NetNews Usenet Archive 1992 #19/NN_1992_19.iso/spool/comp/ai/3226

Text File | 1992-08-26 | 14.7 KB | 280 lines

Newsgroups: comp.ai
Path: sparky!uunet!decwrl!csus.edu!netcom.com!park
From: [e-mail] (Bill Park)
Subject: Re: AI Winter Refugees
Message-ID: <3pbn3![e-mail]>
Date: Wed, 26 Aug 92 13:26:30 GMT
Organization: Netcom - Online Communication Services  (408 241-9760 guest)
Summary: It was hype, greed, Reagan, Bush, and the Commies!
References: <x+an![e-mail]>
Followup-To: comp.ai
Keywords: AI Winter
Lines: 266

In article <x+an![e-mail]> [e-mail] (Steven Vere) writes:

>  I  am writing  an article on AI Winter Refugees:  un(der)employment
> among  AI professionals.

--Right on, brother!

>  I   would like  to  solicit information and
> opinions from the AI community on the following points:
>
> o what were the causes of AI Winter?

Chuck Williams of Inference Corp. places the beginning of the AI
Winter in the first quarter of 1987 (AI Expert Magazine, January 1992,
p. 34).  So it has lasted over five years --so far.

The reasons for AI Winter were frequently discussed at AAAI-92 and
IAAI-4 in San Jose, July 12-16, 1992.  Some of them most popular were:

    * Too many AI vendors chasing too few customers.

    * Hype, on a scale rivaling that of the claims for machine translation
      back in the 1950s.

    * Greed.  See "hype," above.

    * "Low-aspiration" "tactical" expert systems (i.e., anything that you
      could do on a PC in a few weeks) were nothing to crow about because:
    -- They often didn't work out (making YOUR boss look like a putz to
       HIS boss and forever after making AI a four-letter word in the
       whole company).
    -- If they did, you were lucky to see a return on investment
           of 2:1 or 3:1 from them
        -- They usually didn't go any further than the computer-friendly
           environment of the AI lab
        -- If they did, they could usually only be used at a few places
           in the organization (you only had one pretzel-bending
           machine in the whole factory that needed a diagnostic expert
           system for pretzel bending machines)
    -- Once it worked, the guy who developed it got transferred to
           the Oshkosh plant and no one could fix it when it screwed
           up ever again
        -- He didn't get the transfer, but they replaced the model 50
           electric pretzel-bender with the new gas-fired model 65,
           making the expert system totally useless because it always
       wanted you to change the fuses.

     * "High-aspiration" "strategic" AI systems would occasionally
        manage to get built, despite:

        -- The best efforts of hardware and software vendors
           everywhere to sell your company a complete line of
           incompatible products,

        -- The retirement -- or coronary, even -- of your "champion,"
           who was the one guy in the whole organization with the
           clout to keep pouring money down a rathole for three years
           of missed deadlines and incomprehensible dog-and-pony shows
           for the Board (mainly because this was going to be his
           last, flashy coup before retiring with a bang).

        -- Having to rewrite the AI team's mish mosh of Uncommon Lisp,
           Prolog in Forth, and object-disoriented C+- into nice,
           clean, self-documenting COBOL so that the MIS people would
           let you near their precious mainframe with it and you could
           finally get at the real gigarecord corporate database
           instead of the 150 "representative" records that you'ld
           been debugging with for the last three years.

      Six months later, having brought the entire MIS department down
      several times --once requiring a restore from the full backup
      tapes on Christmas Eve-- and with a contract out on your entire
      family from the MIS director, you finally get the damn thing
      on-line with real customer data pouring through it.  This has
      actually occurred enough times in the last 5-10 years for an
      important conclusion to have been drawn around the
      highly-polished tables in the boardrooms of America: When they
      work, strategic AI systems give you ROIs in the 10,000:1 range.
      Higher, some people claimed at the AAAI conference!  Why?
      Because they are the kind of system that is used at hundreds or
      even thousands of "seats" in an organization, not just on one
      pretzel-bender.  To field them, someone has to actually walk
      around the plant and talk to employees and discover how your
      company actually works.  They are thus a path to achievement of
      Total Quality, the corporate Zen Enlightenment of our age.  Just
      ask your local CEO if he'd like to be handed a Baldridge prize
      by the President of the United States.

