These networks seem to be able to perform a few simple forms of learning but clearly there is nothing approaching thinking going on in these comparatively tiny networks.
Present day computer systems can be programmed to perform a limited form of metaprogramming activities including so-called introspection implemented in certain programming languages such as Java, SmallTalk, and Tcl. This use of the term introspection is rather an abuse of the meaning of the term as it was originally meant. This so-called introspection has no awareness or consciousness whatsoever. It is an abuse in the same way that the term intelligent is bandied about when talking about missiles or golf clubs. (We will return to the discussion of introspection as it is an important element in metaprogramming systems.)
It would be useful for you to read Alan Turing’s lectures/publications, to read the correspondence of Charles Babbage and of Ada Augusta Lovelace. To know only what you know of the last few years of computing is to be blindered.
In addition to those items it might be of value to read Searle’s The Chinese Room. Basically it is an argument that symbol processing does not implement intelligence. On the other side of the argument if you read Daniel Dennett, such as Consciousness Explained, you see reasonable argument for what kind of processing can/does make intelligence. It is my view that, in the case of The Chinese Room, it is the book of instructions that is/has the intelligence, and it was generated by an intelligent agent. The man in the room who uses the book has some intelligence too or he wouldn’t be able to read/use the
book would he? The book constitutes an understanding of Chinese (symbols) in that it prescribes how to USE them (operationally). This is the notion that the meaning of something equals how that something is used. A boulder is a chair if one decides to sit on it to rest. We will return to where does the generator of the book of instructions in the Chinese Room get his intelligence so that he can put that into the book.
In the posting of the previous day we said : Second Order Metaprogramming, when its capabilities are applied to the system itself is called Self-Metaprogramming. Present day computer system designs are derived from thinking that goes back to Charles Babbage, if not earlier.
Current computer design, largely the Von Neuman design, has a memory section (ram) which stores bits, packaged as strings of bits, called bytes and words. The circuitry which contains these bits is designed so that the bits can be put there or transfered from there according to an address scheme.
This is essentially a pointer scheme which operates (conceptually) sort of like the turnstile of a railroad yard which turns a section of track so as allow the train to pass along a particular line of rail. Im not faulting this approach merely pointing out that address usage is largely sequential access. Arithmetic processing and logic processing are carried out by specialized circuitry which is different from the kind of circuitry which implements memory. The contents of memory has to be transferred into the arithmetic processing circuitry (section) in order that that kind of processing can be performed. The results have to be transferred back into memory. The same goes for logical processing.
The point is that memory and processing are done by specialized circuitry which exists at different physical locations of the computer system. Any operations that are performed on any data means that there are two sets of transfers required, to move data out of and into memory. [continued next posting]