3 ms·

Suppose you created a machine which, in the style of Solomonoff induction (sorta), was designed to find as short or as "elegant" (in whatever sense of that word

by drdeca 11y ago

Suppose you created a machine which, in the style of Solomonoff induction (sorta), was designed to find as short or as "elegant" (in whatever sense of that word) of a (computation based) description of the world (so, it makes a model of the inputs it receives, and it receives many inputs, including from the internet, etc.) Further, pretend that it has arbitrarily large computational resources. Even if it could completely predict the inputs it receives (modulo quantum mechanics based randomness), would you expect that it's internal model of the world includes some description of consciousness, of an internal experience? I do not mean, would it have some sort of a model of agents, entities that act in a way that tends to optimize some things or other, Rather, I specifically mean the internal experience of things, not just being able to predict that people would say that they experience. I do not think that it would come up with a model of internal experience, And yet, I experience, and so do you. So, I think an understanding of reality based only on computation about physical objects, is incomplete.

Houshalter 11y ago
Yes. If you ran Solomonoff induction on the world, it would generate (countably infinite!) many simulations of people in the world, each of which would act exactly as they do in the world. They would experience the same things, say the same things, behave exactly the same way in every respect. Including having the same thoughts and feelings you are right now, and typing the exact same comment.
- drdeca 11y ago
  Two things: 1) I didn't necessarily mean exactly Solomonoff induction, as what I was describing would be for finding a single good description, rather than making good predictions using a weighted combinations of many descriptions. 2)Well, whether they would have experiences is a topic which is of some debate I think. They would behave as if they did, but, would they? If the simulation is run twice concurrently, are the experiences had twice as much? There are all sorts of potential complications with things like computation with secret sharing schemes, and so I don't think the question is really settled. ____ My point was that even if something describes what occurs, what particles end up in what configurations, that doesn't seem to be a description of what is experienced, even if it predicts how people will self report their experiences, and give causal explanations for their self reporting of their internal experience.

by drdeca 11y ago

Houshalter 11y ago

Yes. If you ran Solomonoff induction on the world, it would generate (countably infinite!) many simulations of people in the world, each of which would act exactly as they do in the world. They would experience the same things, say the same things, behave exactly the same way in every respect. Including having the same thoughts and feelings you are right now, and typing the exact same comment.

drdeca 11y ago
Two things: 1) I didn't necessarily mean exactly Solomonoff induction, as what I was describing would be for finding a single good description, rather than making good predictions using a weighted combinations of many descriptions. 2)Well, whether they would have experiences is a topic which is of some debate I think. They would behave as if they did, but, would they? If the simulation is run twice concurrently, are the experiences had twice as much? There are all sorts of potential complications with things like computation with secret sharing schemes, and so I don't think the question is really settled. ____ My point was that even if something describes what occurs, what particles end up in what configurations, that doesn't seem to be a description of what is experienced, even if it predicts how people will self report their experiences, and give causal explanations for their self reporting of their internal experience.