### Removing Explanatory [Content](Content.md) HM starts with the question: "If I look at a physical system, how would I know whether a simulation has been encoded in it?". He agrees that a simulation is an abstract set of rules and entities that can be instantiated in a physical system. Crucially, the abstract structure matters, not the substrate it’s running on. But this creates a *problem*: how do we get the abstract rules instantiated in the physical system in the first place? Traditionally, this is what encoding is for. Encoding is the process that physically instantiates the abstract rules (simulation) into a physical system. It implies a structured, causal process, as well as a degree of isomorphism between the abstract system and physical system. In this way we were using the word encoding to *stand for a process that solved a problem*—namely the problem of how do we instantiate an abstract system into a physical system. Now HM flips the question around and says "rather than asking how a simulation was encoded into a physical substrate, if we can decode one then we know it had been encoded". This may sound innocuous at first, but it actually shifts the entire explanatory burden from encoding to decoding—decoding is now the [Load-Bearing Structure](Load-Bearing%20Structure.md) . This would be fine if he defined decoding so as to solve the problem that encoding originally solved. But he retreats from explanation and allows any mathematically viable decoding—the look up table. This move HM made *removes explanatory content*. Encoding once referred to the process which explained how an abstract system was instantiated in a physical substrate. He has removed explanatory content by: * Eliminating empirical consequences (you can no longer say which systems are/aren't valid encodings based on empirical structure) * Abandoned causal nature of encoding * Made the claim unfalsifiable So instead of answering how abstract systems are implemented in physical ones, the theory now says: "Anything can be interpreted as anything". This means the theory says nothing. This relates to content as follows: The original concept of encoding had content. It implied: * Certain physical instantiations qualify as representing abstract rules * Others do not (lack of isomorphism, no causality of the encoding process) * We could observe and evaluate whether those conditions were met But, under HM's reformulation, the constraints above disappear. The empirical consequences disappear. Anything can count as a simulation, making the idea meaningless. This is a textbook case of removing explanatory content: loosen the definition so much that it no longer limits or predicts anything. * Encoding originally offered a way to explain how abstract systems are physically implemented—via structured, causal, and observable processes. * Moravec shifts the burden to decoding, then allows any mathematically possible decoding. * This redefinition removes explanatory content by eliminating the theory’s consequences and making it trivially universal. * We’re left with a statement that applies to everything—and therefore explains nothing. To summarize: Encoding solved a problem—namely, it explained and referred to a process constraining how abstract rules get instantiated in a physical substrate. Specifically, via a structured, causal, observable process that lead to an isomorphism. In this way encoding had *consequences*, and therefore *content*. HM then destroys explanation by shifting the burden to decoding, and then never addressing it. Thus, we are left with more problems and questions—ones that our original explanation solved. By doing this HM *chopped off consequences*. Note: he absolutely chopped off consequences that are implied be encoding. Normally, encoding can be used to constrain what counts as an encoding—it will rule some systems in and others out, providing explanatory discrimination. Based on his finagling, he is now saying we must be view *anything* as having been encoded (due to the lookup table). At this point he *has* retreated from explanation—but to really nail him, we need to press further. Where did the look up table come from? If we had reason to believe that the universe we live in hands out lookup tables for free, maybe this would be a fine argument. However, we must create the lookup table! And that would require running the simulation. ### Danger: Removing Empirical Content for Empirical Processes A danger to watch out for is when you are dealing with an empirical, observable process—such as encoding an abstract system in a physical process—and you are removing empirical content (the set of all consequences that could falsify your theory). This is what HM is doing. ### Lookup Table: Description vs Explanation Consider two ways of thinking about encoding. 1. Explanatory: This physical substrate has an abstract system encoded in it because there was a structured, causal process that instantiated it there, in a way that retains an isomorphism between the abstract and physical system 2. Descriptive: This physical substrate has an abstract system encoded in it because there is a lookup table mapping all states of the physical substrate to the abstract system (1) has far more content. We can check to see if the instantiated preserved isomorphism, if it was causal, and so on. On the other hand, (2) has low empirical content. There is almost nothing that could rule it out—lookup tables are easy to imagine. By moving to (2), HM has reduced content. He has also lost out on the explanation—we no longer understand the why or how. Why does the physical substrate correspond to the abstract system (originally it was because there was an isomorphism)? How does the physical substrate instantiate the abstract system (originally due to a causal, structured process)? Additionally, we have a new question: how is his lookup table generated? So he has removed content (we can't falsify him), and lost out on the why and how (as well as added a new how question). # Todos %%TODO: Address how HM is removing explanatory content (this requires addressing what content is, empirical content, consequences, maybe even description vs explanation). See notes in [Content](Content.md)%% **Descent From Explanation: Removing Explanatory Content Means Removing Consequences** * There are two forms this can take: 1) You replace a good explanation with a worse one, and 2) You remove explanatory content altogether, *carving off consequences* so the theory no longer implies anything checkable * Explanatory content means something that is _hard to vary_ and has _many logical, testable consequences_. Theories rich in explanatory power make bold predictions—ones that can be criticized, falsified, refined. * Consider the example: "_The planets move in elliptical orbits because the gods did it._" — What specific consequences follow from that? None. It doesn’t generate predictions. You can’t test it. * In contrast, Newtonian gravity allows you to calculate where a planet should be at any given time. It has vast, detailed consequences that are logically implied and empirically testable. * To descend from explanation, then, is to start replacing nodes in your argument with elements that reduce or eliminate consequence. You _weaken_ the structure—either by vague definitions or by overgeneralizing to the point where the theory no longer says anything concrete. * This is what Moravec appears to be doing. He redefines concepts like “decoding” in a way that removes their constraint—making them so broad that they apply to everything, and thus explain nothing. * The danger here is subtle. It’s not that he’s offering incorrect conclusions—it’s that his framework has no consequences. It lacks the falsifiability, the testable implications, the structure that makes a theory useful. * Think of the theory of computation: it’s broad, but it’s powerful precisely because it’s specific. It offers a universal claim that’s testable—you can try to find a single counterexample in physical reality to falsify it. That’s what makes it strong. * Compare that with “the gods did it.” That kind of claim offers no constraints, no predictions, and no way to falsify it. There’s no structure it implies that we can test. And that’s the danger in Moravec’s approach: his redefinitions move the argument away from structured explanation toward this consequence-free territory. * Thus, Moravec is *chopping off* explanation. He is redefining terms in a way that removes consequences—this is the opposite of what we should be doing. We want bold theories that are ripe for criticism.