[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: starship-design: Suspended Animation


Once upon a time I used to provide a great deal of computer hardware to
Lockheed Space Operation's AI research unit. One of my best friends from
school runs the operation and I can tell you that AI is a lot more difficult
than commonly believed. The hardware is really the smallest part of the
problem. Software is the real killer.

Software development has always lagged behind hardware development and
always will. It is not necessary to even develop an aware AI, simply one
that can be depended upon to perform its job reliably time after time after
time, no matter what input it is presented with. At the moment it is too
easy to crash AI software by presenting it with input it was not programmed
to handle. In a lab this is inconvenient, on a space craft, it could be

Nevertheless, we have made great strides in recent years and I fill
confident that AI will play some role, even if it is not as great as you


> -----Original Message-----
> From: owner-starship-design@lists.uoregon.edu
> [mailto:owner-starship-design@lists.uoregon.edu]On Behalf Of
> Antonio C T
> Rocha
> Sent: Tuesday, May 12, 1998 11:49 AM
> To: starship-design@lists.uoregon.edu
> Subject: Re: starship-design: Suspended Animation
> Kelly,
>     Ship repair is an "easy" task for neural nets. You have a
> "target" state and
> the possible "paths" (methods, tools, etc.) to get there.
> "experience" databases
> and expert system AIs pare the alternatives down giving
> "weights" for each
> procedure - building the branches of your logic tree(s). It
> is a big problem
> divided into a lot of simple parts.
>     Something similar happens today with IBM and other
> companies where the
> technician reads troubleshooting and repair procedure
> step-by-step from a screen /
> pad / whatever, being shunted to "higher-level" servers when
> rarer problems are
> encountered. At the top end end of this line, when a solution
> has not been found,
> lies top-notch human assistance. Aircraft mechanics work a
> lot like that.
> Auto-company car mechanics work like that.
>     The problem today isnt "logic" or problem-solving. The
> problem is real-world
> sensing.
>     The human technicians main job in this case is to provide
> real-world sensing
> and "common-sense". This is stiil too costly or incipient to
> put machines on it in
> everyday chores. Prototypes are being developed-used for
> nuclear plant and
> electric utility plant special jobs in very hazardous
> environments. But we already
> have seeing robots, flying-object catching robots, walking
> robots, best path
> choosing robots, face-fingerprint-handwriting recognition
> systems... etc. - and
> "common-sense" systems are coming along quite nicely.
>     By 2010 they will have become useful / cheap enough for
> supply to fuel demand
> to fuel supply to fuel demand... and so on.
>     Even in emergencies, the multiplying effect of AI, robots
> etc. will ensure
> that you will be able to do without the vast amounts of
> people needed today to
> operate and maintain fleet. They will also buy you
> buffer-time, doing the
> grunt-work and basic troubleshooting routines, enabling the
> humans to concentrate
> on the problem instead of "chasing consequences" when
> something goes wrong.