bigideas

Phil began a discussion of consciousness and computation based on some of the ideas put forward by Roger Penrose in his various books. By computation we mean “anything that a Turing Machine can do”, or, indeed, anything that a modern computer can do. We initially sidestepped a discussion of what we really meant by consciousness (or indeed awareness, understanding or intelligence) and accepted that it was probably “something” of which we had some intuitive knowledge.

We considered the following four extreme viewpoints, A, B, C and D, that one may reasonably hold on the subject of computation and consciousness:

A. (Strong AI) All thinking is computation. The action of carrying out that computation evokes awareness.

B. (Weak AI) Awareness results from some physical action of the brain. This physical action can be simulated computationally.

C. Awareness results from some physical action of the brain. This physical action cannot be simulated computationally.

D. We cannot fully explain consciousness by any physical, mathematical or any other scientific means.

Penrose’s books argue against A and B, and attempt to build a strong case for C using Quantum Theory.

During the course of the evening we enountered, amongst other things, the Turing Test, the Chinese Room, and a Godelian argument…

The Turing Test

A human judge engages in a text-based conversation with another human and a computer program. If the judge is unable, after some sufficiently long time, to reliably determine which is which then the program is said to have passed the Turing Test.

We observe that A and B allow for a computer program to pass the Turing test. Under C, no program could pass; it would eventually reveal its lack of consciousness after a sufficiently long interrogation.

Chinese Room

(http://en.wikipedia.org/wiki/Chinese_Room)

This thought experiment is intended as an argument against A, and attempts to demonstrate that a computer program which appears to be conscious has no actual understanding.
Suppose we have a computer program that can pass the Turing Test in Chinese. That is, it takes Chinese characters as input, performs its computations and outputs Chinese characters. A Chinese speaker can engage in a conversation with this computer program and be unable to tell that it is not human.

This program could be formulated as a book of rules for a human to follow. Suppose that a such a human, a non-Chinese speaker, sits in a room. Through a window in one wall she takes an input of Chinese characters, consults this list of rules, manuipulating various counters and symbols as instructed, and outputs a sequence Chinese characters as the rules dictate. She performs the same function as the computer and the program and is therefore able to conduct a credible conversation in Chinese. However, she does not understand Chinese.

The Godelian Argument

(http://en.wikipedia.org/wiki/Shadows_of_the_mind)

Following along the lines of Godel’s incompleteness theorem, this is intended as an argument against B.

We suppose that there exists an algorthim, X, that encapsulates all techniques available to mathematicians, which can demonstrate that some computations never terminate. The proof goes on to show* how we can find a computation which we know never terminates but for which X is unable to provide a demonstration of non-termination. We conclude that therefore X cannot encapsulate all techniques available to mathematicians.

The idea here is that we have taken a small aspect of human consciousness, namely the ability to provide mathematical proof, and shown that it cannot be performed computationally.

(*) In brief…

1. We enumerate all computations C that act on a single number n: C1(n), C2(n), C3(n), etc.
2. X can therefore be considered as operating on a pair of numbers: If X(n,m) terminates it provides a demonstration that Cn(m) does not terminate.
3. We consider X(n,n) and observe that this is a computation acting on a single number and, by (1) must equal Ck(n) for some k.
4. We now consider X(k,k). By (2), if X(k,k) terminates then it provides a demonstration that Ck(k) does not terminate.
5. However, by (3) X(k,k) = Ck(k), so, from (4) we have that if Ck(k) terminates then Ck(k) does not terminate. We deduce that Ck(k) does not terminate.
6. Therefore, by (3), X(k,k) does not terminate.
   So, we have been able to find a computation Ck(k) that we know does not terminate but X is unable to demonstrate this fact.

Other Matters

The following list is just to jog your memory if you were there…

• What do we really mean by consciousness? Can we even start asking the question until we have a definition? But then, what definition could really capture what we mean by the term? Is it even definable?
• Are animals conscious? We can’t apply the Turing Test to an animal because it can’t use language. If an earwig could talk, could it pass the Turning test? Coulkd a human being fail the Turing Test? What would that mean? Would, for example, an estate agent pass, or would we think all the non sequiturs about things being nice were a dead giveaway that it was a machine?
• Are there, then, many types of consciousness? Computers may possess a consciousness that is simply different to human consciousness.
• If someone wrote a piece of software they claimed was conscious, how would we ever know for sure? This leads us to the other minds problem.
• If there’s something physical about the numan body that augments our computational ability and gives rise to consciousness, why couldn’t we include it in a machine? Could we build a conscious computer using a human body? (Here we digressed briefly on the topic of “posh gravity”)