Author’s note: In a recent interview (June 2026) between cosmologist Brian Keating, a lovable ID skeptic, and Stephen Meyer, the topic came up of a naturally occurring nuclear reactor in Oklo, Gabon, Africa. Keating raised this as a counterexample to design inferential reasoning, suggesting that such a nuclear reactor exhibits the sort of complexity that under other circumstances would trigger a design inference; and yet the reactor seems to be completely natural, having operated 1.7 billion years ago, and thus without any human or otherwise evident intelligence capable of forming such a nuclear reaction. Here is the relevant transcribed dialogue from the interview (starting around the 9:30 minute mark). I’m mentioned in it, but what is unmentioned is that I had addressed this very example in my 2002 book No Free Lunch (pp. 26–27). After giving the dialogue between Keating and Meyer on the Oklo reactor, I include the relevant portion from my book.
The Oklo Trap: A Natural Nuclear Reactor with No Designer (~09:38–11:56)
Brian Keating:
Certainly there are natural fusion reactors that exist, right? There’s one powering at least San Diego. It’s up in the sky. It never takes a break. It never goes on strike and it doesn’t ask to defend its PhD thesis, you know, right when you need it most. So, obviously there are natural fusion reactors. But if I told you that we discovered a fission reactor, which is much more complicated in a certain sense — that we don’t know of effective ways that randomly things can come about — but you know, if I told you we had some uranium somewhere and there was a functioning nuclear reactor that had no human input, there was absolutely no mind in place. Would that be a challenge to at least this branch of the argument from the existence of information originating from a mind? With the existence of a sophisticated neutron-moderated fission reactor that’s self-sustained with slow neutrons for a very long period of time without any human design, without any human intervention. Would that present you a challenge?
Stephen Meyer:
Before jumping into a carefully laid trap like that, I’d like to know what’s the factual…
Brian Keating:
You got me, Steve. I just like you so much and it’s so fun to talk to you because you do have integrity. Again, we don’t agree on a lot of things, but it’s very fun to talk to you because you have a very fine mind. So, yes, you caught the trap. You saw it carefully laid by the devilish Keating. And in Africa a billion years ago there was something called the Oklo natural nuclear reactor and it was a sustained slow-neutron-moderated fission reactor which operated without any, obviously, human intervention a billion years ago and it lasted for millions of years and it was producing heat and light presumably and you could have extracted energy from it and it was all due to changing levels of the water table and this natural material that was causing the… of natural factors that it was, and completely natural. But if you hadn’t known that, you know, I guess the question is if I didn’t tell you that or if we didn’t know about it and then tomorrow I discover it, then the question still stands. I think like we see things that are complex we can’t understand without a mind. But who’s to say that the eventual evidence in a Bayesian sense won’t come and or it hasn’t come already and we just didn’t understand it at the time?
Where the Design Inference Becomes Decisive and Where It Doesn’t (~11:56–15:50)
Stephen Meyer:
What’s interesting about that example is that you have a relatively complex convergence of factors, right, that are producing a functional outcome. So it is a kind of finely tuned system that is producing a functional outcome. And I guess our response to that would be to get a little deeper into the apparatus of design detection. William Dembski, our lead mathematician, has written extensively on this. He would want to know what are the relevant probabilistic resources that are available to generate that convergence of factors. You can suspect design at many levels of complexity. There’s a point past which the inference becomes decisive where the level of complexity is such that it exceeds the probabilistic resources of the naturalistic systems that are relevant. In this case, clearly there are naturalistic resources that were sufficient to produce the convergence of factors that made this kind of functional outcome.
I did run the math on this in my book Signature in the Cell, looking at not a naturally occurring nuclear reactor but the complexity of a single protein fold, and if you take a 13.8-billion-year universe. And first you run the numbers on the complexity of the protein, the degrees of freedom that allow you to calculate the brute probability of getting that protein on a one-off. And it’s infinitesimally extremely small. But you have to take into account, to determine whether or not this could be explained by chance or natural processes, you have to take into account how many opportunities there were for that extremely improbable event to take place. So you run the numbers, and if you say, well, if every event in the history of the universe had been devoted to looking for or generating a unique combination of amino acids in sequence connected by some kind of a bond, and you define an event minimally as an interaction between elementary particles. Turns out that if every event in the history of the universe had been devoted to generating a protein — an amino acid sequence of, in this case we calculated on a basis of 150 amino acids — you would search but a tiny, tiny fraction of the total. So the probabilistic cutoff on that is that it’s much more likely than not that chance would fail than succeed. It’s overwhelmingly more likely that chance or natural processes would fail than they would succeed. Therefore, since we’re looking for a best explanation, it’s more likely that it would fail than succeed; it’s more likely that it didn’t happen that way than that it did. And therefore, positing the chance interaction of natural processes is a much less likely explanation for the information content of the protein than design is.
