend on one’s handwriting skill in the same way that copying the Mona Lisa would depend on one’s painting skill. Digital information is the kind of information in DNA, and the kind referred to in the context of the information argument.
Digital information is measurable. Because it depends on a code, for example the twenty-six letters of the English alphabet, the number of total possible message strings can be compared to the number of message strings allowed by a particular message to get a numerical ratio. So let’s say we use the alphabet with no additional characters to spell out the phrase “METHINKSITISLIKEAWEASEL.” Because there are twenty-three positions in this string, and twenty-six possible characters at each position, we can calculate the ratio of strings we actually have, which is one, compared to the total number of possible strings, which is 26^23, or ~3.5×10^32. Since we only have one string, this ratio would simply be 1/3.5×10^32 or 3.5×10^-32. Notice that if we increase the number of characters in the code, say by adding some punctuation characters like spaces and periods, this increases the denominator and makes the ratio smaller. Whereas, if we increase the number of possible message strings, say by making some of the letters in our message string variables that could be any character, that would increase the number in the numerator, making the final ratio bigger. Intuitively, we know that the amount of information would decrease if there were more possible message strings, meaning the message is less specific, and increase if the code were more complex, thus excluding more possibilities. So in information theory we simply take the reciprocal of this ratio as our raw information measure. Normally it is converted into base 2, represented as a logarithm, and called “bits.”
Notice that DNA can be converted into an information metric in precisely this way. Since DNA has four characters, A, T, C, and G, we can take any length of DNA and turn it into a number 4^n, where n is the length of the DNA segment counted in bases. Protein sequences can also be converted into information metrics similarly. Since there are twenty amino acids normally in proteins (there are a few extremely rare additional ones), we can take any length of protein and turn it into a number 20^n, where n is the length of the protein sequence. We will return to this later.
Conceptually then, information is measured by comparing the number of possible messages represented by our message string to the total number of possible messages for any message string of the same length. This identifies information as a reduction of uncertainty. A character string of “xxxxxxxxxx” where “x” could be any number between zero and nine has no reduction of uncertainty and therefore no information. Whereas, a string of 0123456789 represents only one possible string out of 10^10 total possible strings for this length and this code. That is quite a lot of information. But the key point is that even very short message strings can contain a high amount of information by excluding a large number of possible message strings. And even having the same message string but making the code more complex eliminates more possibilities, so information can increase by changing the code even though the message string itself doesn’t change. Given that the size of the code used, which dictates the number of possibilities excluded, is usually the determining factor in how much information we have, it makes more sense in information theory to talk about the number of possibilities excluded than the number we actually have, which is usually only one, that is, the one identified by our message string. This is why information is defined as the exclusion of possibilities, also stated as a reduction in uncertainty.1 This is called Shannon information after its originator, Claude Shannon.2
Complex Specified Information
Shannon information does not necessarily have to mean anything. It can be a random jumble of characters signifying nothing. But there does seem to be a kind of information that intuitively seems to come from intelligence, like the words on this page. William Dembski noticed this and formulated a definition of this special kind of information. He called it specified complexity, or complex specified information. Complexity is essentially the same thing as Shannon information; it’s a measure of the improbability of an occurrence. This is just the thing that actually occurs compared to the number of possible things which could have occurred. Specification is harder to define. Originally, Dembski defined it as an independent pattern.
More recently, however, Dembski and Ewert have defined a specification as similar to a different kind of complexity called Kolmogorov complexity. Kolmogorov complexity is the shortest algorithm that can reproduce some Shannon information.3 For example, if we have two copies of the same message string, a Shannon information measure would count that as double the amount of information of a single copy, but it seems intuitive that this is not really double the amount of information. Kolmogorov complexity would say that one can make an algorithm which is shorter than the two copies simply by having only one copy of the Shannon information, plus an algorithmic command to repeat it once, reproducing two copies of the same Shannon information. This would be much less information measured as Shannon information than two entire copies, but slightly more than just one copy. So Kolmogorov complexity is like the absolute minimum amount of algorithmic information required to represent some amount of Shannon information. This sounds somewhat redundant, but as the example above shows, it is a useful concept.
