From: Eliezer S. Yudkowsky (sentience@pobox.com)
Date: Wed Nov 20 2002 - 21:46:09 MST
It's an odd thing, but the laws of physics, which describe very well how a
complete system *develops*, seem to make no mention of one particular
event causing one particular other event. I would therefore suggest that
you consider "causation" as a cognitive abstraction from a complete
system, and ask about the uses (i.e., the ecological context and
evolutionary motivation) of identifying one specific regularity in reality
as the "cause" of another. Let's say someone drops a glass and it breaks.
We could describe the "cause" of this event as his opening his hand, or
the car crash outside that startled him, or the presence of gravity, or
the hard floor. The event we'll tend to describe as the "cause" is the
event that's (a) most unexpected and (b) easiest to manipulate. Had
everyone involved been accustomed to a zero-gee environment, it's quite
probable that they would have all said "Damn that gravity!"
Let there exist a background environment of sufficient constancy that it
would not be a good use of computing power for a bounded rational system
to pay attention to it. Suppose you identify a event or set of events A
which, occurring against this background, are necessary and sufficient for
the complete system to develop event B. In this case you can achieve
result B by carrying out A as a subgoal. In other words, the conditions
that you list as required for "causation" are, through some strange
coincidence, precisely the conditions required for a human phenotype of
limited intelligence and computing power to usefully manipulate reality in
a certain specific way. This is why physics impartially describes the
development of complete systems, but cognition singles out specific events
as causing specific other events; causation is a useful and above all
*manipulable* regularity in the development of complex systems.
-- Eliezer S. Yudkowsky http://singinst.org/ Research Fellow, Singularity Institute for Artificial Intelligence
This archive was generated by hypermail 2.1.5 : Wed Jan 15 2003 - 17:58:16 MST