Stephen,

On 4/10/08, Stephen Reed <[EMAIL PROTECTED]> wrote:
>
>  ----- Original Message ----
>  From: Steve Richfield <[EMAIL PROTECTED]>
> To: [email protected]
> Sent: Thursday, April 10, 2008 2:58:09 PM
> Subject: [agi] Comments from a lurker...
>
>
> [snip] BTW, the principles behind Dr. Eliza are rather unique. I'd be glad
> to send some papers to anyone who is interested. Briefly, Joe Weisenbaum's
> original Eliza was built on two concepts, one good and one bad, that no one
> previously separated. The good concept was that individual links in complex
> cause and effect chains could be recognized by the occurrence of slightly
> variable but easily describable snippets of text/speech. The bad concept was
> that text/speech could be usefully manipulated by juggling words around. Joe
> then wrote a book discrediting his own Eliza (with its unseparated
> concepts), thereby causing AI research to take a wrong turn 40 years ago
> from which it never recovered. However, the internals of Dr. Eliza aren't
> really the subject of this posting, other than to demonstrate that AGI now
> already exists, at least in this one potentially useful form.
>
> Any thoughts on all this?
>
>
> Hi Steve, thanks for speaking up.  Is what you call Dr. Eliza, the same
> software program as the well known Eliza<http://en.wikipedia.org/wiki/ELIZA>
> ?
>

Not at all. Then, why the name Dr. Eliza? Our "guiding light" was: Suppose
that the wrong turn 40 years ago had NOT been made. What would Eliza be now,
what with the Internet, modern languages and coding methods, relational
databases, dropping the bad half of the original Eliza and substituting an
engine to reconstruct cause-and-effect chains, etc., etc.? Add to that, our
first strong demo was finding cures for real-world chronic illnesses.

The first demo was written in Access with a generous sprinkling of SQL, but
we then yanked the computationally intensive part out and compiled it under
Visual Basic when we married Dragon NaturallySpeaking to it, for a real-time
conversational speech interface. BTW, speech recognition works GREAT because
it reliably recognizes long words and specific phrases, even when it messes
up everything else. However, people with chronic illnesses wanted nothing to
do with it, as they were unwilling to concede ANY chance of a speech
recognition error that might cause Dr. Eliza to miss a cure for their
illness.

   In any case I would like to see example transcripts of the doctor-beating
> medical dialog you mention.  As perhaps you know, I am building an English
> dialog system to bootstrap AGI, and if successful in realizing its design,
> Texai should easily win the Chatterbot 
> Challenge<http://www.chatterboxchallenge.com/>(some year) and also perhaps 
> the Loebner
> Prize <http://en.wikipedia.org/wiki/Loebner_prize> (some year).  I studied
> existing chatbots <http://en.wikipedia.org/wiki/Chatbot> of course.  If
> you are familiar with 
> A.L.I.C.E<http://en.wikipedia.org/wiki/Artificial_Linguistic_Internet_Computer_Entity>.,
> I would be interested to learn your opinion on how it compares with Dr.
> Eliza.
>

Dr. Eliza is NOT intended to be a chatterbot and will never win any of those
prizes. It rather stubbornly IGNORES where you might quite intentionally be
leading the "conversation" and persistently asks questions that seem to be
quite irrelevant to the problem that you have presented. Of course, the
answers to those questions are EXACTLY what is needed to sort out the
various possibilities of links in the cause and effect chain.

Note that real-world problem cause and effect chains usually look like a
figure 6, with a root cause, several subsequent links, and a self-sustaining
loop at the end. To completely cure a (medical, electronic, political,
personal, etc.) problem, you must attack the closest link possible to the
root cause AND you must interrupt the self-sustaining loop at ANY point.

Steve Richfield

-------------------------------------------
agi
Archives: http://www.listbox.com/member/archive/303/=now
RSS Feed: http://www.listbox.com/member/archive/rss/303/
Modify Your Subscription: 
http://www.listbox.com/member/?member_id=8660244&id_secret=98558129-0bdb63
Powered by Listbox: http://www.listbox.com

Reply via email to