[SIGCIS-Members] details sought on Paul Werbos work on Vietnam for RAND

[Bernard Geoghegan]

Dear colleagues with RAND and Vietnam War expertise!

In an interview in the book *Talking Nets*, Paul Werbos, an earlier
developer of backpropagation algorithms, tells a story (excerpted below)
about a project he worked on for RAND that got leaked to the New York
Times. I've emailed with Werbos, but he doesn't seem to have a handy
citation for the NY Times report. Does anyone happen to know the report he
has in mind or tips on how to locate the report or a gloss on his work for
RAND? There are a few comments in Boden's two volume study of AI, but it
appears her reconstruction is imprecise. Alternately, I'm not planning a
RAND archives deep-dive or site visit, but maybe someone has advice for a
well targeted remote search request I could file? This is just a little
footnote for a larger project, so I'm trying to not get too sidetracked.

Werbos reports: “I went back to Harvard to get a PhiD. in applied math.
Having been fortified with humanity from London, I then descended into the
bowels of the machine , stopping off at the RAND Corporation for the summer
of 1968 along the way....Actually, I did learn a thing that summer that
surprised me. It turned out one of their problems was coming up with a
measure of success in the Vietnam War....I looked for stable invariants ,
and they were things like Vietcong attacks on Americans. That was the best
stable invariant underlying measure. I used that as a success measure and
came to the conclusion that we' should radically change our policy and do
things like small unit actions instead of these large sweeps...Of course ,
it was all classified , and I read it on the front page of the New York
Times two weeks later, which immediately made me very cynical about
American security. The minute it’s useful , it’s on the front page of the
New York Times. That made me very cynical- that event plus a few other
things that happened at the RAND Corporation.”


Best wishes, Bernard
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.sigcis.org/pipermail/members-sigcis.org/attachments/20260330/1174b5a9/attachment.htm>