[extropy-chat] what is probability?
gts
gts_2000 at yahoo.com
Sun Jan 14 17:39:30 UTC 2007
On the subjective theory of probability, two rational agents with the same
background knowledge may hold different judgemental probabilities on the
same outcome, at least in certain situations.
Though I don't necessarily endorse this view, I think I can defend it. Jef
wants to know for example how two rational machine intelligences might
offer different outputs given the same inputs.
Drawing on the cube conundrum, imagine this scenario:
Unbeknownst to each other, two rational agents (machine intelligences,
AI's, whatever) are each given a box containing a cube.
The cubes in the two boxes are identical. The agents know only that their
side-lengths are between 3 and 5 centimeters, their surface areas are
between 54 and 150 cm^2, and their volumes are between 27 and 125 cm^3.
Note there is nothing challenging about these ranges; lots of real cubes
could meet these constraints.
For some practical unspecified reason, each agent must use the information
given to make a best-guess of the probable dimensions of the actual cube
in his respective box. (As often happens in the real world, our agents are
being forced to make decisions under conditions of high uncertainty.)
They make their decisions, then come together to compare their conclusions
and their reasoning. Is it possible that our agents will have reached
different conclusions?
I think so. In fact I'd be surprised if they didn't.
Our agents were forced to decide arbitrarily between 'side-length',
'surface area' and 'volume' as the parameter to use for estimating the
over-all dimensions of the cube. Each of these three methods leads to a
different but perfectly rational judgemental probability. As we've seen,
judgements based on all three parameters are paradoxical.
-gts
More information about the extropy-chat
mailing list