More Options

Darwin in Mind: ‘Intelligent Design’ Meets Artificial Intelligence

Article

Taner Edis

Volume 25.2, March / April 2001

What’s Wrong with the Information Argument Against Evolution?
Proponents of “Intelligent Design” claim information theory refutes Darwinian evolution. Modern physics and artificial intelligence research turns their arguments on their head.

Science no longer treats nature, particularly life, as a supernatural design. Today, the very mention conjures up images of young-Earth creationists with their bizarre scriptural literalism. Even the interesting questions creationists raise (Edis 1998a) are overshadowed by the weirdness produced by leaders such as Henry M. Morris, who can-with a straight face-go on about Satan using psychic powers to deceive Eve (1993).

There are, of course, more liberal views. Theologians interpret evolution as a progressive spiritual development, the creative influence of an infinite God pouring out onto a finite world (Haught 1999). Others speculate about whether the accidents of evolution were supernaturally tweaked to ensure we turn up (Peacocke 1986), or if evolution was set in motion by a creative purpose (Wright 2000). Meanwhile, biologists work with blind mechanisms, and any “progress” in evolution is an artifact of the fact that life started out simple (Nitecki 1988). Liberal notions of design are relatively harmless, mainly because they are only loosely connected to modern biology.

Lately, an “Intelligent Design” (ID) movement has been emerging, trying to steer a course between the inconsequential handwaving of the liberals and the lunatic literalism of the creationists. It too promises more than it has delivered. Phillip Johnson, perhaps their most prominent spokesman, forcefully condemns evolutionary naturalism (1991, 1995) but presents no serious alternative. Michael Behe (1996) claims instances of “irreducible complexity” in biology, which adds up to little more than an old-fashioned incredulity about achieving complex interdependent structures incrementally. The effect of ID on mainstream science has been negligible.

Even so, ID has scored a few philosophical points. Defenders of evolution often hope a tame science and a defanged religion can peacefully occupy separate spheres. Science, we declare, is “methodologically naturalistic,” considering only naturalistic explanations while saying nothing about any deeper supernatural reality (Pennock 1996). But intelligent design is a straightforward fact claim, one which is true about those objects we make ourselves. That an intelligent agent designed some aspects of nature is also a legitimate hypothesis. If science can say nothing about the probable truth or falsity of such a claim, there must be something wrong with our understanding of science. So ID advocates correctly argue that science cannot be restricted to a predefined set of naturalistic possibilities (Moreland 1994). A theoretically sophisticated, empirically well-anchored ID hypothesis can be a serious scientific proposal.

But then the problem is finding such a proposal. Ineffectual complaints about evolution in the Johnson and Behe style are not enough, so skeptics easily dismiss ID as thinly disguised creationism.

Intelligent Design and William Dembski

Enter William Dembski. Already known as one of the better ID proponents, he has recently gathered his arguments in a book that claims to put ID on a solid footing (Dembski 1999). Surprisingly, he is often correct. Though dead wrong in his overall conclusions, he makes interesting mistakes, and his errors highlight how powerful an idea Darwinian evolution is, in biology and beyond.

Dembski sets out to fashion a workable notion of supernatural intervention. One difficulty is that a miracle sounds like an all-purpose excuse rather than a genuine explanation. And even if we allow a design hypothesis in analogy to human creativity, this is easily abandoned at the first hint of a naturalistic alternative. Dembski therefore proposes to detect intelligent action in a way that avoids becoming an excuse or a weak analogy. We distinguish design from accident, he says, by seeing if our data exhibit contingency, complexity, and specification.

Contingency means an information-conveying system must allow many possible arrangements. Not all order is evidence of purpose. Objects we drop fall rather than drift off in random directions, but this only manifests a simple physical law. In contrast, it is as easy to type “urqgkwffferj . . .” as to type a real argument; an isolated string of nonsense-DNA is no different in chemical stability from one that codes for a useful protein.

To rule out pure chance taking over in the absence of simple constraints, Dembski demands complexity. A world of physical laws and random events will occasionally produce something that makes sense, like a monkey at a typewriter banging out “hello world.” But the longer and more complex the message, the more unlikely this is.

Specification is crucial for telling what sort of data is meaningful. Finding (pi) encoded in a radio signal from space would suggest an intelligent source, while any particular random string, though just as improbable, is merely noise. We must be able to specify meaningful patterns before the fact; otherwise, given thousands of crank-hours at work, we can find messages in anything, such as a plan of history in the Great Pyramid.

Dembski argues that such criteria can be made rigorous (1998). Inferring design-or distinguishing messages from noise-is an important problem, from everyday interpretation of ambiguous data in a social context to SETI research. For example, astronomers first wondered if periodic signals from pulsars indicated alien life, but the signals were too simple and soon a physical explanation was found. Dembski formalizes requirements like complexity, defining a procedure to detect design.

