COMMENT ⊗   VALID 00003 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	%design[s90,jmc]		Notes on the design stance
C00011 00003	\smallskip\centerline{Copyright \copyright\ 1990\ by John McCarthy}
C00012 ENDMK
C⊗;
%design[s90,jmc]		Notes on the design stance
\input memo.tex[let,jmc]
\title{Taking the Design Stance to Language and Intentionality}


	Daniel Dennett (1971) introduces the design stance as a way of
understanding a physical system.  It complements the physical
stance, which understands the system in terms of the physical
interactions of its parts and intentional stance which describes
a system in terms of its beliefs and goals.  The design stance
understands an object in terms of what it is designed to
accomplish.

	Dennett's classic example is the alarm clock encountered
in motel.  The traveller sets it and is wakened by it without
ever paying attention to whether its rate is controlled by a balance
wheel, by the frequency of the local electric power or by
a quartz crystal in the clock.  This enables Dennett to argue
that ``alarm clock'' is not a name of a particular physical
system.  Dennett points out that the design stance is likely
to be successful in guiding what properties to expect of
either an artifact or a biological system whose properties
have evolved by natural selection.

	The object of this paper is to carry the idea of design
stance farther than Dennett does.  Specifically, we want to
analyze language and intentionality from the design point of
view.

\section{The Problem of Communication}

	Taking the design stance toward language means emphasizing
what language accomplishes for its users that isn't accomplished
by other kinds of behavior.  We can even imagine ourselves designing
language to accomplish this.

	The general situation is that two or more agents are
co-operating to accomplish some goal.  The goal may involve joint
non-linguistic action or it may be just getting one of them to
understand something the other understands.  We begin with the
joint action case.

	Consider two people (or two robots) accomplishing a task
in each others presence.  Much of their joint work requires no
explicit communication, because each can see the situation and
knows what he has to do and knows what the other will do.
We thus regard the use of language as a perturbation on a situation
in which language is not required.

	Consider a surgeon and an operation room nurse who have
worked together many times.  Although they both know what is
going on, it may not be clear to the nurse what instrument the
surgeon requires at the moment or exactly when he requires it.
So he says, ``scalpel''.  We regard the single word communication
as elementary and not as an abbreviation of a complete English
sentence, say ``Give me that scalpel which is obviously the
correct one for what I am about to do''.

	Taking this view of the matter requires considering the
mental states of the surgeon and the nurse and concentrating
on how the surgeon uttering ``scalpel'' affects these two mental
states.  Thus we have the maxim:

	 {\it Language is a means whereby one person affects
the mental state of other people}.

\section{The Need to Postulate Mental States}

	Mental states are hard to observe, and linguists have
tried to make theories that involve only the
observable sentences and don't involve mental states
that aren't directly observable.  I believe this
kind of linguistic methodology, i.e. almost all that has been
attempted so far, is fundamentally limited in what it can
accomplish.  There are three arguments.

	1. From biology.  Every time an advance has been made
in the observation of the internal state of a biological system
that has previously had to be regarded as a black box, advances
have been made that were unlikely to be made otherwise.

	2. From experience with computers and artificial intelligence.
It is obvious that it would be extremely difficult to make theories
of the activity of computer programs without postulating internal
state, e.g. by regarding the behavior of the program as a mapping
from sequences of inputs to sequences of outputs.  Although there
is not much requirement for analyzing computer programs whose
internal states are unobservable, any such analysis would postulate
internal states as a means of accounting for the behavior.

	3. When we design programs to communicate we make essential
use of the states of their memories.

	I grant that postulating mental states that cannot be
observed at the present level of technology increases the chance
of mistakes.  That's just too bad; we have to do it.

\section{What are Mental States?}

	We don't want to identify mental states with brain states
or machine states in the case of robots.  This is because we would
like to regard some aspects of the brain state or machine state
as irrelevant to mental state.  Moreover, taking the design stance,
we will want to decide what mental state space will be useful and
then consider how to realize it with data structures in a machine.

	
	One straightforward approach to mental state is to make
it a set of sentences that are believed at the given time.  In
general this is too simple, but we might examine how far we can
get with it.
\smallskip\centerline{Copyright \copyright\ 1990\ by John McCarthy}
\smallskip\noindent{\fiverm This draft of DESIGN[S90,JMC]\ TEXed on \jmcdate\ at \theTime}
%File originated on 08-Apr-90
\vfill\eject\end