I think you're missing another possibility (and some other posters are too): maybe it is the case that a simulation can produce consciousness, by simulating a brain for example, but that it takes a sufficiently powerful model of computation to do so.

So, consciousness being computable may not necessarily imply that any given computer could do so. One potential way for this to be the case is if consciousness is tied to some form of hypercomputation (https://en.wikipedia.org/wiki/Hypercomputation), which can provide outputs that are not turing-computable. This would mean that a traditional computer couldn't simulate consciousness, but that another hypothetical machine might be able to.

Yes, exactly.

Bostrom is assuming there's a thing called "computation" which is somehow a fully-specified solved problem. (It isn't).

Worse, he's also assuming that this "computation" can somehow simulate something called "consciousness" - which unfortunately for him, is nowhere near to having the formal rigorous definition that "computation" as we know it today would require.

And finally after all that, you get something which is subjectively indistinguishable from life itself.

Well - obviously. If you don't define any of your terms and assume your argument is correct because you don't show your working or get into specifics, you can argue whatever you like.

> what is special about life that makes it unrepresentable with physics

absolutely nothing

> The physics consciousness relies upon is fundamentally uncomputable

This would be "huge if true", but I don't think it is. At least I haven't seen any indication of why it would be.

> Consciousness _is_ computable and all the weird consequences that follow

I think this is true, and I think the weirdness is a result of misunderstandings on our part (we are notoriously bad when reasoning about systems that contain us in some way).

> What are your thoughts on why a simulation cannot be conscious?

I think this whole consciousness confusion is due to mixing levels of abstraction on our part. Replace consciousness with other higher-level cognitive or social functions, like "love" or "patriotism", and see how silly it sounds. Is "love" computable? Can "loyalty" be reduced to physics? Yes, but the concept makes sense on a whole different conceptual level, which would be silly to reduce to, say, quantum physics, and is much better to reason about at the right abstraction level.

I lean toward the first option for the following reason: I doubt that physical processes can be simulated precisely by a Turing machine.

For example, if I'm living in a simulation and I move my hand an inch to the left, it occupies an uncountably infinite number of intermediate positions during that motion. A Turing machine can't calculate or even represent all of those positions, since the set of Turing-computable numbers is only countably infinite.

Of course, what I've written above assumes that spacetime is continuous. Is it?

That can be tested by experiment:

https://physics.stackexchange.com/questions/9720/does-the-pl...

According to that link, spacetime shows no sign of discretization down to fourteen orders of magnitude below the Planck length. (The Planck length itself is so small that if an object that size were expanded so as to be a millimeter across, then a proton, at that scale, would be larger than the distance between the sun and Alpha Centauri.)

That suggests to me that while Turing machines can provide useful approximations to the behavior of physical systems, they can't represent or simulate them precisely.

I don't think there are an infinite number of positions. The Planck length [1] essentially defines the smallest unit of length. Similarly, Planck time [2] does the same for time.

[1] https://en.wikipedia.org/wiki/Planck_length

[2] https://en.wikipedia.org/wiki/Planck_time

Your first reference states that the Planck length isn't the smallest unit of length in standard physics: "The Planck length is sometimes misconceived as the minimum length of space-time, but this is not accepted by conventional physics, as this would require violation or modification of Lorentz symmetry." It then states that alternative theories do allow for a minimum length.

Experimental evidence (the Fermi satellite data discussed in the link in the grandparent comment) seems to support standard physics.

Computation does not need to be digital. An analog computer could represent positions down to arbitrary precision, though accuracy would still be limited by the information input.

> If a simulation cannot produce consciousness, what is special about life that makes it unrepresentable with physics?

Authors can write stories with believable worlds and compelling characters. Given infinite time, an author could detail how a character would react to every possible situation. But no matter how much they wrote down, their words would just be words. There isn't some breaking point where the author writes enough to create a pocket universe.

Why are turing machines different?

Just because the universe is governed by physics does not mean that physics are sufficient to construct a universe. I expect that "physicality" is a real property of some kind, which doesn't appear just because you described it accurately.

So much of this is guesswork, I may well be wrong—but it strikes me as the most logical conclusion.