Part 1: Definitions
Measure: filtering out all information about the subject except for a useful comparison
Some dictionaries have pretty good definitions of “measure”, and others provide almost no useful information. A good definition should point out that measurements are really only possible by making a comparison, but a really important fact is that it’s not a general comparison, it’s very specific information you want when measuring.
Measuring really is an act of ignoring most information, we want to focus on one quality and try to get the best idea possible of the dimension or extent of that one thing. The way we do this is by comparing what we’re measuring to something else, and if we want to be able to record or share the information we want to compare it to some kind of standard. The obvious example is using a ruler marked with standard lengths, but it could also be thermometer that is known to react in a standardized way or even a comparison to a standard color swatch.
Ultimately though all measurements are estimates, we can never know exactly the properties of something we’re observing, the best we can do is to find a comparison that’s so close we can’t tell a difference. This is one of the major limitations of simulations, at least if we’re trying to exactly replicate the real world, when we’re creating a simulation we can never be sure that we’ve created a perfect duplicate.
Simulation: Takes a group of measurements and applies programmed rules to them to create a new set of artificial measurements.
The key to this definition is that the process of getting information into and out of a simulation isn’t that important, it could be a monitor and keyboard, or VR or pen and paper. What’s important is what kind of information goes in and out, it has to be measurements, and the simulation is run by applying rules to those measurements to create new ones. If the simulation is accurate, then the measurements you get out of it will match the measurements you’d get in the same situation in the real world.
Part 2: Applying Definitions & the Chinese Room
Searle’s Chinese Room is a famous argument against a specific idea of “strong AI”, and it’s one of my favorite thought experiments of all time. I’m going to assume a familiarity with the specifics, the link above has a great overview of how it works. A lot of people have disagreed with the conclusion it comes too, but often that’s because they misunderstand what the conclusion actually is. The argument isn’t that:
- Artificially intelligent machines are impossible
- The Turing Test isn’t accurate
What the argument is proving is actually kind of narrow, it’s just that programs can’t create consciousness. Searle acknowledges that it must be possible to create an intelligent machine with a mind since humans are essentially biological machines, and we have conscious minds.
- Short aside, here for the definition of “program” I’m thinking of it as “an ordered series of settings or limitations with the goal of allowing certain inputs and outputs”. But for this discussion most general ideas of what a program is will work well.
Even Searle’s relatively narrow conclusion has still attracted a lot of criticism and counterarguments. There’s lots of people who think that a program should be able to create consciousness, although I haven’t seen any convincing argument that found a flaw in the Chinese Room. I have a different problem with the conclusion, I don’t think it goes far enough.
The Chinese Room is a great thought experiment because it allows us to imagine a computer that’s perfect in some ways. It has essentially unlimited resources and its processing is completely transparent and can be run on anything. A human carries out each individual step and the medium that they carry those steps out with could be anything – pen and paper, matchsticks, characters carved in marble, etc. This allows us to focus on just the program, and imagine it running essentially any program that’s possible.
One subset of all possible programs is all possible simulations, the Chinese Room can run any simulation that’s possible. This means that the Chinese room can create all kinds of interesting information and knowledge. In theory, it can tell us about how the universe was formed, what will happen to it in the future, how life evolved and how humans became intelligent. It could probably reveal what I’ll have for lunch today or create nearly any mathematical pattern or model any emerging phenomenon.
However the argument also shows us that no matter how carefully we simulate the universe or any part of, no matter what program we use, it won’t create consciousness. But that’s not the only limitation that this perfect simulation machine has. There’s a whole group of things it can’t create, and we can see what these are by looking at the definition and seeing that every simulation needs to be fed some measurements to start.
For example, if we wanted to simulate a plane we could start with a very basic model, enter in the size of the plane, and information about its environment. But we could also go more basic, enter in information about what its made of, and exactly where it’s flying. Or we could go even more fundamental and just enter in information about where all the atoms are and let the simulation calculate everything. In the real world this is impossible complex and slow, but it’s just a matter of time and resources for the Chinese Room, eventually it would give us a simulation of flight. But even with unlimited resources we’d still need to tell the Chinese Room a few important measurements. What’s the strength of gravity, or what’s the formula for the electromagnetic force for example. These are values that are impossible to simulate because they’re just fundamental properties of matter and a simulation can’t create these forces to measure what properties they have. When we simulate a light it doesn’t create actual lightwaves. In the same way the human in the Chinese Room can verify there’s no consciousness created they can also verify there’s no light (or extra gravity or magnetism, etc.) Light could have a different speed, the fundamental force of gravity could be stronger or weaker, these are values that can only be measured in the real world because they’re properties of matter, they can never be measured inside a simulation or any program.
A program can only apply a series of rules, it can never create a fundamental force, so there’s some values it always has to be given and can never calculate. The list of things it can never create are all fundamental properties of matter, the strength of the various forces, the starting position of all particles, information about quantum mechanics, etc. And also what consciousness is like.
The obvious conclusion from looking at this list of limitations is that the reason a program can’t create any fundamental property of matter or consciousness is that consciousness is a fundamental property of matter. This is the next conclusion that we should draw from the Chinese Room, not only can’t a program create consciousness, but why not? What is it about consciousness that prevents it from being created? I can’t see any other conclusion to reach except that consciousness is a fundamental property of matter in our universe. Either it’s a fundamental force we haven’t discovered yet, or a fact about the starting characteristics of the particles, or we’ll need to learn something new about the objective and subjective nature of the matter that is in, and essentially creates, this universe.