[961] in Public-Access_Computer_Systems_Forum
Boolean, Bread and Nonsense
daemon@ATHENA.MIT.EDU (Walt Crawford)
Tue Aug 11 12:25:11 1992
Date: Tue, 11 Aug 1992 11:17:51 CDT
Reply-To: Public-Access Computer Systems Forum <PACS-L%UHUPVM1.BITNET@ricevm1.rice.edu>
From: Walt Crawford <BR.WCC%RLG.BITNET@ricevm1.rice.edu>
To: Multiple recipients of list PACS-L <PACS-L%UHUPVM1.BITNET@ricevm1.rice.edu>
----------------------------Original message----------------------------
Re: Probabilistic vs. Boolean and the Volvo Analogy
The more I think about this, the more it strikes me that the Volvo
analogy makes a rather different point than C. Hildreth may have
intended. Here's the way I look at that:
When you take the Volvo down to the store to buy some bread, you
perform an explicit, ordered, learned series of actions (turn key
in door lock; open door; get in; adjust seat; put key in ignition;
start car; put car "in gear"; look around appropriately; step on gas...)
with considerable confidence that the car will do "what you tell it to."
It's a lot like explicit Boolean searching; not the easiest thing to
learn (we're not born as car drivers), but effective once you learn it.
You may not understand the transmission or the workings of an internal
combustion engine, but you do understand the direct, predictable
relationship between action (step on gas, but only after starting
car and putting it in gear) and reaction (car moves forward).
Let's say that, instead, you get in the car and tell its new voice
interface "I need to get some bread." After all, that's what you need.
Then you wait a few minutes. The car may offer names of nearby banks
and ask if those are relevant to your need...or it might buckle
you up, start (it's recognized your voiceprint already), and take
you down to the nearest bakery...or it might tell you to get off your
duff and bake your own bread; it's healthier.
(This *is* a Volvo, after all.)
Somehow, I think most drivers would find such a user-friendly car
distinctly unsatisfactory...at least in part because they have lost
confidence that they understand what use is being made of their
commands or requests. Something is happening, but they don't know
what it is...or, rather, *why* it is.
That's the point: *not* whether the user understands all the
mechanics involved, but whether the user *feels in control.*
Which is not to say, incidentally, that Boolean is necessarily
good or that probabilistic methods (which have been available in
demonstration versions of online catalogs for quite a few years,
actually) are necessarily bad. Different argument.
I do reject the suggestion that it's "nonsense" to say that
the user must know what's going on. They must be able to understand
the techniques at the gross level--specifically, that what they ask
for has a direct and predictable bearing on what comes out.
-walt crawford-
