Partially Observable Phone Plan Processes

[Michael Oltmans <moltmans@>]

How I learned to stop worrying and embrace the cell phone plan selection 
process via random noisy sampling of a complicated underlying domain 
model.

It seems simple.  I have a phone plan, my girlfriend has a phone plan 
with the same company.  I use 70% of my minutes and she uses 110% of her 
minutes (she apparently has more friends and she actually calls them. 
To my friends on this list, I'll try to call soon, I promise.  Really.) 
  Answer get a plan with pooled minutes.  Simple, yes?  Simple NO!

This simple process, which the naive modeler may think is a simple 
matter of asking simple questions to determine the underlying model, was 
not so easily modeled.  First the assumption of accurate responses to 
questions was false.  Multiple calls and transfers yielded contradictory 
answers to several questions.  Our attempts to decipher which plans we 
could choose from and what would happen to our current bills, our future 
bills and when our new minutes started/ended was reduced to a random 
sampling of a partially observable model with (extremely) noisy 
observations.

Along the way we received such assurances as "As best as I can tell, 
you'll be fine.  So you want to switch now right?"  As best as I can 
tell, the best that he could tell was not worth the time it takes to 
retell the tale.  So I won't and we didn't.  Being a computer scientist 
I took the only logical action to take when your model breaks down and 
the system is in a frozen indeterminate state.  I rebooted.

On call number two we had a serious failure in our knowledge 
representation taxonomy.  We had falsely labeled the concept of 
"included nighttime minutes" as "included nighttime minutes."  At some 
point it became clear that this fell into a different category called 
"Promotional Features."  And since month-to-month contracts are not 
eligible for "Promotional Features" we lost our night minutes.  At least 
we were able to repair our broken taxonomy.

To make a long story long, repeated pseudo random sampling of the 
underlying partially observable model eventually converged and we have a 
plan.  Of course we haven't seen the bill yet so who knows if we 
performed enough iterations.  The chances that we simple fell into a 
local minimum are also fairly high.  Regardless, I can call you, you can 
call me and we can all coordinate a way to make it to tonight's, 
partially observable, edition of...

           +-                                                  -+
             girl scout benefit -+-  5:30 pm  -+- 32-G9 lounge
           +-                                                  -+

              For those coming from elsewhere: Building 32 is
               <http://whereis.mit.edu/map-jpg?selection=32>
          Once you are in 32, just take the G-elevator to the 9th
      floor and we will be in the lounge that you will be looking at
                    <http://projects.csail.mit.edu/gsb>





_______________________________________________
Gsb-announce mailing list
Gsb-announce@
https://lists.csail.mit.edu/mailman/listinfo/gsb-announce