From: Stephen Reed (reed@cyc.com)
Date: Tue Jul 09 2002 - 12:00:09 MDT
On Tue, 9 Jul 2002, Mike & Donna Deering wrote:
> I think not many people appreciate the difficulty of gathering the
> information necessary to get an AI to beyond human intelligence. I've
> heard some say just hook it up to the internet and others claim give it
> a lot of books to read. To begin with, there is a lot of human
> information not on the internet or behind firewalls only available by
> subscription. I have run into this problem many times in my own
> research finding that there is more depth of information at my small
> neighborhood library than freely available on the internet.
My experience has been the opposite of yours. I have lived in Austin now
almost four years and not have not visited any public or university
libraries. I buy many more books online than I find in bookstores, but
the great majority of useful information I find online - more and more all
the time. Google / Google groups / and Citeseer are my favorite info
portals.
> I imagine
> one of the first things you might want to do, once you've got your AI
> functioning, is to input all human knowledge and have it cross
> correlate it and see if we missed anything obvious. That first part,
> "input all human knowledge" is easy to say, incredibly hard expensive
> time consuming and labor intensive to do. As far as technology, just
> use patent applications, right? Send an e-mail to the US patent
> office: "Please sent me a copy of everything you've got. If there is a
> charge, here is my credit card number XXXX XXXX XXXX 1234." When the
> convoy of semi tractor trailer trucks start arriving and the bank is on
> the phone you might see the problem.
Perhaps you should take a community college course on how to use the
internet. Even an old guy like me can learn how. It looks to me that you
are really missing out on what there is to offer...
Here is a sample online patent application, rich with commonsense
background knowledge about bicycle mechanics.
> What about all the textbooks used
> in all the courses at MIT? And subjects MIT doesn't teach get from
> other universities. Still a lot of books and a lot of gaps. I'm not
> saying it will be impossible, just extremely difficult and very under
> estimated by those making predictions.
A surprising number of universities are publishing full courses and
textbooks online. I have an extensive and growing set of bookmarks for
the time (hopefully) when Cyc can read. But I believe that PhD thesis are
an even better source of knowledge for any topic they cover. I have a set
of bookmarks for online PhD thesis (found via web search and Citeseer) -
Chemistry, Physics, Linguistics, Computer Science, Philosophy, Electronics
and much more.
See for example the link site:
http://spot.colorado.edu/~dubin/bookmarks/b/1240.html
Or suppose you want just medical knowledge:
http://www.emedicine.com/
So I hold an opinion the opposite of yours.
Regarding the availability of online reading material for a Seed AI, I
believe that getting an AI to read, understanding what is read,
integrating new, possibly contradictory knowledge - and competently
answering questions about it - is far, far harder than finding useful
reading material.
-Steve
--
===========================================================
Stephen L. Reed phone: 512.342.4036
Cycorp, Suite 100 fax: 512.342.4040
3721 Executive Center Drive email: reed@cyc.com
Austin, TX 78731 web: http://www.cyc.com
download OpenCyc at http://www.opencyc.org
===========================================================
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:40 MDT