[Sigia-l] TMI - and expert-level knowledge

[Donna Fritzsche]

My thinking was headed in the same direction John - but I was 
thinking of a program call ID3 created by Ross Quinlan which used an 
information-theoretic measure to narrow down the set of of attributes 
(very similar to, facets :) ) to the most important ones to consider 
when making a diagnosis.  These then become the set of rules that the 
expert system applies in diagnostic and classification situations. 
(I am not personally advocating this for medicine - but the technique 
is interesting and has potentially interesting applications which 
range from reading medical slides (high human error) to speech 
recognition).

The difference between ID3 and MYCIN is how the rules are entered 
into the system.  For MYCIN they were obtained by interviewing the 
expert, for ID3 they used experts to determine the breadth and depth 
of attributes to consider and also to create a data set. They then 
applied the machine learning techinique (information theoretic 
measure) to derive the rules.

I bring it up in this context because the system figures out what 
questions to ask in order to *efficiently* get to the correct 
diagnosis (classification, etc).  At a cognitive level, it would 
appear that the new technique advocated  for the ER is similar. 
Reduce cognitive overload - so that doctors can access the 
precompiled info, instinct, etc. that they have spent years obtaining.

>
>So are we starting to consider doctors under pressure as logical,
>decision making machines that have to work on rules based deduction from
>a set of limited information?

Medical training is already very decision-tree based (rule based, 
memory based) for just this reason- when on duty for long hours and 
under stressful conditions - precompiled information needs to take 
over (at least I believe that is the thinking).  You can't start from 
first principles when you need to make split second decisions.

Donna