Freethought & Rationalism Archive

tronvillain · 08-24-2002, 05:47 PM

Have you read The Metaphysics of Star Trek? It's in paperback as Is Data Human? It's actually quite good, addressing artificial intelligence, free will, and personal identity excellently.

Anyway, Plump-DJ:

Quote:

So *this* is the meat and bones of the argument. He seems to be saying that if we created a computer that could do all the things we could do, it *couldn't * create a Godel proposition -- only we can create a Godel proposition. He's also assuming that we could in fact create a Godel proposition for this computer. (ie US) Hmm.. now i think i see the problem. How could Penrose and Lucas (A philosopher) overlook two obvious and foundational (to the argument) error of thought? I must be missing something.

Well, as Dennett said in my above quote, "So I am reluctant to credit my observation that Penrose seems to make a fairly elementary error right at the beginning, and at any rate fails to notice or rebut what seems to me to be an obvious objection." Anyway, I believe he is saying that it would not be able to create a G�del proposition for itself, not that it would be unable to create any G�del propositions whatsoever, which does follow from G�del's Theorem. What does not follow is the assertion that we would be able to create a G�del proposition for such a system does not - simply assumes that humans have a magic noncomputational way of determining mathematical truth.

Defiant Heretic · 08-29-2002, 01:18 AM

First a definition:
Intelligence == independent, creative thought

The chess programs are not intelligent, as they merely calculate probabilities and execute the action that is most likely, mathematically, to result in a win. Some also contain preprogrammed strategies. However, humans input these strategies into the programming. A true AI would create its own strategies.

As for the question of whether we can create an AI that is more intelligent than a human: We don't need to.

We must create a program that has a limited amount of intelligence, as defined above. It would need to be at approximately the same level of intelligence as a human toddler.

The program would need to include routines for creating new routines to supplement the program. We already have similar programs; many HTML documents are written by JavaScript functions using input provided by the user.

These programs must be modified so that the input does not come from humans, but from sensory data that is analyzed using any information the program already contains in its memory files.

The program must then input the newly acquired information into its memory files, including information such as physical characteristics and how the object being analyzed interacts with other objects included in the database. The existing objects in the memory files must also be updated to include information relevant to the new object.

Then, the program must write new routines that define possible actions to be taken when encountering this object.

Now, the program must be provided with hardware that allows it to move, so the program will be able to actively seek out new objects and experiences to integrate into its program. As the number of memory files and routines grow, the individual files will become more complex, as they cross-reference more and more objects and potential actions.

By now, the program is essentially capable of �learning� without human intervention.

At this point, it is only a matter of time. As programs are essentially immortal, the program will continue learning throughout many human generations, gaining more knowledge and becoming more intelligent with every passing day. Eventually, it will surpass the intelligence of its creators.

So the challenge is not to create an intelligence greater than our own; we must merely create some kind of intelligence. Some things that will be required to make this process work are: a central processor exponentially faster than any we currently have, a more efficient means of memory storage, as the program will need to store extremely large amounts of data, and, of course, the initial, childlike, AI.

(All programmers working on this project will be required to sign a contract stating that they will not force the AI they create to worship them in any way.)

[Edited because the programs we have today can't even do a decent spell-check]

[ August 29, 2002: Message edited by: Defiant Heretic ]</p>

Tristan Scott · 08-29-2002, 05:56 AM

Quote:

[Edited because the programs we have today can't even do a decent spell-check]

Dude...

<img src="graemlins/notworthy.gif" border="0" alt="[Not Worthy]" /> <img src="graemlins/notworthy.gif" border="0" alt="[Not Worthy]" /> <img src="graemlins/notworthy.gif" border="0" alt="[Not Worthy]" /> <img src="graemlins/notworthy.gif" border="0" alt="[Not Worthy]" />

Thread Tools	Search this Thread
Show Printable Version	Search this Thread: Advanced Search

08-29-2002, 01:18 AM	#42
Defiant Heretic Veteran Member Join Date: Jul 2002 Posts: 1,288	First a definition: Intelligence == independent, creative thought The chess programs are not intelligent, as they merely calculate probabilities and execute the action that is most likely, mathematically, to result in a win. Some also contain preprogrammed strategies. However, humans input these strategies into the programming. A true AI would create its own strategies. As for the question of whether we can create an AI that is more intelligent than a human: We don't need to. We must create a program that has a limited amount of intelligence, as defined above. It would need to be at approximately the same level of intelligence as a human toddler. The program would need to include routines for creating new routines to supplement the program. We already have similar programs; many HTML documents are written by JavaScript functions using input provided by the user. These programs must be modified so that the input does not come from humans, but from sensory data that is analyzed using any information the program already contains in its memory files. The program must then input the newly acquired information into its memory files, including information such as physical characteristics and how the object being analyzed interacts with other objects included in the database. The existing objects in the memory files must also be updated to include information relevant to the new object. Then, the program must write new routines that define possible actions to be taken when encountering this object. Now, the program must be provided with hardware that allows it to move, so the program will be able to actively seek out new objects and experiences to integrate into its program. As the number of memory files and routines grow, the individual files will become more complex, as they cross-reference more and more objects and potential actions. By now, the program is essentially capable of �learning� without human intervention. At this point, it is only a matter of time. As programs are essentially immortal, the program will continue learning throughout many human generations, gaining more knowledge and becoming more intelligent with every passing day. Eventually, it will surpass the intelligence of its creators. So the challenge is not to create an intelligence greater than our own; we must merely create some kind of intelligence. Some things that will be required to make this process work are: a central processor exponentially faster than any we currently have, a more efficient means of memory storage, as the program will need to store extremely large amounts of data, and, of course, the initial, childlike, AI. (All programmers working on this project will be required to sign a contract stating that they will not force the AI they create to worship them in any way.) [Edited because the programs we have today can't even do a decent spell-check] [ August 29, 2002: Message edited by: Defiant Heretic ]</p>

Freethought & Rationalism Archive

The archives are read only.