::scr IA Goldrush (was Ramblings of a Classic Refugee)

Arvid Gidhagen scr@thegestalt.org
Tue, 12 Mar 2002 17:34:07 +0100 (MET)


> Or is it a cultural thing? Once human language enabled machines are
> everywhere, will it become customary to announce, Star Trek style,
> "Computer, bring me my lunch"?

I don't think that will be much of a problem. People are quick to
adapt if they feel there's anything to gain. It didn't take long
for the public to change their attitude to cellphones from
"stupid, unnecessary, annoying toys for yuppies" to a transparent
necessity.

> I think part of the problem is that you can get a long way in creating a
> speaking computer without needing that much actual intelligence. Languages
> are rule-based, however complex the rule set may be, and as such are
> translatable to the machines that we have today, provided we spend enough
> time defining the rules.

This is certainly not an uncontroversial statement. While you can create
a SPEAKING computer without endowing it with much intelligence, a
COMMUNICATING computer has to be incredibly smart. Languages are so
horribly complex and dependent on context and social factors that
noone really has come up with anything even resembling a grand
unifying theory of language and language use.

Take a simple example: What does the word "black" mean? It's not
enough to look it up in a dictionary, because "black" means different
things to different people, in different situations, about different
objects, and so on. It can refer to the actual quality of a physical
object, or it can be used as a metaphor, or an almost unlimited
amount of other phenomena.

See this link for a nice and reasonably short description of situated
cognition:

http://www.ed.uiuc.edu/EPS/PES-yearbook/94_docs/BREDO.HTM


> A computer with a large dictionary and a reasonable set of syllable sounds
> can say a lot, possibly translate a lot back into text too. But give it a
> word like fluctuate which it doesn't have, and it's unlikely to intuit its
> meaning from the fact that it appears related to "flux", which is in the
> dictionary.

Neither is a human, I would think...


- Arvid