Because we want to be less in error rather than more in error, we must employ the most reliable method we know to distinguish true claims from false. We therefore need a means to distinguish reliable methods from unreliable methods, and more reliable methods from less.
Naturalists start with a plausible premise: a method that leads significantly more often than not to the discovery of genuinely true and false propositions will exhibit two particular features, which an inaccurate method will not exhibit: predictive success and convergent accumulation of consistent results. We can even expect that a more accurate method will exhibit these features more often than a less accurate one. And this is how we test out different methods and choose the best from among them, and throw away the ones we don't need.
Of these two tests, first is predictive success. If we use an inaccurate method we should expect our desires and expectations to be routinely frustrated, as what our trusted propositions predict fails to transpire. This failure, in fact, is what it would mean for those propositions to be false (at least before we start making too many assumptions about the underlying reality), so this conclusion follows necessarily from the very meaning of truth itself. Therefore, if our method is correct, then we can expect it to routinely produce propositions whose predicted experiences do in fact take place (since that is what it means for them to be true, prior to our adopting any metaphysics).
This is especially true for those experiences that would otherwise be a complete surprise. Why should we expect this? Because this sort of result would not likely occur if our trusted propositions were false, but could easily occur if they are true. Either way, a bad method will lead us to conclusions that fail to anticipate the future. In short, its results will fail every real test. A good method, because it succeeds in getting at the truth, must necessarily produce assertions that do successfully anticipate the future, to a degree and with a frequency not at all possible by chance. Thus we can identify a good method when we see one.
The second criterion of good method is convergent accumulation of consistent results. If we use an inaccurate method we should expect that when we investigate a proposition from several angles we will get inconsistent results, and the more propositions we accumulate the more contradictions we would encounter and the more complex our belief-system would have to become to accommodate them. But if our method is correct, we should expect that when we investigate a proposition from several angles we will get the same results, which would be an improbable coincidence if there were no stable truth being hit upon. At the same time, the more propositions we accumulated, the more consistent our system of propositions would become, with researches in various areas all confirming and supporting each other and permitting cumulative advances in practical knowledge. In contrast to a bad method, with a good method we would find ourselves eliminating rather than accumulating contradictions, and our belief-system would become less convoluted.
When we test methods by these two criteria, we end up with four that prove more reliable than all others: the methods of logic, mathematics, science, and professional history and journalism. We therefore use those methods to determine what is true. And their greater overall success on the two criteria of reliable method (predictive success and convergent results) is our standard of truth, guaranteeing that what those methods determine is true, is most probably what is true.
If anyone claims there is a better method, we see which one contradicts what we observe and experience and which one does not, and we reject the one and embrace the other. If neither produces conclusions that contradict what we observe and experience, but still they don't produce the same conclusions, then we see which method requires the most unproven excuses to make its conclusions not contradict what we observe and experience. We can prove as a matter of undeniable logic (using Bayes' Theorem) that the approach that requires the most unproven excuses to explain all the same observations is the least likely to be the more reliable approach to finding the correct explanation of things. Therefore we abandon that method and embrace the other.
Epistemological Endgame
Epistemological Endgame:
http://richardcarrier.blogspot.com/2006/11/epistemological-end-game.html
