The following is one of my comments in a debate at Tom Gilson's Thinking Christian blog about the Problem of Evil.
To give you a sense of context, it has been argued up to this point in the debate that there is confirmation of God's goodness in certain events (e.g., people who come out of adversity stronger than they went in), but that perceived injustices (e.g., the Holocaust) should not be regarded as falsifying the claim that God is good because we simply lack the ability (omniscience) to know that such injustices aren't for the best.
In my response, I show that consistency and logic demand we evaluate such questions using a scientific method of local testing and prediction.
Assume good and evil refer to our standard intuitions on morality, from our limited, local perspective. For example, we all agree that the Holocaust was evil, and that emergency rescue services are good. This is analogous to our ability to measure conservation of energy, locally. The meaning of good and evil are the relatively clear and familiar ones.
We can then propose that there are global rules about these concepts. For a physics example, I might propose that the apparent increase in energy of my space probe as it slingshots around Jupiter is exactly balanced by a corresponding loss of Jupiter's orbital energy. If I widen my consideration of experimental observations to incorporate Jupiter and the Sun, I can verify that energy is conserved.
By analogy, we might suppose that there is universal justice, and though (like the energy of the space probe) we see an imbalance of justice in the world, we can theorize that, if we could widen our consideration of the facts, there would be global (universal) justice.
Unfortunately, we could do the same with conservation of ice cream. We could propose that an ice cream cone eaten on Earth is balanced by an ice cream cone popping into existence on the other side of the universe.
So what gives us confidence in conservation of energy, but no confidence in conservation of ice cream?
Our ability to test the proposition locally. If conservation of energy were false, it could have been recognized as such in numerous experiments. We have tested the proposition locally, and that gives us our confidence. BTW, notice that a prediction is required for there to be a test.
In contrast, propositions about God's goodness, universal justice and universally conserved ice cream cones are not locally testable as stated. (In fact, the same would even be true of the claim that energy is conserved everywhere in the universe if we did not limit our claim to observable parts of the universe.)
So, if we are to be consistent, we either have to (Type 1) reject the claims of universal justice or the goodness of God if there are no local tests of such claims, or (Type 2) admit all sorts of similar claims about conservation of ice cream cones and the like.
Alas (er, I mean fortunately) Type 2 evaluations have an even deeper flaw.
Under Type 2 rules, you can describe what experiences constitute validation of your proposition, but you cannot describe what experiences would constitute falsification. For example, you point to Solzhenitzin as a confirming case where an evil was redeemed somehow. If I point to some nameless Holocaust victim, you will answer that my example is not falsification because I just don't have all the facts. Initially, you rest assured that your proposition cannot ever be found to be false.
However, if we stick to these evaluative rules, the negation of the proposition ("there is no universal justice" or "God is not good") is as persuasive as the original proposition. We can find confirmations of the claim in observed injustices, and we can reject local justices as anomalies caused by our limited perspective. So, by Type 2 rules, the negation of your proposition cannot be found false either.
We must conclude that if we are going to accept the confirmation but not the falsification of a proposition, then that proposition has no truth value at all, because neither the proposition nor its negation can be false.
This is why local testing and prediction (aka Science) are necessary for knowledge.