Grokbase Groups Perl ai August 2001
FAQ
Hi Kevin,

I think you have posed an excellent question, "WHAT TO DO?" For some time,
I was thinking that the key would be developing a 'database' that modelled
the world like we do, so that we would have something to talk to. The
computer would be able to parse our language because it would have a
corresponding model of the world to parse it into...

While sleeping on the road (literally) somewhere in the Southwestern United
States on my way to seek AI in San Francisco, an epifany occured to me. The
database is important, certainly, but something else is even more important.

We need to focus more on the 'overhead' part of conversations, which is used
to establish and maintain context, and verify that accurate communication
has occured. This will help a lot in keeping the thread of a conversation
going. As to what to do? If we take the model of how a computer works, we
find that a computer, appearances aside, is doing something all the time.
There is always the idle loop, checking for other processes requesting time.
The idle loop in a conversation is that 'idle gossip' stage where identities
and context are established. More than this, though, 'idle' chit chat
allows the two communicating entities to hone and polish their models of the
world (or at least the way in which language is commonly mapped into it).
Various key sentences or questions can act as triggers to go off into other
loops which fulfil requests for information or to perform tasks, but once
these tasks are verified and fullfilled, then they fall back into the basic
idle loop. Determining the effective timing and language of this 'chit
chat' is an area where psychologists, philosphers and flim flam men may
well come in handy... (Maybe even former German teachers...)

As I work up a fuller and more formal write up on this, I will post it on a
common web-site and announce the location to everyone.

Regards :) Chris

----- Original Message -----
From: Kevin Watt <kevin@allpoetry.com>
To: <perl-ai@perl.org>
Sent: Wednesday, August 22, 2001 3:38 PM
Subject: facts & RE: Project Earth

Hi All,
From what I've seen, our unstepped stone in AI isn't formulating a
believable network of facts - such systems as wordnet and framenet work as
great categorical placement of words + their meanings, and there are many
great semantic parsers out there (Automatic Labeling of Semantic Roles,
Gildea / Jurafsky), but the hardest part is deciding WHAT TO DO - imposing
some will upon a system, past even just a list of projects. To have IT
start a conversation, and lead it, rather than just reacting as so many
chatterbots can do / fake intelligence with today.

Ciao,
Kevin
--------------
Kevin Watt, AllPoetry.com Community manager: Poets Unite!
kevin@allpoetry.com <http://allPoetry.com>
"Here, write it, or it will be erased by the wind." - Isabel Allende

-----Original Message-----
From: steve@ladon.host4u.net On Behalf Of
Steve Vertigan
Sent: Wednesday, August 22, 2001 1:57 AM
To: perl-ai@perl.org
Subject: Re: Project Earth


Lee Goddard wrote:
The objective of Project Earth is to build a database of factual
information
relevant to any aspect of life by collating voluntary contributions
made
by
project participants.
Does anyone remember the name of the project that tried to
program a database with the equivelant knowledge of a three
year-old child? It's still running, after I think 30 years.
Did that project utilise a world-wide network of contributors like this
one aims to do or did it involve a limited team? Of course then you
have the problem of people entering junk data in the database. I
suppose one solution would be to have a review system like the PGP web
of trust where a given fact could be trusted in accordance with how many
people have confirmed it and how trusted a given person is considered to
be (probably has a neat parallel with how humans choose to believe
something).

But the main problem with this approach in general IMHO is surely the
best you can hope for is a sophisticated database searching program that
can tell you $string is/= $otherstring. By itself it wouldn't even have
much hope of passing the turing test let alone be arguably
'intelligent'. Or am I missing something?

Regards,
Steve

--
OpenBSD maelstrom.dyn.dhs.org GENERIC#399 i386
12:50AM up 1 day, 16:44, 1 user, load averages: 0.61, 0.64, 0.64
God may be subtle, but He isn't plain mean.
-- Albert Einstein

Search Discussions

Discussion Posts

Previous

Related Discussions

Discussion Navigation
viewthread | post
posts ‹ prev | 8 of 8 | next ›
Discussion Overview
groupai @
categoriesperl
postedAug 22, '01 at 8:30p
activeSep 10, '01 at 7:01p
posts8
users6
websiteperl.org

People

Translate

site design / logo © 2021 Grokbase