Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!uunet!seismo!sundc!pitstop!sun!quintus!ok
From: ok@quintus.UUCP (Richard A. O'Keefe)
Newsgroups: comp.ai
Subject: Re: Who owns the output of an AI?
Message-ID: <181@cresswell.quintus.UUCP>
Date: Wed, 11-Nov-87 20:35:34 EST
Article-I.D.: cresswel.181
Posted: Wed Nov 11 20:35:34 1987
Date-Received: Sat, 14-Nov-87 04:51:30 EST
References: <1778@svax.cs.cornell.edu>
Organization: Quintus Computer Systems, Mountain View, CA
Lines: 39
Summary: yes it is fair

In article <1778@svax.cs.cornell.edu>, houpt@svax.cs.cornell.edu
(Chuck Houpt) writes that the new British intellectual property law will
rule that the output of an AI program will belong to the person running
the program, not the person who wrote the program.

Yes, this is fair.  The basic reason that it is fair is that it is much
easier to discover who ran a program than to discover who "wrote" it.
Suppose H1 makes a useful discovery using program P1.
	P1 is written using an knowledge representation system P2.
	P2 is written using object-oriented extension to Prolog P3.
	P3 is running in Prolog system P4.
	P4 is running on UNIX.
In a typical AI system, all the programs P1..P4 are physically present
in a single memory image, and to some extent all are active.  Let's face
it, the output of the interesting result might be critically dependent
on the correct behaviour of printf(). Why shouldn't AT&T get a slice?
Why shouldn't the typist who filed the purchase order for the program
get a slice?  The program is almost certain to rely heavily on ideas
that were published in the open literature.  Shouldn't Knuth & Tarjan &
the rest get a slice of almost everything?  You have to draw the line
*somewhere*.

One of the most important requirements about the law is that people
should be able to find out what it is and take it into account in their
planning.  Any rule which assigned a share of the rights to a discovery
to the authors of (say) the underlying Prolog system would be so hard
to operate that it would probably end up being a "whoever can hire the
best lawyers has ownership" rule.

Any software vendor who is unhappy about this will probably be able to
do something about it.  For example, instead of selling P1 to H1, the
author(s) of P1 might sell H1 the right to operate P1 as their agent.
If you can sign a contract with an employer granting the employer the
rights to your inventions, you should be able to sign a contract with
a software vendor granting a share of your rights.

And think about the other side of it.  Do the authors of AI programs
really want *liability* for everything their programs might come up
with?  Fair is fair both ways around.