    * Word got around the MIS departments of the land, and quite a few
      decided that they wanted in on this kind of high-visibility
      bacon-saving.  Pretty soon, high-priced AI consultants were
      finding they weren't so indispensable; the boys in the back room
      could handle everything now.  Client-server architectures were
      coming in, too, so you didn't have to make a $50,000 Lisp
      Machine pretend to be someone typing on a TSO terminal (and
      "reading" the screen, error messages included!!).  In fact,
      those 680X0 and 80X86 and RISC chips got so fast, you could
      actually run 20- or 30-megabyte AI programs quite nicely in
      them.  In fact, the MIS folks would rather you did, and let them
      get on with processing 1,000 tiny transactions per second.  And
      the rout of other GUIs by the free X Window software didn't
      hurt, either.  By then, you didn't need to advertise for Lisp
      Wizards any more; any old C/Unix hacker fresh out of school
      would do.  And many of them wore normal clothes.

    * In an effort to keep selling $50,000 software packages to large
      clients, the AI vendors pretty much all put up new signs saying
      "Software Engineering Systems Sold Here," and took the tattered
      old "AI, Take It or Leave It" sign out back and burned it.  They
      learned to say "Verification and Validation," though they still
      haven't figured out what those strange words mean.  They are
      probably French.  Of course, many AI shops tobogganed into
      Chapter 11 bankruptcy, or got bought by a chain of Indonesian
      Yak meat restaurants before they could clean up their acts.
      This further decreased the need for people who knew a car from a
      cdr, in favor of those who could relate to a data base or work a
      net.  People in the street began to nod wisely to each other and
      say, "A Little AI goes a Long Way."

    * Just when we were starting to get the boys in the banks and
      brokerage houses in New York to buy a few expert system shells
      and get used to searching around a screen the size of a
      cafeteria tray with twenty windows all scrolling madly at once,
      some damn fool had to spill the beans to them about neural
      networks.  Once they realized that, of all SIC codes, the money
      manipulation industry has the best historical data to train a
      net with, it was all over for knowledge-based systems.  Who
      wants to answer a lot of crazy embarassing questions about the
      possibility and plausibility values you estimated for that
      $10-million-dollar trade you took the wrong side of last week?
      Especially when the kid doesn't know a butterfly from a straddle
      and thinks soybeans are things you buy in the organic food coop
      in the Village.  Put a $10,000 neurocomputer board in a PC, hire
      some nerdy little guy to feed it all the numbers in the house,
      and you can retire to your own Greek island in a year.

    * By the end of the 1980s, realization began to grow that the
      Japanese 5th Generation Project was not going to be all that
      fatal a death blow to he occidental computing industry, after
      all.  In 1982, the U.S. was stampeded into setting up the MCC in
      Austin, Texas under Bobby Inman with a budget of $65
      million/year (of course, it took another two years for President
      Reagan to sign the bill allowing IBM to participate in saving
      the nation's bacon without being sued for antitrust violation!).
      Likewise, the Brits spun their science policy around 180 degrees
      and spent 350 million pounds on the Alvey Projects, after they
      had nearly strangled all their typically first-rate computer
      scientists in response to the Lighthill report at the end of the
      1960s.  The Europeans, in turn, spent about half as much on
      their ESPRIT projects.  By the end of the 1980's people were
      beginning to wonder when the promised breakthroughs from all
      this expensive hustle-bustle would begin to appear on store
      shelves.  At least we here in the land of the free (or at least,
      the marked down) can look forward to 1995, when the Cyc system
      will start reading by itself.  It will then rapidly complete its
      education about common-sense things without further human input
      (said Doug Lenat, during a panel session at the conference).  The
      question is, if Cyc really does develop common sense, will it be
      allowed to vote?

> o why has AI gone out of fashion in the US industrial sector?

   * Earnings are down, thanks to (1) the recession, and (2) the
     untimely departure of those awful commies that the Pentagon found
     so very useful for prying our tax dollars out of congress to
     scatter around their favorite military contactors for DIVADs,
     B-1s, Aguilas, and --oh boy, oh boy, oh boy!-- Star Warts!!  Used
     to be we'd have either a recession or an outbreak of peace that
     had to be put down, but not both at once, so we were always able
     to tighten our belts and squeeze through.  This time, we've got
     'em both at once and certain nonlinear effects are begining to be
     felt in our politico-econo-industrial institutions.  Like, a
     million layoffs is bound to get noticed sooner or later.

