Eliezer S. Yudkowsky, <firstname.lastname@example.org>, writes:
> I think we all agree that even if a million monkeys typing on a million
> typewriters generated the paragraph above, it wouldn't make the Hal who
> wrote it real. Only Hal's brain or a simulation thereof - being
> instantiated - can put actual qualia behind the statement. It's not
> enough to have the inputs and outputs in a Giant Lookup Table. So the
> question is, if the playback was generated by random chance, can the Hal
> it "records" be said to exist? Will he ever, even once, have said
> "cogito ergo sum", or is there only a text-based representation of the
> words? For the purposes of this argument we are *assuming* that an
> actual neuron-by-neuron simulation would make the Hal real; the question
> is, *given* that, does a randomly generated recording also make Hal real?
It seems that you are combining two ideas here: playback and randomness.
In speaking of "playback" it is simplest to consider making a record of a calculation, and then somehow replaying that record. As you described, there are a number of variations of that idea.
But the question of random actions which "just happen" to mimic (or create) consciousness output is a different matter. This would not have to be a replay of a previously existing consciousness.
You proposed a system with a little bit of randomness, one which is usually lawful but occasionally makes an unplanned transition. We could have a knob to vary the amount of randomness, from a totally deterministic conscious calculation to a totally random, uncontrolled series of states.
Clearly, in practice, as we move the knob from the deterministic towards the random position the consciousness is going to begin to go bad. It would be like zapping the neurons in your brain more and more often, perhaps taking stronger and stronger drugs. Eventually you go insane, dissolve into total, chaotic madness, and (presumably) are no longer what we would call conscious.
In practice, then, we have a clear relationship between determinism and consciousness, controllable by the knob. A random machine is not conscious.
Now, it is theoretically possible that, even though the knob has dialed in a significant amount of randomness, that we "get lucky", and the machine happens to continue to work right. As we move the knob more towards the random position, this becomes less and less likely. With the fully random position it is astronomically unlikely.
I have seen it suggested that exploring the behavior of systems so far from "reasonableness" is outside the range of our intuition. Our common sense fails as we move to the quantum level, or out to the range where relativity holds sway. It may be equally true that when dealing with events which are of such extraordinary low degree of probability, that our common sense fails there as well.
It could be that a fully random state machine which nevertheless managed to instantiate in full and complete detail the complexity of a human mind, thinking thoughts over an extended period of time, would in fact be conscious. Whatever intuition leads us to think otherwise must be distrusted, in a realm so far from that where common sense is proven to work.