Special difficulties of AI terminology

From: Eliezer S. Yudkowsky (sentience@pobox.com)
Date: Tue Feb 19 2002 - 08:25:48 MST

Somewhere in an alternate universe, there is a version of Eliezer Yudkowsky
who publishes "leisurely" scientific papers, and one of those papers is
entitled "Special difficulties consequent to research on generally intelligent
supersystems." I'll probably never get to write that paper. But I'd like to
publish a small excerpt from that alternate universe.

Right now I'm trying to use standard words, without neologism, and yet this
involves a new problem: I'm rapidly running out of words. If you discuss
generally intelligent supersystems, then not only will you need to overload an
immense number of words, but furthermore they will be very abstract words,
ones that would otherwise play a strong supporting role in abstract language.
In my paper, I can no longer say "the past concept that children have no
mathematical abilities"; I have to say "the past notion that children have no
mathematical abilities". "Concept" has already been overloaded as a technical
term. I have to use "notion" or "idea".

Recently I wanted to discuss the second part of a trigger - you know, the
thing that happens after the condition part is satisfied - and I found that I
couldn't. "Action" was long since overloaded to describe a deliberate action
of the AI. "Sequitur", which would have been appropriate in this case, had
been used somewhere more important. I'd used up the space of causal
terminology and the only words left didn't really describe what I wanted to
say - "consequence" and "effect" and "outcome", for example, treat the action
(sigh) as static, rather than as an active force. And at that I was lucky the
paper didn't go into causality in more detail, or I would probably have been
forced to use up "consequence", "effect", and "outcome" as well. At one
point, faced with the prospect of being forced to overload the term "problem"
to describe a virtual-environment challenge presented by the programmer to
teach a concept or belief, I took the coward's way out. I called it a
"microenvironmental challenge", abbreviating it as "MEC".

So you can avoid neologism, but it's going to come at a cost; describing
thought processes abstractly often consumes a term that we use in ordinary
abstract thinking. Maybe when I'm done with the first draft, I'll be able to
go back and reparse the language space more precisely and get rid of some of
the current problems. For example, on reflection (and some
thesaurus-hunting), I think that "microtask" may do to replace the neologistic
"MEC". But don't be too quick to blame the creators of neologisms until
you've walked a few miles in their thesauruses.

-- -- -- -- --
Eliezer S. Yudkowsky http://singinst.org/
Research Fellow, Singularity Institute for Artificial Intelligence

This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 13:37:40 MST