# Moravec's Argument Let me start by restating Moravec's argument as clearly and strongly as possible. He constructs it as follows. First, he describes that a simulation is defined by a set of abstract, intrinsic rules and entities that must then be instantiated and executed in some physical system. In other words, simulations are *substrate independent*, but require a substrate to run: > [!quote] > Inside the simulation events unfold according to the strict logic of the *program*, which defines the “*laws of physics*” of the simulation...The simulation’s *internal relationships* would be the same if the program were running correctly on any of an endless variety of possible computers, slowly, quickly, intermittently, or even backwards and forwards in time, with the data stored as charges on chips, marks on a tape, or pulses in a delay line, with the simulation’s numbers represented in binary, decimal, or Roman numerals, compactly or spread widely across the machine. There is no limit, in principle, on how indirect the relationship between simulation and simulated can be. He then identifies that this creates a problem: given some set of abstract rules how does one instantiate them in a physical substrate, and once instantiated, how does one interpret them? These two problems are addressed by *encoding* and *decoding* respectively. He never explicitly defines encoding, but we can roughly make out that it is the process of instantiating the abstract rules and entities that govern a simulation in some physical substrate. As long as the intrinsic, abstract rules are followed, the encoding process can be arbitrarily complex: > [!quote] Just as a literary description of a place can exist in different languages, phrasings, printing styles, and physical media, a simulation of a world can be implemented in radically different data structures, processing steps, and hardware. If one interrupts a simulation running on one machine and translates its data and program to carry on in a totally dissimilar computer, the simulation’s intrinsics, including the mental activity of any inhabitants, continue blithely to follow the simulated physical laws. Only observers outside the simulation notice if the new machine runs at a different speed, does its steps in a scrambled order, or requires elaborate translation to make sense of its action. He then defines decoding is then the process of interpreting the physical system in a meaningful way: > [!quote] > Today’s simulations, say of aircraft flight or the weather, are run to provide answers and images. They do so through additional programs that translate the simulation’s internal representations into forms convenient for external human observers...A simulation, say of the weather, can be viewed as a set of numbers being transformed incrementally into other numbers. Most computer simulations have separate viewing programs that interpret the internal numbers into externally meaningful form, say pictures of evolving cloud patterns. At this point it is helpful to orient around the *problem* that Moravec is trying to address: if you look at some arbitrary physical system, how can you identify if it has encoded and is running a simulation? Moravec responds with: > [!quote] > What does it mean for a process to implement, or *encode*, a simulation? Something is palpably an encoding if there is a way of *decoding* or *translating* it into a recognizable form. So in Moravec's view if a physical process can be decoded into a recognizable form, we can say that the physical process had encoded a simulation. This then brings up a new question: what exactly does it mean to decode something? Moravec writes: > [!quote] > As the relationship between the elements inside the simulator and the external representation becomes more complicated, the decoding process may become impractically expensive. Yet there is no obvious cutoff point. A translation that is impractical today may be possible tomorrow given more powerful computers, some yet undiscovered mathematical approach, or perhaps an alien translator...Why not accept all mathematically possible decodings, regardless of present or future practicality? This seems a safe, open-minded approach, but it leads into strange territory. > Given that Moravec will accept any mathematically possible decoding, a natural follow up question arises: what is the most general mathematically possible decoding possible? Moravec argues that it is a lookup table, and that this has startling implications: > [!quote] > A small, fast program to do this makes the interpretation practical. Mathematically, however, the job can also be done by a huge theoretical lookup table that contains an observer’s view for every possible state of the simulation... > > The observation is disturbing because there is always a table that takes any particular situation—for instance, the idle passage of time—into any sequence of views. Not just hard-working computers, but *anything at all can theoretically be viewed as a simulation of any possible world*! > > A simulation, say of the weather, can be viewed as a set of numbers being transformed incrementally into other numbers. Most computer simulations have separate viewing programs that interpret the internal numbers into externally meaningful form, say pictures of evolving cloud patterns. The simulation, however, proceeds with or without such external interpretation. If a simulation’s data representation is transformed, the computer running it steps through an entirely different number sequence, although a correspondingly modified viewing program will produce the same pictures. There is no objective limit to how radical the representation can be, and *any simulation can be found in any sequence*, given the right interpretation. Thus Moravec has arrived at his claim that *anything at all can theoretically be viewed as a simulation of any possible world*. One specific consequence of this is that a rock could be viewed as a simulation of a *conscious mind*: > [!quote] > Perhaps the most unsettling implication of this train of thought is that anything can be interpreted as possessing any abstract property, including consciousness and intelligence. Given the right playbook, the thermal jostling of the atoms in a rock can be seen as the operation of a complex, self-aware mind. How strange. Common sense screams that people have minds and rocks don’t. But interpretations are often ambiguous. > > No particular interpretation is ruled out, but the space of all of them is exponentially larger than the size of individual ones, and we may never encounter more than an infinitesimal