My likely response to the hypothetical, or what’s a real example, would be to look at, well, what are the conditions that were necessary to make this happen? Do they exceed the probabilistic resources of the entire universe? Clearly in this case they don’t, because it happened within a natural system on Earth. There are features of living systems, of information systems that we develop, that clearly do exceed the probabilistic resources of the universe, and therefore there is a sound basis for inferring to design in those cases, but not all cases where you have complexity and function.
Brian Keating:
Listen to what Stephen just did. Stephen just did something very high in terms of integrity, which a lot of people on either side might not do.
From No Free Lunch
William Dembski:
Here, then, is the relevant portion from my book (pp. 26–27). Note that reference to the complexity-specification criterion is another way of talking about specified complexity and thus about the two features, specification and complexity, that within my design-inferential method triangulate reliably on intelligent causation. Note also that Meyer’s response to Keating was on target, though as the subject of an ambush by Keating, Meyer didn’t have the luxury of developing his answer as I did below.
Is there a convincing application of the complexity-specification criterion that employs this universal probability bound [i.e., 1 in 10^150] and in which a coincidence better explained by undirected natural causes gets attributed to design? Perhaps the best candidate is the natural nuclear reactors at the Oklo uranium mine in Gabon, West Africa. Producing a nuclear chain reaction these days is not simple. It requires getting the right proportion of uranium 235 and uranium 238 and placing sufficient quantities in a controlled environment. Indeed, all the nuclear chain reactions on earth these days result from human design. Nonetheless, two billion years ago nature produced its own nuclear chain reaction at the Oklo mine. What’s more, it was a controlled reaction, with water facilitating the reaction and continually cooling the uranium to prevent an explosion—just as in today’s power plants.
Many highly specific and seemingly unlikely conditions had to be satisfied for the Oklo reactors to form. For instance, because the half-life of uranium 235 is about six times as short as that of uranium 238, the proportion of uranium 235 to uranium 238 on earth is nowadays much less than required for the enriched uranium that powers artificial nuclear reactors. Nevertheless, two billion years ago, the proportions would have been just right, and thus a necessary condition for nature to produce nuclear chain reactions would have been satisfied back then. Likewise other highly specific conditions had to be satisfied for Oklo to produce natural nuclear chain reactions. The Oklo reactors are remarkable. To date no other natural nuclear reactors have been identified.
What are we to make of the Oklo reactors? My own view is that although the conditions these natural nuclear reactors had to satisfy were highly specific, they were not that improbable. For instance, with respect to the proportions of uranium 235 and 238, given their differing decay rates, there was bound to come a time when the proportions would be ideal for a nuclear chain reaction. That the time should be two billion years ago, coinciding with the Oklo reactors, is therefore not the sort of coincidence that triggers a design inference. Rather, it lands on the necessity node of the Explanatory Filter. Other conditions that needed to be satisfied, even when considered jointly, do not appear to trigger a design inference either. Although these conditions are specified, they do not appear that improbable. The precise probabilities for such conditions have yet to be ascertained. Consequently, it is not possible at this time to decide whether the Oklo reactors satisfy the complexity-specification criterion. My own very strong suspicion, however, is that should such probabilities be ascertained, the complexity-specification criterion would not be satisfied.
But suppose the Oklo reactors ended up satisfying this criterion after all. Would this vitiate the complexity-specification criterion? Not at all. At worst it would indicate that certain naturally occurring events or objects that we initially expected to involve no design actually do involve design. This would no doubt seem counterintuitive in light of the naturalism that currently dominates science, but it would not provide a counterexample to the complexity-specification criterion. Rather, it would constitute a borderline case, one that without the complexity-specification criterion would be consigned to the category of unexplained coincidences, but with it would be attributed to design. If design is as pervasive in nature as design theorists claim, then we certainly need to find it in such traditional repositories of design like biology, but we should also not be surprised if we find it in unexpected places.
Cross-posted from Bill Dembski on Substack.









































