My thesis: the answer is “yes, but who cares?”
But first, let me give a few definitions. In this context, “religion” will be defined as any kind of belief in an uncreated intelligence (“God”). Since most people are not explicit about the probabilities they assign to beliefs, and because often — especially when it comes to religious beliefs — they descend into belief in belief — it is hard to quantify it, and there will certainly be cases of “religious” people who do not qualify. It is for this reason that I am explicitly stating a definition in this post.
So let’s try again:
Are beliefs in an uncreated intelligence (with high probability, and accounting for belief-in-belief) compatible with science?
That will serve as unpacking the first part. Now we unpack the second part. Science is defined as using the scientific method to understand the universe:
- Devise a theory
- Generate predictions (hypothesis) from the theory
- Design experiments to test those predictions
- Accept theories which are “elegant” and conform to the experiments
There are many nuances here — experiments must be carefully designed, and confounding factors must be accounted for. This is to be expected — if learning how to be a scientist was as simple as reading four lines in a blog post, we would not need PhD programs.
So let’s try again:
Are beliefs in uncreated intelligence (caveats included) compatible with using the scientific method to explain the world?
Well, sure they are. After all, dab a little bit of rationalization on the belief, and you can make it compatible with anything. But wait — where does the intuition that there is a conflict come from? Well, at first rate, the scientific method seems like a ritual. You follow the ritual, the social convention, and science comes out. Science turns out to be useful in manufacturing technology. The world seems to behave in the following way:
Science allows us to have power to manipulate the world to our liking
This is a fact, as far as we can see. Can we turn the method of science on itself? Can we generate predictions, and test them? Classically, the philosophy of science was a distinct tradition of science apologetics, distinct from science. That is, the conclusion was forgone — science is good — and arguments in favor of science were sought. Popper’s falsificationism and Kuhn’s paradigm are both of that ilk — they’re not science!
Properly understanding science requires several more auxiliary facts:
- We know humans find it hard to change their minds
- Human beings are not Bayesian reasoners
- Reality seems to not care what we think about it
So here is a theory: science is a set of social conventions to allow imperfect human beings to approximate, collectively, Bayesian reasoning about the world. Of course, that Bayesian reasoning has to start from some priors. Science has always espoused Occam’s Razor — which sounds suspiciously like an a-priori way to distinguish between theories. Can we formalize Occam’s Razor into a prior? It turns out we can! It’s called Minimum message length and has its own wikipedia page.
Let’s restate the theory again:
Science is a set of social conventions to allow imperfect human beings to approximate, collectively, Bayesian reasoning with MML priors about the world
Here is a concrete prediction:
Other philosophies of science will justify approximations to Bayesian/MML reasoning
Well, do they?
- Falsificationism — if P(theory is true|current knowledge) is close to 1, experiments that confirm it can help very little. On the other hand, if an experiment “falsifies” it, meaning P(theory is true|current knowledge+experiment)<<1, then that’s “enough” to drop it. Contrariwise, a theory that cannot be falsified, meaning P(theory|experiment succeeds)=P(theory|experiment fails) means, P(experiment succeeds|theory)=P(experiment fails|theory), so the theory says nothing about any experiment. A theory with no predictions is longer than the MML “anything can happen”, and so with a lower probability and no positive evidence, it is not accepted.
- Paradigm shift — if P(theory is true|current knowledge+experiment)<<1, but it’s just one experiment, a message that corresponds to “theory is true except in that one experiment” is not much longer, so it’s likely that P(theory modified with accounting for that one experiment) will be close to 1. This grows worse and worse the more exceptions are added, until a “paradigm shift” occurs, and a new theory is selected for. Given that in Falsificationism there’s no positive response except “back to the drawing board”, it is only to be expected that a bounded intelligence will prefer searching for close-by messages if we can find ones with P(message) close to 1.
Now, let’s assume we accept that theory. We have
Science is a social approximation implemented on human beings to Bayesian/MML reasoning
It is perfectly possible to do science without understanding that. But a scientist who reflects on the scientific method is likely to come to similar conclusions, and then ask themselves the obvious question: “If Bayes/MML turns out to be good at explaining the world, why not apply it everywhere?”
That’s where the problem with religion comes in. The MML for “God” seems, on the face of it, to be really really long. Adding insult to injury, “God” has poor predictive strength. The two of those combine to make sure that P(“God”|what we know), as far as we can roughly approximate it, is close to 0. This is what people who are aware of Bayesian reasoning mean when they say “The statement ‘God exists’, as far as I can see, is false.”
In conclusion: Science is compatible with religion — but only so long as we do not reflect on science. As soon as we reflect on science, and apply it to itself, the distance from there to rejecting religion is short.
This work is licensed under a Creative Commons Attribution 3.0 Unported License.