Dembski’s information-theoretic work is fairly respectable.(1) The controversy begins when he applies his criteria to biology, finding that life exhibits just the sort of specified complexity that is supposed to signify intelligent design. ID proponents claim to improve on the classical design argument by providing a rigorous procedure to identify a particular sort of order indicating intelligent origin. When tested on objects we know the origins of, they say, this procedure reliably sorts out artifacts from the haphazardly cobbled together, even when we know little about the functions of the artifacts. So it looks like organisms are also, at some level, products of design.

ID needs more, since its criteria might fail to distinguish between explicit design and evolution-both may generate specified complexity (Elsberry 1999). ID proponents attack this in two ways. One is to produce the usual litany of alleged failures of Darwinian “macroevolution": the origins of life, the Cambrian explosion, Behe’s “irreducibly complex molecular machines,” and so forth. This is the tedious, disreputable side of ID. The second way, however, extends the information-theoretic argument, promising to show why a Darwinian mechanism cannot create specified information.

Darwinism must fail, Dembski says, because information is conserved. Unintelligent processes that transform and transmit information can never add new content. Consider a message string, “3:45 p.m.” This might be translated into “15:45"; no information is gained or lost thereby. Or it might be degraded by a process that rounds times to the nearest hour, leaving “4:00 p.m.” If the message was input to a computer program that e-mailed meeting times to a department staff, it might be converted to “Next department meeting: 3:45 p.m.,” but the additional comment, though useful, is not really new. Such a program could only be used to transmit meeting times; this information is built into its initial design.

Random processes do no better. A noisy channel might, with a lot of luck, produce “Christmas party: 3:45 p.m.,” but there is no reason to trust it. Variation-and-selection can add no meaningful novelty to a message because all it does is reveal information in pre-programmed selection criteria. According to ID, the creativity producing information-rich structures like living beings cannot be captured by blind naturalistic processes.

Physics and Intelligent Design

To see what is wrong here, we can cast ID as a physical claim. First, take a universe with dynamical laws like those of Newtonian physics. These conserve information at a microscopic level; a complete description of particle positions and velocities at any time also determines all past and future states. Following Dembski, we might suspect that if complex structures appear at some point, this is not a genuine novelty, since these were implicit in previous states.

However, such a scenario does not preclude evolution. It suggests a clockwork deism, where the information provided through the initial design unfolds in time, manifesting in complex macroscopic structures. This still leaves the question of how these local pockets of specified complexity are assembled. Variation-and-selection may still do the job.

This issue is related to one of the classic problems of physics: understanding an irreversible macroscopic world, which does not appear to conserve information, when our basic microscopic dynamics are reversible. Part of the answer comes from realizing we never have a complete description of any system. What approximate knowledge we have rapidly becomes obsolete due to dynamical chaos, as even the smallest error grows exponentially. We can only keep track of statistical properties of systems, through macroscopic variables like temperature, which behave irreversibly (Gaspard 1992). For example, if we bring objects at different temperatures into contact and let them reach equilibrium, they will end up the same temperature. No measurement can recover their original temperatures, and they will not spontaneously acquire different temperatures again.

Such loss of information does not challenge ID; it even plays into creationist suspicions that the second law of thermodynamics precludes evolution. But the same physics also underlies the emergence of order from chaos. If a system behaves such that its maximum possible entropy increases faster than its actual entropy, it will be driven away from equilibrium. This creates space for order to form. In particular, Darwinian processes can take hold: simple replicating structures can mutate and diversify, exploring more complex configurations along the way. All this takes place under ordinary physics, without outside intervention (Brooks and Wiley 1988, Edis 1998a).

The information-based arguments of ID, then, allow design to be confined to setting up initial conditions. Hence they are too broad to support a critique of evolution. In fact, the situation is worse, as the deistic view is itself highly dubious.

Focusing on microscopic information and deterministic dynamics can give the impression the physics of complexity is a nuisance foisted on us because of our imperfect knowledge. Actually, much of what we have learned about complexity is valid under a wide range of dynamical laws and initial conditions: concepts like irreversibility, self-organization, and Darwinian variation-and-selection are not very sensitive to the underlying microscopic physics. So studying complexity requires more than traditional physics, calling on fields such as biology and computer science (see Badii and Politi 1997). What exact history is realized in a universe does, of course, depend on microscopic details. But just obtaining local pockets of specified complexity is not too difficult. When a variety of dynamical laws can generate complexity from random initial conditions, it is quite a leap to conclude there must be an intelligence behind it all.

Modern physics provides even less of a peg to hang ID upon. With general relativity, random boundary conditions are no longer tucked away in the distant past; a black hole is as much a source of true randomness as the Big Bang (Hawking, in Hawking and Penrose 1996). And quantum mechanics is notorious for its pervasive dynamic randomness. Randomness also makes physical systems haphazardly explore their possible states, leading to irreversibility. And now, it makes no sense to speak of predetermined order. Random data is patternless (Chaitin 1987), so no cause behind it can be inferred; certainly not intelligent design.

