Simulating a brain on a computer

[David Bailey]

Paul C. Anagnostopoulos has suggested that I start a new thread to consider the gedanken experiment of simulating a brain (or whole body) on a computer. Unfortunately this will involve some repetition of ideas already presented, but please bear with me.

We normally suppose that any physical system can, in principle, be simulated on a sufficiently powerful computer. This statement is more or less equivalent to stating that we know the relevant laws of physics for the system in question. The simulation of quantum phenomena will also require the use of a random number generator, but computers come equipped with good deterministic pseudo-random number generators.

Of course, if we simulate a stellar explosion or an electronic circuit, we obviously don't expect the computer program to be completely equivalent to the original thing - just to provide a good numerical approximation to its behaviour.

Now consider a human brain (possibly including the rest of the body). Suppose we simulate that in a gedanken experiment - say for a period of 30 mins. We somehow take a snapshot of the system at the start of the period - converted to numbers - and then allow the simulation to proceed on the computer.

Q1. Is the computer program consciously aware in the same way as the original brain? If the answer to this is no, what is the essential difference, and can it be rectified?

A lot of materialists will answer yes to Q1 because they consider that a brain's behaviour is only defined operationally.

Perhaps the body in question is suffering internally generated pain, so those who answer yes to Q1 should presumably worry that the simulation may also be suffering pain.

Q2. If the program is run again with the same inputs, does it suffer all over again, or are the two simulations somehow equivalent?


Now consider this. Given the combination input/program/output, what you have is a rather quirky theorem. Like any other theorem, this is true for all time (assuming the semantics of the programming language are not wilfully altered!). So we have a situation in which this emotional program has actually become a theorem - something like:

{assorted bodily inputs}=>P=>{output to glands, etc}


Now, it seems reasonable to me to ask just exactly when (and how often) does this program feel all these emotions of pain?

Q3. Is it every time the program is run, or every time someone proves this theorem, or maybe these experiences are completely de-localised over space-time, as with any other theorem?

My point is that computer is really just a very specialised theorem prover. It repeats a transformation from input to output that is already completely defined once the program has been defined. It is merely a tool for reproducing that transformation. Looked at in that light, does it really make sense to credit the system with emotions, or qualia of any sort?

Note that the above argument is extremely general - it works for any theory of consciousness, provided it is possible in principle to perform the simulation. Of course, if the brain acts as a filter then the simulation step might well not be possible - analogous to trying to simulate the behaviour of a TV set without including the incoming electromagnetic radiation.

Comments please

David

[anonymous]

The human body is an analog computer. It calculates the behavior of atoms and molecules very accurately. The universe is just an analog theorem prover, it calculates the probability of wave function collapse.

[David Bailey]

Analog processes are routinely simulated on a digital computer. Do you think QM is a vital part of the explanation of consciousness - I don't really understand your point.

David

[anonymous]

Originally Posted by David Bailey

Analog processes are routinely simulated on a digital computer. Do you think QM is a vital part of the explanation of consciousness - I don't really understand your point.

David

I don't think we will learn anything about consciousness by asking if a theorem prover can produce consciousness because it is exactly the same question as asking if a biological system can produce consciousness because a biological system is an analog computer.

However, there is probably a lot to learn about consciousness by trying to simulate it on a computer.

If we really knew enough about the human brain/nervous system/endocrine system to simulate it correctly I think we would know if it was capable of reproducing human behavior.

If we exactly simulated a human being at the biological level and it would not wake up from a coma that might prove something.

I think the real experiment would be much more valueable than the thought experiment.

[Paul C. Anagnostopoulos]

Originally Posted by David Bailey

Q1. Is the computer program consciously aware in the same way as the original brain? If the answer to this is no, what is the essential difference, and can it be rectified?

Assuming we can really simulate the entire state of affairs of the brain, then I would say yes.

Q2. If the program is run again with the same inputs, does it suffer all over again, or are the two simulations somehow equivalent?

