Re: bias

From: Joshua Fox (
Date: Fri Mar 23 2007 - 05:58:49 MDT


But does it make sense to make bias into the highest priority? Isn't it more
important to find a good design for artificial intelligence than to identify
the problems in something quite different -- human intelligence?

> see that essentially *all* errors are attributable to cognitive biases,

Biases are design features in human brains. They generate errors that
consistently go in a certain direction; there is always some reason for
this, such as computational efficiency.

But cannot there be non-bias errors, those generated by bugs in an
intelligence, e.g. in humans, mental illness and problems caused by brain
damage; or in computers, problems caused by software bugs and hardware

It just seems strangely over-ambitious to hope to re-make the entire
scientific method, when a very difficult task but more specific task stands
before us.


2007/3/22, Michael Vassar <>:
> We focus on bias because we know of abundantly numerous instances of
> historical advocacy of proposals that seem like good ideas initially due
> to
> biases but which were later recognized to be omnicidal.
> More generally, once you know a certain amount about normative reasoning
> you
> see that essentially *all* errors are attributable to cognitive biases,
> and
> when you are aware of UFAI you see that essentially *no* errors in FAI
> design can be tolerated. It must work perfectly the first time. Science
> can't do that. We need something better. Science is a set of evolved
> heuristics for avoiding bias. Via scientific study of bias we hope to
> create somethign better, e.g. systematically designed debiasing
> procedures.
> _________________________________________________________________
> i'm making a difference.Make every IM count for the cause of your choice.
> Join Now.
> _________________________________________________________________
> Need personalized email and website? Look no further. It's easy
> with Doteasy $0 Web Hosting! Learn more at

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:57 MDT