Path: utzoo!utgpu!jarvis.csri.toronto.edu!cs.utexas.edu!usc!apple!limbo!taylor
From: eugene@eos.arc.nasa.gov (Eugene Miya)
Newsgroups: comp.society
Subject: Re: Looking Backwards
Message-ID: <321@limbo.Intuitive.Com>
Date: 11 Jan 90 16:59:44 GMT
Sender: taylor@limbo.Intuitive.Com
Organization: NASA Ames Research Center, Calif.
Lines: 63
Approved: taylor@Limbo.Intuitive.Com

Being with the Government, we engage in exercises like this all the
time.  So we get called in for opinions by other Agencies (some times
in the Black) and various companies (computer).  So my comments are
neither complete nor will I say any inside information.  Planning.

In predicting trends, you want to be able to deduce and justify
your forecasts.  The more quantitative, the better.  Seek those
areas which have refined methods of measurement like electronics
and physics.  This is why wafer fabrication went so far: controlled
fine measurements.  Unfortunately, communication is more vague.

> The keyboard will go the way of the card reader.  Voice-and-pointer
> will be standard; the pointer may be a dataglove or merely a camera
> pointed at your hand.  You have to have something to do with that
> 100 mips, after all.

I don't quite think so.  Neck ties will disappear first. 8)
A good question is whether the pocketbook size machine will have
a keyboard?  Does to make sense to build a device which a human
can't easily operate?  Yes and no.

> Software for language comprehension will emerge from a synthesis of
> spelling and grammar correctors, OCR for scanned text input, and
> talkwriters.

I use an OCR weekly.  The AI people predicted this in the 1950s.
We are still waiting.  Find an OCR and observe all the errors
they make, you will get laugh.  I do not know if you have ever seen a
typewriter without a "One" key "1," you used an "l" "ell" key.
Even neural nets won't help, just more of the same.  We have to
learn to store more/better context.

>   *Robots recognizing policemen*?

I doubt this.  An attempt was made at the end of the 1960s in Vietnam to
design a technological fence called the McNamera Line.  Designed to prevent
enemy infiltration from N to S Vietnam.  Again the AI recognition problem
is a lot harder, and I do not see a solution in 10 years.  Can you use
AI to tell a good Panamian from a baddie?

> .. sometime in the 90's the information available in electronic form 
> will catch and exceed that available on paper...

There is a difference between information and knowledge versus data.
But I believe I will have more of both, and a lot of junk as well.
My NeXT box shows me this.  I have "thrown away" my hardcopy Websters.

> * Voice, text, and CAD systems will be pointed to both by AI researchers
> * and their critics as supporting their positions.

It is time, critically, to establish firm minimum baselines of performance.
Vague notions like the "Turing testing" will be moving targets firing at
other moving targets (technology).  Start with qualitative constraints.
Then move quantitative.  Beware that things do not scale linearly.

If we don't do the above, then when the Emperor walks by with his fine
new robes (or Mind, if you have read Penrose recently), then you
will clap with the rest of them.  I hope by that time, I am still
a little boy, "Er excuse me, but ..."

Science is about skepticism.

						--eugene miya