It suffers all over again. I don't know what it means to say they are equivalent.

Q3. Is it every time the program is run, or every time someone proves this theorem, or maybe these experiences are completely de-localised over space-time, as with any other theorem?

I don't understand this question. What does it mean for the computer to feel pain when someone proves the theorem?

My point is that computer is really just a very specialised theorem prover. It repeats a transformation from input to output that is already completely defined once the program has been defined. It is merely a tool for reproducing that transformation. Looked at in that light, does it really make sense to credit the system with emotions, or qualia of any sort?

That depends entirely on what you think qualia are.

Now my question to you: If you think that a mind cannot be simulated on a computer with appropriate random inputs, then I believe you are saying that a mind is more than some combination of deterministic and random events. Could you please describe the third thing that is included? If it is easier, you can describe the third thing that is involved in libertarian free will.

~~ Paul

[David Bailey]

Originally Posted by anonymous

I think the real experiment would be much more valueable than the thought experiment.

I am a bit short of resources to try that

David

[David Bailey]

Originally Posted by Paul C. Anagnostopoulos

I don't understand this question. What does it mean for the computer to feel pain when someone proves the theorem?
~~ Paul

My point is the triple input/P/output is a theorem!! Given deterministic random number generators, the output is fixed just as much as the answer to 22+17 is fixed. Why associate consciousness with the act of merely checking this theorem with an actual computer?

Now to your question - "If you think that a mind cannot be simulated on a computer with appropriate random inputs, then I believe you are saying that a mind is more than some combination of deterministic and random events. Could you please describe the third thing that is included?"

First, I am not alone in this, Roger Penrose thinks brains cannot be simulated on a computer - his famous Godel argument.

My answer might be that the non-physical part of our brains must have a quality which makes it impossible to simulate - otherwise it too would be subject to this argument - mental stuff must be very subtle.

Again I remind you, I am not proposing an alternative theory, merely poking holes in the orthodox one!

David

[Paul C. Anagnostopoulos]

Originally Posted by David Bailey

My point is the triple input/P/output is a theorem!! Given deterministic random number generators, the output is fixed just as much as the answer to 22+17 is fixed. Why associate consciousness with the act of merely checking this theorem with an actual computer?

I still don't get it. The simulator is not a theorem checker. It actually simulates the computation of the outputs from the inputs in the same way the brain would do it.

First, I am not alone in this, Roger Penrose thinks brains cannot be simulated on a computer - his famous Godel argument.

But he is talking about a certain kind of conventional computer. We are talking about one that can simulate the brain in exact detail, including any quantum uncertainty. It's possible that we need to conclude that such a simulation is impossible, but that blows our thought experiment.

~~ Paul

[David Bailey]

Paul,

Any triple input/program/output is a theorem. Here is a trivial example:

input= {22,17}

program=

x=readinput();
y=readinput();
print(x+y);
stop;

output={39}

This is a relationship that is true however many times you run it - it is a theorem that you presumably proved for yourself without needing to execute it (and anyway, I wrote it in a simplified form for clarity).

The above example could be elaborated so that the program contained loops, recursion, whatever, and you would still have a theorem.

Quantum calculations certainly require vast amounts of computation, but they can be done on a classical computer - so I don't see this as a get-out.

David

[Paul C. Anagnostopoulos]

Originally Posted by David Bailey

This is a relationship that is true however many times you run it - it is a theorem that you presumably proved for yourself without needing to execute it (and anyway, I wrote it in a simplified form for clarity).

The above example could be elaborated so that the program contained loops, recursion, whatever, and you would still have a theorem.

I'm sorry David, but I just don't understand. Certain classes of programs can be proven correct, but there is no automatic way to prove an arbitrary program.

But more important, what does this have to do with simulating the brain? No one claims that proving an input/program/output results in consciousness. The claim is that running the simulation in the same way the brain actually works might result in consciousness. It's a process, not a result.

~~ Paul