Enter Artificial Intelligence

Our physical world is a realm of accidents, of seething, mindless dynamism-the unpredictable twists and turns of history. Yet expecting a combination of laws and chance, however elaborate, to be genuinely creative may be too much. ID, after all, is not just an exercise in information theory; it also draws upon deep-seated intuitions that machines cannot display creative intelligence. Without some account of the place of intelligence within nature, it is still possible to suspect naturalistic explanations of complexity overreach.

Many a science fiction tale tells how a hero defeats a computer by posing a problem it was not programmed to deal with. It then starts saying “does not compute!” in a synthetic yet anxious voice, and finally goes up in smoke. Unlike the rule-bound machine, however, we think human intelligence at its best is flexible, innovative. We confront situations beyond what we have prepared for, and if we do not always succeed, we still often come up with novel approaches to the problem.

As Dembski’s argument that information is conserved makes clear, it is difficult to see how new content can be generated mechanically. Artificial intelligence (AI) researchers ask us to imagine machines that perform a variety of complicated tasks, learning about and responding to their environment in sophisticated ways. But if these machines remain within the bounds of their programming, it is natural to attribute intelligence not to them but to their designers. ID voices this suspicion: that no pre-programmed device can be truly intelligent, that intelligence is irreducible to natural processes.

Such intuitions underlie not only ID but some respectable criticisms of AI, including those based on Gödel’s incompleteness theorem. This has recently been championed by Roger Penrose, the eminent physicist (1989, 1994); Gšdel’s theorem is attractive because it reveals how any rule-bound system has blind spots because it is unable to step outside of a pre-defined framework. And though Dembski considers Penrose to be insufficiently anti-naturalistic, ID requires at least some such critique of AI to be sound.

It turns out, however, that all that is needed to add the required flexibility to a machine is to let it make use of randomness. A random function, because it is patternless, can be used to break out of any pre-defined framework. It serves as a novelty-generator. Plus we can prove a “completeness theorem” showing all functions can be expressed as a combination of rules and randomness. So if all we claim is that humans are flexible in a way not captured by rules, randomness alone does the trick. There is no other option (Edis 1998b).

Now we need to use randomness for actual creativity. And we already know an excellent mechanism for putting bare novelty to work: natural selection. Dembski’s claim that randomness does not help create content is incorrect; a Darwinian process is different from altering a message through fixed selection criteria. Everything is subject to random modification-there are no predetermined criteria; nothing but mindless replication and retention of successful variants.

A fuller understanding of something as convoluted as human creativity is a long way off. But fundamental objections like those ID raises have largely been overcome. It is almost certain that randomness and Darwinian processes are vital in the workings of our brains. So our current sciences of the mind are full of ideas like neural Darwinism, Darwin machines, memes, and multiple levels of Darwinian mechanisms depending on competing processes to assemble our stream of consciousness (Dennett 1995). Variation-and-selection, today, is beginning to be vital for theories of mind as well as biology.

A Darwin Detector

What, then, are we to make of ID? It now seems like a bad argument, concocted of pointless complaints against evolution on one hand, and flawed intuitions about information and intelligence on the other. Discarding ID, however, would be hasty. Important theories about the world convince us by ruling out serious alternatives. Historically, evolution took shape against then-compelling notions of design. ID may be wrong, but it is also a decent update of Paley with a real intellectual appeal. Its errors provide a useful contrast, highlighting what is correct in evolution.

Confronting the information-based arguments of ID is especially helpful in revealing how profound an idea evolution is. As ID proponents suspect, Darwinian thinking is not confined to biology; it anchors a naturalistic understanding of all complex order, even including our own intelligence. Hence today, Darwinism is central to a thoroughly naturalistic picture of our world.

So in defending their religious views, ID proponents pick the correct target. They are also right to emphasize how designed artifacts and living things are similar. And Dembski’s criteria of contingency, complexity, and specification do reveal a special kind of order they share. The irony is, what these criteria actually detect is that there were Darwinian processes at work. The complexity of life is directly produced through evolution, but an artifact also is an indirect product of the variation-and-selection processes that must be a part of creative intelligence.

Defenders of evolution can now allow themselves a wry smile. Intelligent Design is as close to respectable as anti-evolution intuitions are likely to get, and Dembski has made a good stab at making ID rigorous. And what we end up with is a Darwin detector.

Note

  1. Dembski’s work has been criticized (Fitelson et al. 1999), but these objections do not seem fatal. In any case, Dembski’s criteria are not signs of design as he understands it, even if we were to ignore all such criticism.

References

Taner Edis

Taner Edis is an associate professor of physics at Truman State University. His latest book is An Illusion of Harmony: Science and Religion in Islam (Prometheus Books, 2007).