Similarly, Dembski and Ewert have defined a specification as a short description, or perhaps the shortest description.4 This is another information measure of how much information we have. Instead of being measured according to the number of possibilities excluded, a specification refers to the information measure of its description. They mean this quite literally in terms of some language and the minimum number of words, like Kolmogorov complexity, needed to describe the artifact in question. They discuss how information measured in this way is somewhat relative to the language used but show that most real languages give similar results. For example, English has about 200,000 words in common usage, and therefore one word in English is about 17.6 bits, which they round up to 20 bits.5 They give an example of how to apply this to a protein function like “binds ATP” whose two words would constitute a 40-bit specification.6 They give an equation for how to relate complexity metrics with specification metrics to give a numerical result for specified complexity. A large result means a large amount of specified complexity. In words, a low probability (which is a high complexity) combined with a short description constitutes specified complexity.
Dembski has always argued that specified complexity is the property of objects we intuitively recognize as the artifact of intelligent design.7 Specified complexity represents the feature we see in letters drawn in the sand, or Mount Rushmore. These arrangements are highly improbable, but of course virtually any arrangement of grains of sand on a beach or rock faces is highly improbable. It is not just improbability but improbability paired with a specification that indicates design. Other examples, say a vortex swirling in a liquid or a crystal growing into a precise geometric shape, have short descriptions, i.e. specifications, but with high probability. The high probability here is due to laws restricting possibilities so that the occurrence of the vortex or crystal is a highly probable result. So even though it seems to have a pattern that we might identify as a specification, a design inference is not triggered because the event is not highly improbable. But if we see both, as in Mount Rushmore or a message drawn in the sand, we infer design. Specified complexity is a rigorous way to show this feature of designed objects.
The famed fine-tuning argument is a special case of specified complexity. Two things are required: a very low probability and a specification. In the fine-tuning argument, the low probabilities of the various natural constants are combined with the specification of a natural world that supports complex life. Luke Barnes has shown how the range of possible values compared to the range of values which are life permitting can be calculated.8 The improbability of all these constants fitting into those ranges, especially in combination, is more than astronomical. “Universe supporting complex life” constitutes the short description required for a specification. Extremely high complexity (low probability) corresponds to a short description (specification), and we are justified in inferring design.
The same fine-tuning argument can also be applied to biology.9 In fact, complexity calculations, as already shown above, are much easier in the context of DNA and the genetic code. They are even more straightforward in the case of protein sequences. In the case of specifications, proteins are also quite straightforward, as most of them already have short descriptions in the existing scientific literature that usually are a succinct description of their function. Naturally occurring proteins almost all have high specified complexity and result in a design inference according to this formulation of the information argument. There are some exceptions which generally prove the rule due to either their short sequence length, low specificity of function, or both, such as the ice-fish anti-freeze proteins which are best explained as the result of random mutations.10
Various objections have been raised to this general argument, but they usually fall into two categories. They both claim that Dembski’s theory is wrong and that natural, that is, non-designed processes can generate complex specified information. One way of trying to show this is through the use of computer models called “evolutionary” or “genetic” algorithms. Programs like Ev and Avida were claimed to have created exactly this sort of complex specified information through evolution-like processes programmed into the algorithm involving random variations acted on by a simulation of natural selection. Dembski, Robert Marks II, and Winston Ewert over several years published a series of papers showing how these programs all cheated by including what they called “active information” into the program ensuring that it would succeed.11 Active information is information added by the programmers or designers of the program that made it look like it was being generated by these simulated evolutionary computations. In this same research, they proposed the Law of Conservation of Information.12 Essentially, this says that any information output of such an algorithm must be equal to or lesser than the information input. Dembski and Ewert are working on a forthcoming book on all of this research.13
Complex Specified Information in Biology
Another class of objections appeal to work in biochemistry rather than computer simulations that shows evolutionary or biochemically simulated evolutionary processes creating complex specified information. By the early 2000s, most of these types of objections were not much more than hand-waving or anecdotal. But then a biochemist named Douglas Axe published a series of papers culminating in hard evidence that the number of stable, functional protein folds is extremely rare in the protein sequence space. This is similar to Luke Barnes showing how rare the values for the fine-tuned constants are in the possible range of values. Both show that the actual values and sequences we have are extremely rare, thus exhibiting extremely high complexity. Axe was even kicked out of his lab just prior to completing his research for apparently ideological reasons.14 He was able to publish his final paper showing that the rarity of a stable, functional folding domain for a protein called TEM-1, a commonly used beta-lactamase enzyme, was between 10^-64 and 10^-77.15 This immediately suggested that functional protein folding domains are designed by an intelligence. Axe also calculated that any stable, functional protein fold of sequence length 153, the same length as the folding domain he studied, would have a probability in the sequence space of 10^-74.
