It seems to me, an old man with a lot of rudimentary computing experience that the current hulla-balloo about Artificial Intelligence, it's promise, is either polly-ann-ish, or outright deceptive. Let us discuss for a minute the public perception in my close-by neighborhood.
The
generalized fear is that AI will replace people. End of story. These
machines will somehow become self actuating and self replicating.
Ultimately they will do whatever we human types do and we will be out
of a job; whatever that particular job is. So we better get busy
finding a new line of employment before the AI figures out what we
are doing and gets there before us.
The much brighter vision
is that AI will begin doing the menial work we loathe. AI will do the
cleaning and the cooking, begin doing the hard tasks, the hot work,
and the danger will go out of our lives. Nothing but improvement for
humanity and the only people effected are those who do the menial
labor, the hot work, and the dangerous stuff; you know . . .
immigrants. Those people need to find better work or become useless
and superfluous back in their old countries.
Of course there's
a spectrum of possibilities between these two and the AI hub-bub will
most certainly bleed over the outside edges of these two visions as
well. But the fact of the matter is that AI cannot do any of these
things, and likely won't be able to do much of this for many years,
likely decades, possibly a century or so. Despite the hype, this
isn't happening. Let me explain.
I have been in the computing
world for many decades. I did manual labor directly for Steve Jobs.
My Step mother had Wosniak over to the house before his name became
well known. I spent my early teen years punching card stock on a big
machine to program a much bigger machine that couldn't compete for
computing power with a modern blood pressure monitor. I built my own
computers through the 8086 chipset days, with their 5.5 inch floppies
and 256K of storage and continued to engage in this until it just got
cheaper to buy a laptop which did everything I ever dreamed of. I
knew personally the guy who helped create the MP4 format and waited
patiently for the 1982 promise of streaming video and broadband
internet to deliver the movies in 2012. I adopted cellular telephony
as soon as I could afford it and bought a smart phone as soon as they
got into the price range I desired. I witnessed the entirety of the
Internet's birth while in the military and watched it grow for almost
fifty years. I paid attention to this extraordinary period of
informational growth and access with curiosity and took the good with
the bad. Not once was I filled with overwhelming wonder because I saw
the actual bones of the baby before the flesh got added; saw the
bubbles bursting even as the pundits huffed and puffed. So much
soap.
And AI, for all it's promise, is just another thing of
this fictional genre. Possible, but not probable. Certainly not
practical. All the bubbles people seem willing to invest and believe
in will be found to be more bubbles in the making once the current
flotilla fails to fly. And here's why it will sink.
At the
root of computing is the idea of data. Computers keep this in files.
Files are simply numerical things filled with organizations of
letters and numbers designed to keep things where things go for use
when they are needed. Computers access these things when ask, develop
these things as instructed, and output whatever the result is when
the result is found. Back in the old days (back when Luddites worried
about computers taking their jobs) much of what went into the
machines was trash, so when we ask it to tell us something we didn't
know we got trashy answers. Things have become better since then.
Faster computers, loads of data,
billions of lines of code. If you hadn't seen the beginning you
wouldn't have reason to suspect that what these wondrous things do is
just the same as they have always done, only faster. The machines are
much smaller, and they do so much more and, strangely, unemployment
decreased because we still needed people to put the data in, tell the
machine what to do, and read the results. Result which are likely
going to need correction because the inputs were trash in the first
place.
This is not to say the machines are trash, they are
not, but the paperless promise cut down a lot of trees and the
wireless promise installed a lot of wires. If you were to guess how
much wrong information, both accidental and intentionally included
was contained in the database we call the Internet you might have to
say about half of what we know is suspect and all of it will take a
human with an education to know the difference. If you were to ask an
AI, the best ever created, to tell you how much of the data is false
it would wait for more instructions. It would need to know what the
scope of the project is, where to find the data, what does false
mean, and then give you an answer which is likely needing analysis
and correction by human minds.
AI is a machine trick incapable
of making a moral choice and having not stake in the decision. It
might someday be able to follow a well defined set of instructions
and procedures for making a pseudo moral choice and might one day do
this so as to be thought of as good at it. But humanity still has no
such set of moral definitions to work from which might allow the
programmers to commence work writing code. Eventually these machine
choices will fail and be amended to call for other programs and
fuzzier statistical manipulators which check the work and seem to
decide better. But these machine will only give you what you gave
them and do with it what they were told to do with ever increasing
specificity. Don't expect this to happen any decade coming soon to
your lifetime.
AI has limits as to what it possibly might do.
It might become a light switch which knows when to turn on the lights
in many situations. But the switch will still need to be given
instructions to work the way the user wishes it would; instructions
which might take as much time as flipping on the lights a thousand
times. There will be many thousands of man hours dedicated to making
this basic system of light switching to actually perform as expected.
The promise of the intelligent light switch, if we ever get
around to it, is potentially achieved in the efficient replication of
the invention which lowers costs, (which will take a bit of time to
be set up before promises come true). Today all you need do is walk
over to the wall and flip the switch and this might always be the
case whether they build smart switches or not; but saving humanity a
few hundred thousand hours and countless piles of treasure takes a
different moral choice.
Automobiles and airplanes might
someday move themselves with speed and accuracy, but they will get to
do this because thousands of people made millions of moral choices
and hypothetical questions were eventually answered to our collective
agreements. Until then self-flying planes will crash, self driving
cars will find themselves driving in circles or into rivers, both
economic decisions. The tragedies will continue and moral decisions
about the effective usefulness of the machine will be sorted out by
legislators and invisible market based hands. Until then employment
of drivers and pilots will continue. The AI promise un-kept for the
time being despite the hype.
Today we are being ask to allow
the building of thousands of expensive data centers where the promise
of AI is said will be found. We are ask to invest in the bright new
future of AI light switches and talking bathroom mirrors, none of
which we need, and most of which will fail on design or marketing
made by people capable of doing what AI can never really do: make a
moral choice which will hold up in the real world of life, legality,
and longevity. But it isn't going to happen they way they sell it
today. It might happen some day, but not how they planned it. And we
don't have to allow these hypothetically driven dreamers to gamble
the life we have on their promise of the life we someday might.
