NetNews Usenet Archive 1992 #31/NN_1992_31.iso/spool/comp/ai/philosop/7188

Text File | 1993-01-01 | 3.7 KB | 83 lines

Newsgroups: comp.ai.philosophy
Path: sparky!uunet!tcsi.com!iat.holonet.net!ken
From: [e-mail] (Ken Easlon)
Subject: Re: Drawing the entity/environment boundary
Message-ID: <[e-mail]>
Organization: HoloNet National Internet Access BBS: 510-704-1058/modem
References: <[e-mail]>
Date: Fri, 1 Jan 1993 15:03:25 GMT
Lines: 72


In article <[e-mail]> ,
[e-mail] (Marvin Minsky) writes:

>1).  The notion of "understanding" is itself defective if we try to
>define "A understands X".  Read Chapter 30 of The Society of Mind.
>A major point is that this is relative to someone's judgment, so one
>really should be discussing "O agrees (thinks, believes, assumes) that
>A understands (in some sense proposed by O) X.  So Joe may think Jack
>understands Physics, whereas Gell-Man thinks Jack understands nothing
>about it.

I think if we apply enough tests we can determine for all practical
purposes whether or not A understands X, per a standardized definition of
"understand".  I agree that without the standardized tests and definitions
there might be some confusion.

>2) the notion of "boundary" should also be relative to some observer...

I think what Mark is searching for is a standardized definition of the
entity/environment boundary that is useful for attaining a functional
degree of understanding of the process of "grounding".  I guess my point in
bringing up all the other possible uses of the boundary concept is that
there are a number of ways of thinking about grounding.

I think if we are to determine if an AI configuration (notice my clever
avoidance of the flamed words) is sufficiently grounded for practical
purposes, we need to know the job(s) the AI is designed for.

>Read Dawkins' The Extended Phenotype for many examples of where it is bad
>to draw a boundary around a particular cell, organ, or individual
> organism's body.  See his discussion of Beaver-caused lakes, for example.

I haven't read Dawkin's, but I can easily visualize the human body as a
structure of proteins and glyco-proteins, with cellular membranes being
little more than sort of an access control system.

In this view (which only consider's a cell's interphase function), cellular
boundaries are only important for determining the local chemical
environment in microscopic regions.  A membrane enclosed area would be an
entity for a biochemist interested in studying its chemical content.

Of course when we look at the cell's whole life cycle, the structure
enclosed by the cellular membrane takes on many more entity defining
functions than simple access control.

Likewise, if we look at a functioning AI configuration we might draw the
boundary around a rather large area based on job description.  If we look
at other phases of the design-manufacturing-deployment-junk life cycle we
might draw the line some place else.

In other words, a functioning AI might have to interface with a large
number of other "entities", but when it comes time to scrap that pile of
junk we only have to replace one robot.

I think Mark is interested in the training phase.  How much equipment,
software, and developmental interaction is necessary to provide the "A" "I"
with the "I".

In my opinion, once you go through the training process for one system, you
can build an exact copy without going through all the interactive
development, which means you can eliminate some of the equipment that was
used exclusively for the training phase.

I wonder what can of worms I've opened with THAT statement?


--
Ken Easlon                | "...somebody spoke and I went into a dream..."
[e-mail]           |    -Paul McCartney
Pleasantly Unaffiliated   |