A couple of weeks ago the tech news, at least the online Mac version,
was full of reports of Macintosh pioneer Jef Raskin's complaining
about Mac OS X, that there's no difference between it and Windows XP
and how what the computing world really needs is a big does of
Artificial Intelligence so users can say
do what I mean and
the computer will figure out what the heck that is. I thought it was
nonsense at the time, a no-longer-listened-to prophet getting old and
cranky. And having spent a bunch of years around Artificial
Intelligence, I was a lot less confident than Jef that a computer's
idea of
DWIM would look anything like the user's.
A small example is the SB-600 flash I got for my Nikon D70. It offers
two automatic modes when coupled to the D70. The cool mode is called
i-TTL Balanced; it uses the camera to determine just how much light to
apply to the scene, balancing the need to light the subject of the
picture as well as the background. Then there's plain old i-TTL; it
lights the subject and lets the background fend for itself. And one
guide to the camera warns you not to use Balanced mode, because
you won't be able to predict the results. By trying to light
both subject and background, you may end up with both lit poorly.
Whereas just doing the subject, you can be pretty confident that it
will be lit correctly.
And this is the problem with AI solutions to normal problems: that
what you gain in the quality of the solution (assuming the solution is
actually good), you lose in predictability. And for humans, getting
the answer we expect is more important.
What made me think of all this today is an article on
As The Apple
Turns, my favorite humorous Apple news site, that talks about the
surprising product suggestions you can get from Amazon's own brand of
AI. To wit, add Mac OS X 10.3 Panther to your wish list and
you'll get a suggestion of Pee Wee's Playhouse Christmas
Special. Ask for the upcoming Mac OS X 10.4 Tiger
instead and you'll be told about The Tripleets of
Belleville. And the magazine suggestions are even better:
interest in Panther makes you a target reader for The New
Yorker, while Tiger buyers are candidates for
Playboy.
All of which suggests something I figured out around 1989: this AI
stuff is fun, but I sure wouldn't stake anything important on it.
(Too bad I got started in it in '83. But some of us take a while to
discover important truths. And some of us never do.)