What step in this argument is the least valid?

Question

I know this is not a totally formal argument, and it is not mine, I just tried to distill it as much as I could to find where the weakest point is. Please help me if you want. It goes like this:

H (hypothesis): we can create self-conscious entities in a virtual reality simulation

T (thesis): H implies either that (a) there are non-physical (in the sense of Platonic) self-aware entities; or, (b) The universe itself is non-physical.

How to go from H to T:

1) the simulation runs in discrete time steps. By symmetry (or lack of a preferred time scale), the separation in time between time steps does not change H. Nor should the time it takes to complete a time step, thus the size of a time step itself. Nor if all time steps are simultaneous.

2) From (1) we can run the simulation at any speed even the micro-steps within every time step.

3) Every micro step corresponds to a configuration of the computer. It should not matter how we reach each step, either by "computing" it or by a look up table (that is, there is no meaningful difference between initial data and computed data).

4) The look up table and a computer reading it is all needed. The H self-aware entities become conscious as the computer performs the reading. Then: is it important "what" performs the reading? Should encryption matter? the data from the table can be transformed by a computable function before being put into a computer. Does that transformation affect the consciousness? It should not because the simulation should be arbitrary up to at least a computable function. It should not matter if we are able to watch the beings in a screen or if the simulation is encrypted so no outsider can "see it". Same with the computer: it can be any arbitrary machine that is equivalent to the original as far as we can computably pair the states of the original machine with those of the new one.

5) It follows from (1), (4) : it only matters the existence of the lookup table itself, not the specific implementation. The machine is arbitrary, it can be the book itself where the look up table is printed.

6) A book with a printed look up table should generate self-conscious beings. But how does it operate? Not in time certainly. Or, it could be regarded as a computer that operates at infinite speed, meaning all states of the simulation (all time steps) are on memory simultaneously.

7) Does the physical existence of the book matter at all?

UPDATE:

I think now that the H I am assuming is actually (a). So this is not really an argument at all! (or it is rather some kind of tautology). But still I would like to hear more opinions. Thanks!

Answer 1

I have a problem with (1). Why do you assume the time scale doesn't matter? If a human brain ran really really slowly, we would not be conscious. Or at least it's an open question. So I think (1) is the first problem. But really T does not follow from H and I don't think any of your argument holds water.

But the idea that running the simulation slowly would still give you some sort of (slow-witted?) consciousness, I find highly doubtful.

As evidence, consider chemical reactions. If you slow down a reaction (for example by lowering the temperature of the apparatus) you kill the reaction. You can't build a slow fire. Many physical processes, especially biological ones in the brain, involve time in a qualitative manner.

If I understand your argument, you are saying that we can write down the state-transition diagram for the simulation, and then read it really, really slowly. Clearly there would be no consciousness generated. Your example disproves your argument. You have provided a counterexample to your own idea. Because it's manifestly absurd that I would generate consciousness by reading down the transition table and "playing computer," as we are taught to do in programming class.

Answer 2

I notice two things that run through your steps. The first it is brought up again and again that the method (tool) used to implement consciousness does not inherently matter, but I believe it does. The book is not conscience because it can take no action (it is inert). The computer may be conscience because it can take action. I believe the ability to take action is at least one requirement for being conscience. The second is that the steps accept input from an external source. Is a remotely controlled robot conscience? Is there a difference between the "lookup table" and a remote pilot to a robot? I doubt people would see a remotely controlled robot as conscience, rather they would look to the controller.

The flaws come in part from a poorly defined term "conscience". First, let me offer that I believe you mean consciousness. Consciousness is the state or quality of awareness, or, of being aware of an external object or something within oneself. We sometimes say being self-aware. Implied in the definition is that this awareness is a self-contained action. It does not come by virtue of a look-up table or any external other input. I am not saying that it is not learned, just that if learned, it does not require input to be maintained. Assuming that you accept that definition then step three is not valid because the system is not self-contained.

Finally, all of the above steps assume a computational model for consciousness. That is to say that consciousness can be mapped into specific steps and choices based on inputs. You have not demonstrated that this is true. The requirements for the tool to implement consciousness depend on the answer to this question. So steps 1,2,4 and 5 are ambiguous.

Answer 3

It follows from (1), (4) : it only matters the existence of the lookup table itself, not the specific implementation. The machine is arbitrary, it can be the book itself where the look up table is printed.

A book is not a machine.

You are imagining that the book contains, essentially, a giant state machine diagram. You are then asserting that any implementation of the state machine could fill the role of the machine. I am willing to allow this assumption for the sake of argument. But it all goes off the rails when you assert that the book itself is an implementation of the state machine. A state machine has multiple components:

1. The diagram or lookup table itself.
2. The current state.
3. Some input data which the machine is currently processing.

The book has (1), but it is missing the rest. So it is not an implementation of the state machine, and the rest of the argument fails.

(I should also note that, in practice, we usually use the more complex Turing machine model to describe computation in the most general sense. However, a true Turing machine is infinite, and when you constrain it to be finite, it is effectively reduced to a very elaborate state machine, so this is not wrong by itself.)