RE: Books on rationality

From: Ben Goertzel (
Date: Fri Jun 07 2002 - 11:09:18 MDT

> "Words are like ball-bearings on a skating rink:
> to get anywhere, tread lightly and avoid putting
> too much weight on any one of them."
> The easy part: adopting the truth of that prescription
> and seeing how in the recent discussion several people
> we'd always describe as "quite rational" scarcely knew
> what they meant by the term.
> The hard part: getting somewhere, especially in software,
> I suppose, without heavy reliance on hard and well-defined
> concepts. (I'm confident from what I've read, however,
> that no one is likely to be doing a better job than some
> of the savants on this list.)

It's true. I re-read Wittgenstein recently and was newly impressed by his
discussion of "language-games", i.e. of the fact that "word meanings" are
just conventions of social discourse and rarely capture any truly "natural"
or "absolutely meaningful" concept.

In this view, debates over the real meaning of truth, beauty, rationality,
etc. are not terribly useful. There are going to be many overlapping
conceptions of such things, and the human language term is going to be a
fuzzy mixed-up combination of them (furthermore a combination that shifts
over time).

My approach in my own work has been to create relatively precise definitions
of concepts like "intelligence", "mind" and "reason" -- without claiming
that these capture some kind of ultimate essential concept and are thus
better than alternative definitions. More in the mathematician's mindset,
where first you make your definitions and then you do some work relying on
them. e.g., is intelligence "really" captured in my def'n "Achieving
complex goals in complex environments"? I don't know what this means,
because I don't have access to the space of Platonically Ideal Correct
Definitions of concepts like intelligence. All I know is that when I make a
definition, at least I know what I'm talking about reasonably well ;>

So, yeah, debating what rationality really is, is useless. What's valuable
is to make a reasonably crisp definition of rationality, explain it, give
examples, and then give guidance as to how to achieve it for those who are
interested. I guess that is what Gordon plans.

One thing the recent discussion uncovered, is that Gordon's notion of
rationality made more assumptions about the mind's underlying goals than my
own notion of rationality. My notion of rationality is, basically:

A system X is rational *relative to a system Y* if

-- the conclusions it draws are (in Y's judgment) about as accurate as
possible given the constraints on X's cognitive abilities

A corollary of rationality is that, if a system has explicitly stated goals,
it should draw accurate conclusions (within the limits of its abilities)
about how likely its actions are to help it achieve its goals, and about how
its short-term goals relate to its long term goals

Often, in judging whether a human X is rational or not, we assume that the
judging entity Y is a kind of composite consensus-reality cultural belief

To make an objective definition of rationality, one has to assume one knows
the answers, e.g. one can postulate a judging system Y that makes perfect
probabilistic judgments based on the available evidence (though one can
never actually build such a system).

It is worth remembering, however, that we are not this perfect reasoning
system Y, ergo when we judge X to be irrational, we are doing so only
relative to our own judgmental ability, which may be flawed.

I do not think that my definition of "rational" (loosely given above)
captures the full natural language concept of rationality. Natural language
concepts can almost never be captured in crisp definitions. My definition
is useful for me, perhaps useful to others, and that's enough :>

Note that my definition does NOT specify that a system must have any
particular goals to be rational. In my definition, an evil serial killer
could be rational if, given his emotionally determined goals, he drew
accurate conclusions about how to achieve them. This seems in line with
common usage of the term, where we have the notion of the "cold-blooded
calculating killer" ....

-- Ben G

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:39 MDT