Several objections were raised over the years since 2004. Some had technical objections, but none have been published. Axe has publicly stated that one was submitted for publication which he reviewed and subsequently was not published.16 Other objections claimed that another way of doing the experiment gives a different result, namely a much larger probability for stable functional protein folding domains on the order of 10^-11 to 10^-12. Dembski and Ewert mentioned such examples and argued that the function in question is common enough that design is not indicated in such cases.17 But a more direct answer to this objection is that these studies were in vitro, not in vivo. Axe’s study was done inside a real living organism, and these other studies were done in test tubes. The difference is that in a test tube there are no other biological molecules around to interfere or be interfered with. A protein in a real biological context doesn’t just have to perform its function. It must also avoid performing other functions; that is, it must avoid interactions with other biological molecules that would interfere with their functions or have its own function interfered with. This likely means that the constraints on the proper functioning of proteins in a real biological context are much higher than in a test tube. In other words, it doesn’t matter if a protein can bind ATP in a test tube if it would bind nearly everything inside a real organism, thus killing it. That is why real, naturally occurring proteins require such high specificity. It is not only the function that must happen; it must also exclude all other functions. This objection fails to recognize a key feature of information: excluded possibilities are much more numerous and more important in information measures than the possibility that actually obtains. These in vitro studies generate a large number of random sequences from scratch and test for bare function. Most of them are not naturally occurring and were not tested in a real organism.
In 2006, Dan Tawfik’s lab published a study18 which generalized and confirmed Axe’s number of 10^-74. Tawfik studied ten proteins of different types and found that two out of every three random mutations reduced the thermodynamic stability of their folds by 1 kcal/mol or more. Since Axe’s version of TEM-1 was barely stable, this means that in this context any such mutation would likely have destabilized the fold. Since two in three mutations destroy the fold and therefore the function resulting from the fold, it is a simple matter to calculate the number of sequences which would likely work versus not. The number of amino acids that would work at each position is 1/3, and if the sequence is 153 amino acids long, simply take the number of characters that work to the power of the number of positions in the sequence, just as we did for DNA above: (1/3)^153 = 10^-73.19 This is within a single order of magnitude of Axe’s calculation, confirming Axe’s result.
Intelligence is the Best Explanation for Complex Specified Information
But why does specified complexity indicate intelligent design? This is a more philosophical question that was taken up by Stephen Meyer.20 Meyer takes Dembski’s and Axe’s work, along with others, and puts the finishing touches on it. Meyer studied the philosophy of science of Charles Darwin and one of his major influences, geologist Charles Lyell.21 Meyer notes that Lyell argued that historical science, the science of understanding natural history and origins, 22 should utilize causes now in operation to explain past events. Meyer simply asked the question, “What is the cause now in operation that explains complex, specified information?” He argues this cause is intelligence, and so we can infer intelligent cause for complex, specified information by the same rules that Lyell, Darwin, and the historical sciences generally use.
Scientific inferences today are understood as inferences to the best explanation, which is technically called an abductive argument.23 Abductive inferences, or inferences to the best explanation, cannot rule out all other possible explanations. All they can do is compare competing explanations to see which one is better. In order to do this, scientists weigh competing hypotheses and try to test them against each other to see which one survives the tests. Meyer argues that intelligent design is the best explanation for information in precisely this way and therefore follows the rules of scientific inference and succeeds as the best explanation for complex, specified information. 24
The argument from information is thus well supported both scientifically and philosophically in two major examples: the physical constants according to fundamental physics and the information in DNA, both required for life to exist. This argument persuaded the famous atheist philosopher Antony Flew to admit that God exists25 and has been mentioned by many prominent atheists as being the most compelling argument for theism. The current argument from information is different and stronger than past versions of the design argument. It has identified the rigorous logical, mathematical, and philosophical basis of the intuitive design inference we all use every day. This argument has been applied to artifacts in nature which could not have been designed by human beings, some of which could only have been created by a supernatural intelligence simply because no natural intelligences could have been around at the time. It is one of if not the best current argument for the existence of God.
Footnotes
[1] William A. Dembski, Intelligent Design: The Bridge Between Science and Theology (Downer’s Grove, Illinois: InterVarsity Press, 1999), 154.
[2] William A. Dembski, The Design Revolution: Answering the Toughest Questions about Intelligent Design (Downer’s Grove, Illinois: InterVarsity Press, 2004), 135.