fraction. The rock-minds may be forever lost to us in the bogglingly vast sea of mindlessly chaotic rock-interpretations. Yet those rock-minds make complete sense to themselves, and to them it is we who are lost in meaningless chaos. Our own nature, in fact, is defined by the tiny fraction of possible interpretations we can make, and the astronomical number we can’t. And just like that Moravec has argued for a complete overhaul of our world view. Not only are rocks and pinecones and coffee mugs conscious, but any simulated world can be found in any simulated object, and they are all equally real. This is a deeply unsettling proposition. Now you may be thinking "I am unfazed, this argument is clearly nonsense. Rocks don't have minds and simulations don't exist everywhere we look. That doesn't make sense." But *that* is an exceedingly poor counter argument. This was the exact structure of the argument that [The Inquisition made in response to Galileo's Heliocentric Theory](Galileo%20vs%20the%20Inquisition.md). The earth does not *feel* like it is hurtling through space at 66,000 miles per hour, while moving at a speed of 1040 miles per hour around its rotational axis. To say that it does is to make our theory more complex and contradict clear intuition. What grounds could we possibly have for doing that? And yet today we know that Galileo was correct. So clearly arguing against an idea on the grounds that it is not intuitive and contradicts common sense is not an approach we want to lean on. ###### Further Support We can provide even more support to Moravec's argument. He is trying to reason from first principles, from fundamentals, about what we can say in regards to simulations being encoded in physical substrates. Reasoning from first principles means you remove extraneous, parochial constraints and work with only those that are absolutely essential. He sees that if we classify possible decodings only based on our present knowledge, we will certain end up missing some—we will claim that a physical system is not encoding anything when in fact it is, we just didn't have the knowledge to identify it. So he reasons that our level of current knowledge is a temporary, extraneous detail that should not factor in to our encoding classification criteria. Instead, we should be as general as possible so as to not rule out any possible encodings that we may just not be able to identify today. And on this point he is most definitely *correct*. If we try come up with a criteria for what is a valid encoding based on our present knowledge, we will be sure to rule out some valid encodings—let us call this the *false negative problem*. It is simply that we falsely classify certain physical systems as *not* encoding a simulation when in fact they were. ###### Looking At Binary To make this sink home even further, imagine you are sat down at a computer and shown a list of binary numbers. You are asked to identify whether this is encoding the states of a simulation or if it is just random noise. You may rightfully say you have no idea, you don't have enough knowledge of the situation to answer. But now imagine I slip you a piece of paper telling you that I had programmed a chess simulation prior to you entering into the room, and those binary numbers correspond to the board states from a recently simulated game. Now you may say that yes you do believe that those numbers correspond to a simulation—you have new knowledge that allows you to claim this. Or perhaps I pass you a USB stick that has a little decoding program that takes in those binary numbers and produces a graphical representation of the game, displayed on the screen. You again will be able to confidently claim that the binary numbers represented a simulation. In both cases your ability to say anything about the physical process was dependent on your current knowledge. And since Moravec wishes to remove any parochial, extraneous details of our problem, he removes this constraint of knowledge. This is just classic first principles thinking in action, is it not? ###### Godel Numbering and The P-Q System There is yet another beautiful example to make this point even more clear: Godels Incompleteness Theorems. Via a method known as Godel-Numbering, Kurt Godel showed that statements of number theory can be interpreted on two levels: as statements about numbers and as statements about the system of numbers itself. The details of Godel-Numbering are outside the scope of this essay, but we can walk through an analogous example. Consider the following expression: $\text{— — p — — — q — — — — —}$ What might we interpret that to mean? In other words, can we decode this expression into one that we understand? What about the following decoding: $2 + 3 = 5$ We can argue that this is a valid decoding because there is a *correspondence* between the statements. We can interpret the second as having the string '$\text{— —}$ ' corresponding to $2$, the $p$ corresponding to $+$, the $q$ corresponding to $=$, and so on. But is this the *only* interpretation? What about the following: $\text{2} = \text{3} \text{ taken from} \text{ 5}$ It most certainly is! There is yet again a correspondence between the symbols, only now $p$ corresponds to $+$, and $q$ corresponds to $\text{taken from}$. What this shows is that even in mathematics, a single statement can have multiple valid interpretations. ###### Man of Steel Thus we can summarize Moravec's position in it's strongest form as follows. He started by asking the question: If you look at some physical system, how can you identify if it is encoding a simulation? After laying out a set of definitions, he reasons from first principles to show that a physical system encodes a simulation if it can be decoded into a recognizable format. Reasoning from first principles, in an attempt to prevent any false negatives (ruling out valid simulations due to lack of knowledge), he argues that *any mathematically possible* decoding is valid. Our examples then showed that this lack of knowledge is a genuine problem: we are guaranteed to incorrectly classify certain physical systems as not encoding a simulation when in fact they do. This is unacceptable, for Moravec's entire *goal* was to determine a criteria that can correctly identify if a simulation is encoded in a physical system—and if we don't accept all mathematically possible decodings, we will have a criteria that is known to make mistakes (the false negative problem). %%TODO: Read through OG draft and see if there is anything else worth including ([*Draft 1 - Moravec's Mistake*](*Draft%201%20-%20Moravec's%20Mistake*.md))%%