Quote:
Originally Posted by Paul C. Anagnostopoulos  You're oversimplifying with this theorem thing. First of all, the inputs to a brain simulator are incredibly complex, including interactions with arbitrary parts of the external world. Some of the inputs are clearly random.
Second, a computer program is different from the execution of the program. A computer program that blows up my car will not blow up my car if it is sitting on a shelf. I cannot ask a program questions to see whether it passes the Turing test if it is not running.
Third, a computer program is only a complete statement of a theorem if it has no state. If it collects data as it runs, then only the running program can be said to be the whole story.
The question is: How close do we have to come to building a human brain in order for it to pass the Turing test? If it's impossible, as I think Ian says, then that's pretty interesting. Otherwise it's just a question of accuracy of simulation.
~~ Paul |
The complexity of the external world doesn't really matter, because I am taking the input as fixed. Let's say you get out of bed and stub your toe, and yell out. There will be various inputs to your brain (or your whole body simulation, if you prefer), and the simulation will process them in a completely fixed way. Random number generators can be added to the input if required. We are considering one particular instance, with one set of inputs, one set of inputs coming later in the event, one initial brain state. So whatever you felt on the first occasion, should (according to you) replay each time you run the program to check the corresponding theorem!
I am not quite sure what you mean by the program that blows up your car - is this an Al Qaeda type thing, or something that inflates the tyres? However, whatever it is, the point is that the computer program is still a theorem checker, but it is connected to a bomb or a pump, as required. You want your computer to be conscious in itself - not to send signals off to a 'consciousness unit' to experience the qualia because that gets us back to square one!
I also think the Turing test is a red herring. It was devised over 50 years ago when computers were quite primitive, and I am sure that, had he lived, Turing would have revised the test. One problem with the Touring test, is that no thought was given about the issue of cheating. Suppose, for example, that the computer had access to a huge volume of human written discourse. I type it a sentence such as "Hello, how are you?" - let's just call that sentence A. Suppose the computer scans its text looking for conversations that start with A, and simply returns the next sentence B. I reply with a further sentence C, and it looks through its text for a conversation that starts with ABC, and returns the next sentence D. With a very large body of text and good luck, it might pass the test, but in no possible way could it be said to be conscious.
David