Re: Model and sources of uncertainty

Vipul, All,
On Jun 18, 2007, at 1:14 PM, Kashyap, Vipul wrote:

>> Dissonance happens when you have distinct pieces of evidence
>> supporting contradictory or conflicting views.
>> If a given piece of evidence supports hypothesis H and another
>> supports hypothesis notH, then we have contradiction, since H and
>> notH are mutually exclusive. A Knowledge Base that has two axioms
>> supporting contradictory hypothesis is inconsistent.
>> Conflicting evidence is another form of dissonance, this time
>> supporting hypothesis that are not mutually exclusive. Therefore, a
>> Knowledge Base that has two pieces of evidence supporting conflicting
>> hypothesis is not necessarily inconsistent.
>> Ambiguity happens when you have one or more pieces of evidence that
>> are subject to different interpretations, which can lead to distinct,
>> sometimes incompatible conclusions.
>
> [VK] I guess, we need to model "evidence" and "interpretation" as  
> well in our
> ontology in addition to agent, statement, belief and uncertainty

Agree!

>
> It's interesting how different notions of uncertainty presented  
> above depend on
> the notion of "evidence" and "interpretation".

This dependence comes from the fact that we perceive the world via  
sensors, which give you evidence about the world that you will have  
to interpret using your prior knowledge about it.

> BTW, at this time my position on
> these definitions is "uncertain", but I do find them clarifying and  
> interesting.

As I've seen in this group:  +1  :-)

>
>> In other words, dissonance (either contradiction of conflict) refers
>> to the relationship between distinct pieces of evidence, whereas
>> ambiguity is more related to how a given piece of evidence is
>> interpreted. To explore this idea further, two pieces of evidence may
>> not be ambiguous (each one is clearly supporting a specific
>> hypothesis), but they might be dissonant when analyzed together.
>
> [VK] So looks like you are identifying the underlying causes of  
> uncertainty
> rather than the various types of uncertainty themselves?

Yes!
Types are Epistemic and Existential, although there's a deep  
philosophical debate on this issue.

>
>> In short, if we see uncertainty as the inability to predict the
>> outcome of an event, then such inability can be caused either by our
>> imperfect knowledge on the event or by the nature of the event
>> itself.
>
> [VK] What about the case where imperfect knowledge about an event  
> can lead to
> the prediction of a certain event?
>
> For e.g., If the patient's blood pressure is between X and Y then  
> the doctor
> will prescribe ACE Inhibitors.
>
> So the event(or knowledge thereof) is ambiguous, but the prediction  
> is certain?

I'm not sure this event is ambiguous. I don't need to know the exact  
blood pressure of a patient (if that number really exists) in order  
to prescribe ACE inhibitors. The measurement is a standard procedure  
that gives me an estimate of the patient's average pressure among its  
many blood vessels. If that estimate falls into a given range then  
the physician has an (strong)  evidence that this patient needs ACE  
inhibitors.
I may be purposely forcing a situation here as a way to make the  
point that we don't have perfect precision in any of our  
measurements, but that doesn't mean we can't use those measurements.  
For the purpose of prescribing ACE inhibitors, a measurement between  
X and Y is enough to make a decision (which might not be a certain  
prediction, just a highly probable one).

>
> Also, ambiguity could be context sensitive? For e.g., the same  
> thing can be
> precise and ambiguous in multiple contexts?

I believe so. Preciseness is a context sensitive concept. A specific  
level of preciseness may be sufficient for your needs, but might be  
insufficient for other purposes.

Paulo

Received on Wednesday, 20 June 2007 03:12:49 UTC