[FoRK] A word about bias...

Jef Allbright <jef at jefallbright.net> on Thu Dec 20 16:44:47 PST 2007

On 12/20/07, Jeff Bone <jbone at place.org> wrote:
> I have a great deal of admiration for Robin Hanson;  somewhat less
> so, though an enduring interest in the activity of, Eli Yudkowsky.  I
> read overcomingbias.com on a daily basis --- yet every time I do, I
> have to chuckle some at the name.
> Bias, of course, can be a perfectly fine thing.  Some biases provide
> a fitness advantage in various fitness landscapes;  they are
> evolutionary shortcuts, means of achieving statistically correct
> behavior in the absence of complete information or enough time to
> process whatever information is available.

It's perhaps a subtle point, but it seems to me that **any and all**
meaning is strictly dependent on context -- and this is at the center
of the understanding gap between Robin and Tyler in the present

I've appreciated and respected Robin and his writings for well over a
decade now, but I've always been slightly jarred by what appears (to
me) an implicit assumption of objectivity in his models, whereas for
me, all models are understood within the context of the observer, all
observations are relative to some prior and inherently uncertain, with
a background assumption of a consistent, rather than "objective"

At this point, for the record, I should state that I find relativism
and post-modernism annoying, and that's not what I'm about.

> Many kinds of bias, of course, are so specialized as to confer no
> general benefit --- and many can be anti-fit when the host is
> suddenly in a novel fitness landscape.
> Bias by itself isn't the problem.  It is bias which reduces fitness
> in some important fitness landscape that's the problem.

What is the difference between a human embryo and a duck embryo, if
not a set of biases?
What is at the root of any intentional action, if not bias?
What is "good", "right", or "moral" without bias?

But of course their point is (or should be) that one should prefer
one's instruments (especially Rationality) to be effectively unbiased.


- Jef

More information about the FoRK mailing list