# Falsification and False Negatives Consider the link to falsification: No matter how many white swans I observe, I cannot conclude all swans are white. However, a single black swan can rule out that theory. Now, it *could be* that all swans are indeed white—making that a true statement. In that case, our not classifying it as *true* could be a considered a *false negative*. However, consider now that we had an explanation for why all swans were white. We could argue that all swans are indeed white and tentatively accept that conclusion. If we have a good explanation it may be entirely reasonable to proceed this way. Note that the way we were able to proceed and move forward was via a good explanation. We were not able to move forward by appealing to the number of white swans we have seen. They are corroborating evidence, certainly, but without an explanation we can't confidently say anything about the lack of black swans. We have observation, description—but no explanation. HM is afraid of false negatives. He does not want us to rule out possible decoding mechanisms, for it exposes us to having created a false negative. So he proposes we simply accept *all possible* decoding schemes. Why? Because we have evidence in the past that certain things were not able to be decoded until we found the correct scheme. This is an inductive argument (trying to generalize from past observations) being applied because HM is afraid of the shortcomings of induction. He is: * Abandoning criticism in favor of blanket acceptance! He no longer has a means of criticism * There is no comparison of explanations—all are treated as equally viable * It turns fear of error into acceptance of nonsense * Instead of taking an approach where one can correct against errors as time progresses forward, his approach ensures no error correction can ever take place, for he refuses to eliminate any decoding schemes! This is an over correction against possible error An interpretation is an explanation We can put this another way: HM has the symmetry *backwards*. ### How Science Actually Works Let us consider how science actually works. We do not start by saying "all theories are equally valid until they have been ruled out". Rather, we only consider theories that are good explanations. Given a theory that is a good explanation, we then look at it's *consequences* and attempt to criticize it. One form of criticism is falsification via empirical observation—if our theory predicts that $A$ will happen and $B$ will no, but we observe that $B$ happened and $A$ did not, then we can rule out our theory. But if we observe $A$ that does not make us more confident in our theory so to speak. It just means that it has been more effectively *corroborated*. Perhaps our theory had a rival and some evidence did not corroborate the rival. This leads us to prefer our theory—we accept it, tentatively of course. For example, we don’t seriously consider ridiculous theories like “eating one kilogram of grass causes consciousness.” Why? Because it’s not a good explanation. We don’t need to perform experiments to rule that out—it’s never even on the table because it doesn’t explain anything. In real science, we don’t just accept all logically possible theories until we have evidence against them. We start with good explanations and then use observation to rule them out when their predictions fail. That’s how progress works. Note that this approach will clearly have false negatives all the time! For example, consider the class of all possible theories (of which we know only an infinitesimal number of them currently). Based on the way that science actually works, we do not consider all to be equally valid until they have been ruled out. Rather, we only consider those that are good explanations. This means that many true theories in that set will be "ruled out" (deemed to not be the best explanation) until we have determined that they actually are good explanations. Note: this doesn't even account for what it means to have a set of objects of which you don't know some of them. It is nearly incoherent to say: "Ah well let's just accept all logically possible theories as equally valid until we have ruled them out". In general to construct a set you must have a way of *creating it*—but how could we do that for explanations, of which we may not even have the necessary knowledge. ### What HM Is Doing HM is wanting to accept all logically possible decoding schemes in order to avoid false negatives—e.g. ruling out an interpretation (decoding mechanism) that is actually valid but has just not been discovered yet. In science that would look like: "well, we have seen that theories have been proven wrong before. So the safest bet is to allow all theories that don't conflict with logic". In science this doesn't work because there are methods of error correction and criticism. We could say: "well this logically possible theory is ruled out many times by observations—thus to accept it would be introduce an inconsistency, a conflict". Also note that this is an incredibly poor choice from a practical sense. Yes HM will avoid ruling out any true theories and thus any false negatives, however he is effectively extending his class of false positives to be infinite. This means that any theories that are actually good explanations will be overwhelmed and lost in the sea of garbage (but still logically possible) theories. He does not realize that his process is far more analogous to science than meets the eye. He’s using an inductive generalization from past misses to justify accepting literally everything, even stuff that makes no explanatory sense. That’s like saying, “Since we missed General Relativity for a long time, let’s just accept every possible physics theory right now so we don’t miss anything ever again.” That’s not how science works. We don’t accept everything. We accept the best explanations we have, and we treat them as tentative, always open to improvement. Accepting every possible theory or decoding isn’t being open-minded—it’s abandoning the very idea of explanation. Note that in science, just because we have constrained our space to search based on current known physical laws (e.g. the speed limit imposed by light), does not mean that no one looks *outside* those limits. You can look anywhere you'd like. His mistake is not in saying that logical possibility describes an absolute bound—it does when we seek physical laws as well. But rather in saying that we should accept all logically possible interpretations (theories). Moravec tries to dodge this by retreating to logical possibility—he argues that since what’s physically possible changes with our knowledge, we should anchor ourselves in what’s logically possible. But that just shifts the problem. Logical possibility is so broad and unconstrained that it collapses any standard for criticism. It opens the door to accepting all sorts of nonsense, just to avoid the risk of ruling something out too early. But that’s not how progress works. We don’t make progress by avoiding mistakes at all costs. We make progress by making bold conjectures, subjecting them to criticism, and moving forward based on the best explanations available, all the while knowing we might improve or replace them later. Notice the absurdity of retreating to the logically possible—many things are logically possible that we would agree are entirely absurd.