Path: utzoo!utgpu!water!watmath!clyde!bellcore!faline!thumper!ulysses!andante!mit-eddie!uw-beaver!cornell!rochester!ur-tut!sunybcs!sher
From: sher@sunybcs.UUCP
Newsgroups: comp.ai
Subject: Re: AIList V6 #86 - Philosophy
Message-ID: <11191@sunybcs.UUCP>
Date: 13 May 88 00:57:36 GMT
References: <1579@pt.cs.cmu.edu> <3200016@uiucdcsm> <523@wsccs.UUCP>
Sender: nobody@sunybcs.UUCP
Reply-To: sher@wolf.UUCP (David Sher)
Organization: SUNY/Buffalo Computer Science
Lines: 63
Posted: Thu May 12 20:57:36 1988

It seems that free will vs determinism is being nicely beaten to
death.  To abstract out the part of the discussion that I personally
find interesting: those who object to determinism and those who object
to any deterministic form of intelligence seem largely to object to
the fact that deterministic entities can not carry "responsibility".
Having intelligent yet irresponsible entities is indeed a scary
thought.  But I still am hazy about what "responsibility" might be.

Well first the dictionary ("The Oxford Minidictionary"):
    responsible:
1.	obliged to take care of something or to carry out a duty.
2.	liable to be blamed for loss or failure etc;
3.	having to account for one's actions
4.	capable of rational conduct
5.	trustworthy
6.	involving important duties
7.	being the cause of something.
An unusually long definition for a pocket dictionary probably
indicating an unusually difficult word.   

So now we can consider which of these definitions can be applied to a
machine, intelligent or not.
Definitions 1,4,6, and 7 can be applied to any machine.  3 can be
applied to software with a debug mode so that actions can be explained
as a result of input and state.  Thus 3 can be applied to axiomatic
systems but perhaps not so much to connectionist systems.  However 3
is applied to human beings (the ultimate connectionist system) so
perhaps this is not a valid objection.  I have encounted machines I
trusted and machines I wouldn't trust with a bit so I guess 5 can be
applied to machines.  That leaves definition 2.  

Can a machine be "liable to be blamed for loss or failure etc?"

Currently when a machine fails blame is either accrued to the operator
of the machine if it is improperly operated, or the manufacturer if 
it failed due to an internal flaw or to G-d if it failed in a
completely unacountable way.  We never really blame or punish a car
for hitting someone.  If it hit someone because the driver was drunk
we blame the driver.  If it hit someone because of a failure in the
brake system we blame the manufacturer or G-d depending on the nature
of the failure.  

Would intelligent machines be different in this respect?
If so how?

I am personally interested in this problem since I currently have a
small project (that I hope will grow larger) in medical imaging.  If
this project bears fruit I may be directly confronted with the issue
of responsibility for intelligent machinery.  

One last note: we seem sometimes to apply responsibility to things in
inverse proportion to our understanding of it.  Thus a more complex
and difficult to understand system will tend to carry more
responsibility than a simpler system.  Thus we will place more
responsibility in our cars than in our screwdrivers.  We also place
more responsibility in the economy (the ultimately complex system)
than in our cars.  This may be why people want to assign
responsibility to intelligent(=complex) machinery but not to simpler
machinery.  Is this reasonable behavior?

-David Sher
ARPA: sher@cs.buffalo.edu	BITNET: sher@sunybcs
UUCP: {rutgers,ames,boulder,decvax}!sunybcs!sher