[Top] [All Lists]

[ontolog-forum] Toward Human-Level Artificial Intelligence

To: "'[ontolog-forum] '" <ontolog-forum@xxxxxxxxxxxxxxxx>
From: John F Sowa <sowa@xxxxxxxxxxx>
Date: Wed, 23 Apr 2014 09:47:33 -0400
Message-id: <5357C475.10506@xxxxxxxxxxx>
Phil Jackson's thesis defense was successful, and he is now Dr. Phil.
Following is a 27-slide intro:    (01)

http://www.philjackson.prohosting.com/TowardHumanLevelAI20140422.pdf    (02)

The last slide has a pointer to the full dissertation:    (03)

http://www.philjackson.prohosting.com/PCJacksonPhDThesis20140422.pdf    (04)

I recommend both.  The intro summarizes the arguments for the
controversial claim that human-level AI is possible with a knowledge
representation based on natural language.    (05)

That hypothesis sounds like heresy to anybody who has been using
logic-based notations.  But it's not implausible -- especially since
we do have at least one example of a successful implementation: people.    (06)

It's also consistent with a claim I frequently make:  every version
of logic or other artificial language is a generalization from the
syntactic and semantic features of NLs.  I analyze that point in
http://www.jfsowa.com/pubs/rolelog.pdf    (07)

I also show how various logics (EGs, CGs, Kamp's DRS and others) are
related to NLs and to each other:  http://www.jfsowa.com/pubs/eg2cg.pdf    (08)

Phil's notation, Tala, is similar to DRS, EGs, and CGs in that it has
a fairly direct mapping to NLs and it uses symbols that have a closer
mapping to NL words and phrases than to URIs.    (09)

In fact, this is a point I have often made:  URIs are useful for many
purposes, but they do *not* facilitate natural language understanding.
It is *impossible* for anyone -- human or computer -- to relate a term
from an NL text to a URI that adequately represents that term until
*after* they have analyzed and understood the NL text.    (010)

I admit that a document that contains annotations (with URIs or other
identifiers) may be helpful for NL processing.  But annotations based
on a well-designed terminology for humans would be as good or better --
especially since any humans who do the annotating depend on the humanly
readable terminology, not on the formalism.  No identifier of any kind
can be more reliable than the system that generates it.    (011)

However, I do have some criticisms or at least caveats.  Phil does not
devote enough attention to the importance of mental models or imagery.
Many animals without language have an impressive level of intelligence.
Scientists are just beginning to appreciate their intelligence.    (012)

I believe that a system without imagery can achieve a much greater level
of intelligence than current AI systems.  But humans cannot -- just by
reading -- fully understand other humans who have gone somewhere and
done something that is very different from anything they have ever
experienced.  Neither can a computer.    (013)

I also recommend the dissertation as a good reference.  It has a 26-page
bibliography with URLs for many of the documents.  It also has useful
one or two-page summaries with citations for further reading.    (014)

John    (015)

Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/  
Config Subscr: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/  
Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
Shared Files: http://ontolog.cim3.net/file/
Community Wiki: http://ontolog.cim3.net/wiki/ 
To join: http://ontolog.cim3.net/cgi-bin/wiki.pl?WikiHomePage#nid1J    (016)

<Prev in Thread] Current Thread [Next in Thread>