Re: [EMOXG] Confidence

Bill,
 
 Very interesting context you set out there. Maybe I need to change my view?
 
 I am fine with no confidence meaning "not known", then it can be left to the developer how they interpret it (actually both ways can do that).
 
 I dont like semantics (i.e. text) much at all but your explanation of cyc highlights problems of numeric values.
 
 To add some more context in support of numeric (although I am not convinced myself now that it is the best solution):
 
 a. Easily Machine readable
 b. MULTINATIONAL (if we state one default format it will no doubt be english semantics, while this is generally universal it does bother me somewhat. Imagine how we would all feel if the default set was Chinese? Perhaps jianhua can assist here ;)
 c. more interoperable
 d. Unambiguous (except when they are biased, or relative or .... )
 
 Perhaps we do need both options?
 
 Ian
 
 Ian Wilson
 CEO
 Emotion AI
 
 w: www.emotionai.com
 e: ian@emotionai.com
 p: +44 (0) 704 040 3198
 
 Bringing the power of emotional communication to your products
 
 
 
 On Wed Aug  6 4:32 PM  , Bill Jarrold <jarrold@AI.SRI.COM> sent:

Hi Ian,


Thanks for the concise reply and apologies for my lengthy one just a few minutes ago.


Here (below) I give a succinct version of my previous email..




On Aug 6, 2008, at 5:48 AM, Ian Wilson wrote:

All,
 
 Points of agreement:
 
 The stated core set (2,3,4,5,7) should have confidence but I dont think any of the others make sense to have confidence.
 
 It should be an attribute.
 
 It should be optional and set by default as 1 (Andys suggestion) 


As in my previous email, there are some problems with this.  Can we just be agnostic about confidence if it is not set?


 
 It should be a uni polar 0 to 1 value, not fuzzy text (Andys suggestion) 


Agreed that text is fuzzy.  Having a fixed set of of ratings, say 5, to choose from is less fuzzy.  


Alas, human certainty of confidence is a fuzzy thing.  Thus a numeric rating imposes an artificial sense of security.


To be sure, allowing arbitrary text labels seems worse than uni polar 0 to 1 value.  But having a fixed set e.g. "very low", moderately low", "medium", "moderately high", "very high" is in some ways better than 0 to 1.


Thanks,


Bill




 
 Points of divergence:
 
 The example for core 5 uses the attribute for the enclosing tag, I am not sure if this makes sense, especially if the sub tags are set as confidence = 1.0 by default.
 
 Best,
 
 Ian
 
 Ian Wilson
 CEO
 Emotion AI
 
 w: www.emotionai.com 
 e: ian@emotionai.com 
 p: +44 (0) 704 040 3198
 
 Bringing the power of emotional communication to your products
   
 -------
 Sent from Orgoo.com - Your communications cockpit!





-------
Sent from Orgoo.com - Your communications cockpit!