They should make risky and narrow predictions
Naval: Brett, would you say that a scientific theory is a subset of a good explanation?
Brett: Yes. They’re the testable kinds of good explanations. Falsifiable theories are actually a dime a dozen. This doesn’t tell you anything about the quality of the explanation you’re being given.
The example that’s used in The Fabric of Reality is the grass cure for the common cold. If someone says, “If you eat 1 kg of grass, it will cure your common cold,” then they have a testable theory. The problem is that no one should test it. Why? Because they haven’t explained the mechanism that would enable grass to cure the common cold. And if you do eat 1 kg of grass and it doesn’t cure your cold, they can turn around and say, “1.1 kg might do it.”
Naval: Right. Or you need a different kind of grass.
Brett: It’s always testable, but you’re not making any progress.
Naval: The second piece of a good explanation is that it’s hard to vary. It has to be very precise, and there has to be a good reason for the precision.
The famous example used in The Beginning of Infinity is the explanation for why we have seasons. There’s the old Greek explanation that it’s driven by Persephone, the goddess of spring, and when she can leave Hades. There was this whole theory involving gods and goddesses. Not only was that not easily testable, it was very easy to vary. Persephone could have been Nike, and Hades could have been Jupiter or Zeus. It’s very easy to vary that explanation without the predictions changing.
Whereas, if you look at the axis tilt theory—which says Earth is angled at 23 degrees relative to the sun and therefore we’d expect the sun to rise here in the winter and over there in the summer—the facts of that are very hard to vary. It makes risky and narrow predictions. The axis tilt theory can predict the exact length of summer and winter at different latitudes, and you can test that precisely.
Beyond it being a creative theory that is testable and falsifiable, it should be hard to vary the pieces of that theory without essentially destroying the theory. And you certainly don’t want to vary it after the fact—like in your grass example, “Oh, it was 1 kg? No, now it’s 1.1, now it’s 1.2.”
Finally, the predictions that it makes should be narrow and precise, and they should be risky. For example, I believe in relativity it was Eddington who did the experiment and showed that starlight gets bent around an eclipse. And that was a prediction that Einstein had made in relativity, which turned out to be true. That was a risky prediction that took a long time to confirm.