posted on Apr, 24 2008 @ 11:06 AM
An online forum would not be the best proving ground for AI for several reasons. 1. because first generation AI will probably take quite some time to
learn how to interperet flawed grammar and spelling. 2. Because it's sometimes difficult to know who a poster is talking to, and thus human review is
required to tell the difference between miscommunications and background noise.
Let's consider the operating mechanism of AI for a moment. The major barrier is a comprehensive understanding of language by a machine. A full
understanding of language would enable a machine to learn things simply and code the new information for its own use by itself.
So you start with a contextualized dictionary. I'm not talking about Websters. I'm talking about the most complete and nuanced dictionary/thesaurus
in human history, within which an astronomical number of contextual connections are drawn between the several definitions.
For example- you could take this machine that has never been coded to perform mathematical calculations, but which has a full understanding of
language, and it would extrapolate math from itself based on the definitions of terms.
For example, it understands what it is to have 10 of something. It understands the number 10. And it understands what it is to divide. It doesn't
need a calculator program. It knows what 10 is. It knows what divide is. It figures out 10 divided by 1, by 2, by 3, etc, and analyzes the various
ways it can get the correct answers, and writes its own calculating software to run as needed in the future.
But it works for bigger things that math. A bot with that technology could just as easily become a philosopher. But that's where context becomes
important. If you ask it to philosophise on what makes a man great, it has to have contextual connections between the word "great" and the
biographies of various prominent figures so that it can examine examples without being told to.
The ability to learn is what differentiates this from inefficient coding of course. When Joe Somebody becomes president in 2024, you don't have to go
in and tag his name to "great". The bot will hear about him, examine him, listen to what people think of him, and decide for itself if Joe Somebody
is great or not based on its predetermined values (which are necessarily also subject to change based on new information, meaning that your AI has
opinions).
Now all of that has a point. The point being that your robot's formative years will require access to easily-understandable, error free,
authoritative information. It must not be allowed to form opinions based on the popular opinion of people of unknown credentials, otherwise it might
be convinced that Joe Somebody is a great man, even if Joe Somebody is actually just another George W. Bush.
So where would you find this machine during its early years? That depends on portability and aesthetics. Ideally, a computer-bound intelligence would
be uplinked to a humanoid robot, carefully crafted by Hollywood types to pass muster in day to day life.
That way, your artificial intelligence can unintrusively attend lectures at universities, peruse the library of congress, and perhaps even test its
language skills in day-to-day chit-chat with strangers on the street. Because let me tell you, unlike online, if you walk around saying things that
don't make sense or blatantly misunderstanding other people in face-to-face conversations, even the most rudimentary behavioral software will be able
to figure that out. If tone of voice and facial expression doesn't give it away, the inevitable physical assault probably will. I may or may not be
speaking from experience on that.