[3] William A. Dembski, No Free Lunch: Why Specified Complexity Cannot be Purchased Without Intelligence (Lanham: Rowman & Littlefield Publishers, Inc., 2002), 59-62.
[4] William A. Dembski and Winston Ewert, The Design Inference: Eliminating Chance through Small Probabilities, 2nd ed. (Seattle: Discovery Institute Press, 2023), 151-152.
[5] Ibid., 348.
[6] Ibid., 427.
[7] William A. Dembski, The Design Revolution: Answering the Toughest Questions about Intelligent Design (Downer’s Grove, Illinois: InterVarsity Press, 2004), 81-82.
[8] Luke Barnes, “A Reasonable Little Question: A Formulation of the Fine-Tuning Argument,” Ergo 6, no. 42 (2019-2020). https://quod.lib.umich.edu/e/ergo/12405314.0006.042/–reasonable-little-question-a-formulation-of-the-fine-tuning?rgn=main;view=fulltext (accessed December 10, 2024).
[9] Steinar Thorvaldsen and Ola Hossjer, “Using statistical methods to model the fine-tuning of molecular machines and systems,” Journal of Theoretical Biology 501, (Septermber 2020). https://www.sciencedirect.com/science/article/pii/S0022519320302071 (accessed December 10, 2024).
[10] Abirami Baskaran et al. “Anti freeze proteins (Afp): Properties, sources and applications – A review,” International Journal of Biological Molecules 189 (31 October, 2021): 292-305.
[11] William A. Dembski and Robert J. Marks II, “The Search for a Search: Measuring the Information Cost of Higher Level Search,” Journal of Advanced Computational Intelligence and Intelligent Informatics 14, no. 5 (2010): 475-486.https://evoinfo.org/publications/search-for-a-search.html (accessed December 10, 2024).
[12] William A. Dembski and Robert J. Marks II, “LIFE’S CONSERVATION LAW: Why Darwinian Evolution Cannot Create Biological Information,” in The Nature of Nature, ed. Bruce Gordon and William Dembski (Wilmington, Del.: ISI Books, 2009), 381.
[13] William A. Dembski and Winston Ewert, The Design Inference: Eliminating Chance through Small Probabilities, 2nd ed. (Seattle: Discovery Institute Press, 2023), 472.
[14] Douglas Axe, Undeniable: How Biology Confirms Our Intuition That Life Is Designed (n.p.: HarperOne, 2016), 49-51.
[15] Douglas D. Axe, “Estimating the Prevalence of Protein Sequences Adopting Functional Enzyme Folds,” Journal of Molecular Biology 341, no. 5 (August 2004): 1295-1315. https://www.sciencedirect.com/science/article/abs/pii/S0022283604007624 (accessed December 10, 2024).
[16] Sean McDowell. The Debate over Evolution and Intelligent Design (w/ Doug Axe). https://youtu.be/YGggxHBqPRc?t=1557 (accessed July 3, 2025)
[17] William A. Dembski and Winston Ewert, The Design Inference: Eliminating Chance through Small Probabilities, 2nd ed. (Seattle: Discovery Institute Press, 2023), 427.
[18] Shimon Bershtein et al. “Robustness – epistasis link shapes the fitness landscape of a randomly drifting protein,” Nature 444 (2006): 929-932.
[19] Brian Miller, “A Dentist in the Sahara: Doug Axe on the Rarity of Proteins Is Decisively Confirmed,” Evolutionnews.org, February 18, 2019. https://evolutionnews.org/2019/02/a-dentist-in-the-sahara-doug-axe-on-the-rarity-of-proteins-is-decisively-confirmed/ (accessed December 10, 2024).
[20] Stephen C. Meyer, “The Origin of Biological Information and the Higher Taxonomic Categories,” Proceedings of the Biological Society of Washington 117, no. 2 (2004): 213-239. https://www.discovery.org/a/2177/ (accessed December 10, 2024).
[21] Stephen C. Meyer, Signature in the Cell: DNA and the Evidence for Intelligent Design (n.p.: HarperOne, 2009), 160.
[22] Ibid., 409.
[23] Ibid., 152-154.
[24] Stephen C. Meyer, Signature in the Cell: DNA and the Evidence for Intelligent Design (n.p.: HarperOne, 2009), 347-348.
[25] Antony Flew, How the World’s Most Notorious Atheist Changed His Mind (n.p.: HarperOne, 2008)
© Adam Lloyd Johnson and Convincing Proof.