   * A manager's first duty when earnings are down is to fire someone.
     Not himself, that's unsanitary; someone under him.  A couple of
     months ago, during a panel at the expert systems SIG of the
     Software Entrepreneurs' Forum, some folks who used to sell expert
     systems to financial organizations said that the only thing the
     executives of those companies wanted to know was "How many people
     can I fire if I buy your software?"

   * In tough times, the best people to fire are people who don't do
     anything important and whom no one else likes.  Research staff
     fit these criteria.  Can 'em and nobody even knows they're gone,
     right?  In fact, a lot of people are glad to see those overpaid
     snotnose draft dodgers having to get out and find a real job for
     a change.  After they're gone, then you can cancel the
     ridiculously expensive support contracts for all those wierd
     computers they insisted on buying instead of using the company
     mainframe like the guys in payroll do.  Jeez, aren't PCs good
     enough for them?  That'll help stanch the haemorrage of cash in
     your department and maybe the V.P. will decide to close down old
     man Carpfutzer's area instead of yours.  Maybe.  With mortgage
     rates so low, it might be a good idea to refinance the house
     now, while you still have a job and can qualify.

> o stories of AI professionals or recent graduates who are or have been
>   unemployed or underemployed due to AI Winter;

I'm a 48-year-old Ph.D in control systems from U. of Penn, Dean's list
several times at MIT after skipping freshman math, physics and
chemistry --in other words, not as dumb as I look.  17 years of
experience in robots, expert systems, neural networks, and waving of
vu-graphs in clients' faces.  I've had one interview in 14 months of
unemployment, and I'm filling a scrapbook with shaft-O-grams from the
finest companies in Silicon Valley.

I've seen a lab full of talented roboticists in a local aerospace
company having to bootleg "overhead" time to work on their robot
system because there is no money for such work any more.

A top-notch roboticist in another aerospace company here is only
working half time and is just waiting for his hearty handshake.

And of course, you, Steve Vere, are the fellow who wrote the DEVISER
program, a landmark in automatic planning, which is going to be such a
big industry that the expert systems boom will look like the great
Edsel buying frenzy.  Are you working?

> o alternative career paths followed by AI Winter refugees.

A friend of mine of similar advanced years who knows object-oriented
data bases inside out has had to take a job as a sales clerk in a
hardware store.

>    Countervaling views  are welcome, but please be prepared to explain
> the following phenomena: near-zero employment opportunities advertised
> in AI Magazine and similar forums; near-zero recruiting at recent AAAI
> conferences; major  declines   in  attendance at AAAI   over the  past
> several years.

Another friend of mine has started a business that arranges for good
programmers in India to write your software at a far lower cost than
U.S. programmers would work for.  Like the auto manufacturers, we are
faced with unbeatable price competition from Asia.  Protectionism is
doubly ludicrous when your are talking about bits entering the country
through a satellite dish on someone's roof.

A triple ludicrousy: The engineers we do have are wandering the
streets in a daze, out of work, yet the NSF is predicting a shortage
of engineers (yes, they're doing it again!!), and both political
parties want to improve the education system so we can turn out more
educated people.  I ask you, friends, is the invisible hand of
economics playing pocket pool?

The capital formation system in the U.S. dictates that managers ignore
long-term strategic planning for short-term stabilization of earnings
growth.  Otherwise the stockholders sue them.  Big Japanese companies
don't have to operate under this constaint as much, since each
Zaibatsu has its own bank, resorting to a lawyer is a disgraceful sign
that you have failed to manage your own affairs well, and Japan has a
critical shortage of lawyers (can't we help them out by sharing a few
boatloads of our bountiful surplus with them?).

>   Steven Vere                                    [e-mail]
>   Boulder Creek, California                               [e-mail]

Bill Park, Ph. D., President, CEO, CFO, COO, CIO
William T. Park & Associates    (me and the cat)
Computer-Aided Quality        (me and the Macintosh)
Park Research Institute        (me and the cat and the Macintosh)

An equal oportunity employer.  We do not discriminate on species.
--
Grandpaw Bill's High Technology Consulting & Live Bait, Inc.