--On 9 November 2006 19:08:57 +0000 Chris Lightfoot <chris@???>
wrote:
>
> there's a distinction between random errors and actual
> errors of classification. obviously there is a random
> error rate for both human and machine classification of
> mail. however, there are also systematic errors. a person
> will not make systematic errors of this kind in
> classifying spam (by definition -- if they decide that
> they want a certain mail, then they did, end of story).
Clearly, the human is the final arbiter, and can't really be gainsaid on
their decision. Still, they might live to regret it in the case of phishing.
However, as a classifier, working under pressure, humans do make mistakes.
It doesn't really matter how mistakes are made, the point is to try to
build a system that (a) takes load off the human and (b) reduces the number
of mistakes.
--
Ian Eiloart
IT Services, University of Sussex
