Logic: The Wittgenstein problem

This is a problem which should have puzzled every person who has ever ventured into philosophy of any flavour.

The following is a slightly edited and altered text quoted from a post by Bollinger, which he calls The Wittgenstein problem.

"The central questions are:

1. How do we manage to coherently label the world into particular objects through the use of language?

2. (a) And if labelling is a purely conscious process then why can't I explain how I manage this labelling?
(b) In other words how can I not know something I am doing consciously?

This I call the Wittgenstein problem in honour of the great Austrian thinker who posed these types of questions in his work 'Philosophical investigations'."

(Before beginning, I must congratulate "Bollinger" at www.ephilosopher.com for a very clear statement of the questions.)

These are two beautiful questions that have always puzzled me. And I must admit that I am in no position to give a definite answer to them. I do agree that we, in general, manage to coherently label the objects we observe; and that we have difficulty explaining how we manage to do this labelling.

I think I encountered this problem quite a few years back when I started wondering about what to call a living thing. That was during my schooldays when I had started pondering over the subject of artificial intelligence and artificial life. To be able to create what could be termed living or to even explain the origin of life, one must have a clear conception of what a living thing is. Such a concept or definition I have failed to arrive at.

Similarly, one could consider even the simpler objects and still stumble upon the same hurdle. To define X as a class of all the objects that are labelled X serves no great purpose. Because this definition would not guide you whether a new object that you come across could be labelled as X or not. Also, such a definition will keep getting revised each time a new object (i.e., one that is not already in the class) is labelled X.

I will not go into detail of how the difficulty of explaining the method of labelling arises. Anybody who has given some thought to this subject will easily understand this and raise the above questions. However, a clearer idea of the problem can be obtained by reading the first post in the discussion on The Wittgenstein problem.

Let me repeat that I am not suggesting a solution. I will try to show one way in which it is possible for us to coherently label without managing to explain how.

Consider a very complex neural network N1 that is designed to be able to classify things which it may be able to do quite satisfactorily. Let N1 also have a feedback mechanism by which it can occasionally tell that its classification was wrong and, at times, it can make corrections too. But N1 is not designed to output the exact configuration of weights that allows it to do the classification. The result is that N1 can coherently classify the objects (and hence, label them) without being able to explain how it does so.

What are the lessons one can draw from N1? If our brains and the associated paraphernalia are designed/wired/evolved to coherently label things (among other faculties) but not to bother about the design then it explains how the two questions arise. The N1 example has been kept very simple to illustrate the essential elements only. Better examples can be formed without adding much.

However, N1 misses one part of the question completely. That is, "if labelling things is a conscious exercise, how can one not know how I label things".

The introduction of the term "conscious" (and its related forms) into the question necessitates understanding consciousness and its capabilities.

Consider a very complex neural network N-omega (many generations higher than N1) which can, of course, classify and label objects. N-omega can also do some maths. But what distinguishes N-omega from other generations of neural networks is that it also asserts an identity in its communication with its human users. Assume that this assertion of identity results due to its high level of complexity (I am not sure if that is possible).

It then follows from the design of N-omega that it can label things, and it asserts that it is conscious when it does such labelling. But it may still not know how its consciousness works. To be conscious of a process is quite different from understanding that process.

In stating all this I am not saying that this is how the brain works and hence the questions. What I am saying is that such questions can co-exist in certain scenarios, one of which is rudimentarily shown above.
----------------------------------
Conclusion
----------------------------------
1. The ability to coherently label things could be a result, among other things, of
(a) knowing explicitly the criteria to be followed in doing so and then using the classifying apparatus (brain, neural network, computer etc.) to algorithmically (or otherwise) put the object to the test of the given criteria, or
(b) only the structure and complexity of the classifying apparatus.

2. If the criteria is explicit, there is no difficulty in explaining how the coherent labelling is possible. But, if the labelling is a result of only the structure and complexity of the apparatus then it will be difficult to tell how the labelling is managed so coherently. The degree of difficulty should increase with the complexity. As you can see, this difficulty has nothing to do with being conscious of the process or otherwise.

Also, to be conscious of a process is quite different from understanding that process. Therefore, the only knowledge that can be inferred from being conscious of a process is the knowledge that one is conscious of that process. Being conscious alone does not imply any further knowledge about that process.
----------------------------------