Predictive Power: Evidence Versus Tuning

       I’ve heard several people claim that there’s plenty of evidence for God out there, if you’re just willing to look for it and remain open to the possibility. Unfortunately, this “evidence” typically comes in the form of personal experiences that are relatively unpredictable, subjectively interpreted, lacking in independent confirmation and repeatability, and are only made sense of after the fact. It is exactly the kind of thing that hard science does not use as evidence to support theories. This is why, in Open Your Mind, I was careful to outline the process by which solid evidence should be produced, instead of simply asking for evidence.

       The kind of “evidence” that theists keep presenting is used in science, but not as a means of substantiating a theory. When the subjectivity is taken out and independent confirmation is worked in, you have a postdiction. But even then it still doesn’t meet most scientists’standards for conclusive evidence. An excellent example of this in practice is General Relativity (GR). Einstein published this theory in 1915. Many physicists expected great things from GR, and it had already postdicted the unusual orbit of Mercury, but it took several years for GR to gain a “solid empirical foundation,” as Wikipedia puts it. It wasn’t until predictions were made, tested, and verified that scientists were satisfied.

Describing how a belief in God is consistent with or even highly plausible from a particular observation after the fact is not completely convincing to a scientific thinker. At most, it could convince me that the concept of God is a good idea that should be tested, but the tests never seem to come. All I get are more post-hoc explanations. What theists don’t seem to realize is that these sort of delayed analyses are supposed to be used to tune a theory, not to verify it.

The problem is similar to the problems incurred by unfalsifiability. A fairly good description of hard scientific evidence is an event that could have falsified a theory but did not. If your theory hasn’t committed to a conclusion ahead of time, then the “evidence” in question does not serve as a potential falsification of the theory. Put another way, most of the after the fact explanations I hear from theists are tailored to the event. If the event had gone differently, the explanations would have been tailored differently. Thus the event cited isn’t being used to support the theory, it’s being used to fit the theory. To generate the theory. It isn’t evidence, it’s tuning.

The problem with trying to support a theory in this way is that it leads to theories that are extremely adaptive. This may sound like a good thing, but it actually spells doom. If you keep bending your theory to fit event after event after event without any genuine testing, you’ll end up with a theory so flexible it can fit most any data. At this point, your theory will probably have become profoundly vague. Even worse, you can do this multiple times and generate many, many flexible theories. If Islam and Christianity and karma and dogma and reincarnation and Satanism and Wiccan Voodoo can all be flexed to fit the known facts, then “fits the known facts” isn’t good enough. It isn’t a good selection criteria because it’s not selective. In order to show that your theory is better than the others, you must have something the others don’t. You need predictive power.

Not any old predictions will do. You need to avoid making your predictions profoundly vague. Your predictions must be specific enough that you can get some use out of them. And you need to establish success/failure criteria before testing your prediction. Otherwise you’ll just end up tuning again. And your predictions must be consistently accurate. Well above the standard deviation of some randomized method like drawing predictions out of a hat. And you need independent verification to demonstrate convergence and objectivity. All of this is crucial to demonstrating that your idea has more merit than the competition. And all of it is completely missing from apologetics. Any theory can get a series of ad-hoc patches to match observations after the fact. Only the best of the best are able to tell you the observations before you even look.

It’s hard to do this, because you will eventually have to face up to the fact that you are wrong (and you will be wrong at some point). But the payoff is HUGE. The electric motor would never have been built if we hadn’t first built up the expectation that electric coils carrying current will push magnets. And we would never have had that expectation if our ideas didn’t predict such an event would occur reliably and consistently. It’s predictions – reliable, specific predictions – that allow us to build technology. And not just technology. If I want to develop a good business, I am much better off with a system that predicts which events will turn out to make or lose money than with a system that is only able to “explain” why events made or lost money after the fact. (And s it even an explanation if dozens of conflicting theories meet the same criteria?) There are a wide variety of ways that predictive power can provide incredible benefits. But you have to be willing to commit to predictions in order to get there, and that means subjecting your theory to possible falsification.

So if you are confident in your beliefs, so confident that you think you and others should base their lives around these beliefs, if you really believe that they are the best of the best of the best, then put them to the test. Don’t hide behind unfalsifiability. Don’t get stuck tuning indefinitely. Don’t be content with after the fact patch-ups. Anyone can do that, Muslims, Christians, Pastafarians, Jedi, Scientologists. If you really think your theology is better than the competition than show it. Do what they cannot. Show that your theology has genuine predictive power, and you will convert more free-thinkers than all the fliers in all the world.