On 21 May 2008, at 13:40, Albert Schlef wrote:
> Eleanor McHugh wrote:
>> [...]
>> I started out in physics and I've spent a fair amount of my
>> professional career arguing with very capable CS experts over all
>> kinds of problems which are elementary with my background but have
>> them scuttling back to graph theory
>
> Could you elaborte here? It sounds interesting.

Not easily without getting sued for breaching various NDAs ;)

Generally though there's a deep-seated belief in CS that because  
computers deal in defined states formal abstractions such as graph  
theory are the best way to tackle a number of often-intractable  
optimisation problems. For simple systems that's true, but as your  
state-space explodes you end up with solutions which are easily  
expressed elegantly in Lisp or whatever but which in practice are  
completely useless due to their processing requirements.

Anyone who comes from a science or engineering background will  
recognise this as a granularity issue: a state-space has a level of  
granularity beneath which additional precision in differentiating data  
points becomes completely irrelevant in determining real-world  
behaviour. The classic example is the disconnect between classical and  
quantum mechanics, but it occurs in all experimental fields.

Unfortunately many very good CS people still have a mathematician's  
bias towards a perfect answer rather than a working approximation and  
look to maths for their abstractions rather than seeking them in the  
real world and instead getting the granularity of their systems right.  
In my personal experience this then leads to long and involved  
discussions which go round and round in circles for months on end  
until projects get cancelled because the perfect answer is too costly  
to deploy, and the working approximation isn't provable for all cases.


Ellie

Eleanor McHugh
Games With Brains
http://slides.games-with-brains.net
----
raise ArgumentError unless @reality.responds_to? :reason