[rdfweb-dev] foaf:Person identification

[Stephen Pollei]

On Wed, 2004-01-07 at 15:36, Mike Higginbottom wrote:
> Sure, but they suffer from two problems.  Firstly, none of them 
> necessarily act as a unique representation of a single person.  Secondly, 
> they aren't congruent with a sensible definition of a person's identity 
> (see below).
Nothing by itself that isn't artificial(guid/tag uri/opaque string) is
unique representation. And even then people sometimes steal more than
one, or steal someone elses;-> Think Identity Theft:-( Like Libby said
in other email all identifiers have problems. You should use more than
one, as much as possible. cross-reenforcement.
The most intrinsic seems to be birth_time, birth_place, parents, and
biometrics. death_time and death_place; however the people to which
those apply are generally dull;-> For twins you had best include some
randotypic biometric(s) and birth order(or precise enough birth_time to
disambiguate)
> But you're not your e-mail addy, your home page or your blog either ;).  
> None of the existing properies fit the bill but there is an implication 
> that they do.  I can't help finding this a bit of a kludge.
Yup. Especially their use of mbox without time information and without
other collaborating evidence. mbox is more of a contact information than
identifier. Hey can I call you 812 Walton St. ? Why wouldn't snail mail
addresses work? You could even sha1 those to keep privacy! Same
arguments against street address not being used as identity work for
email addresses in principle. In practice more people leave together
than share addresses.. However think cheap $19.95 one email 56k modem
dial-up account a family might use. I guess they are too cheap for foaf
to grace them with distinct identities though:-(
> In effect 
> we're saying 'There's no single thing that truly represents a person's 
> identity so we'll take something and squish it and pretend that it does'.  
> It would seem more honest to me to say 'There's no single thing that truly 
> represents a person's identity so we'll use something completely abstract 
> that can have no semantic meaning attached to it accidentally or 
> otherwise'.
Thats an approach govs use a lot. has problems as well.
However having enough detailed and redundant information can work well.
Plus topology itself can work well. You need redundant self-sufficiency
and proper connectivity to work best.
In some ways it's a k-means clustering problem with billions of slots.
> Yeah, agreed, sort of.  If you're marking up a picture of Danbri then 
> you're naturally going to use his name.  That's the way we identify people 
> in practical terms.  I can see plenty of situations though where that 
> might not be enough.  What if you know two Danbris?
The George Foreman problem. George has lots of sons all named after
daddy. How cute is that? And to think they all know each other pretty
well. And people that know one of the Georges probably knows at least
one other George as well. I hope none of them are cozy enough with each
other to share email addresses; I think that would give Libby and Morten
a Coronary;-> foreman_family at example.com -- how cute is that;-?
> What if, a couple of 
> years down the line, we've got to a stage where everybody is FAOFd up and 
> our FOAF aware apps go away and find all the Danbris in the world, present 
> you with a list and ask you to select which Danbri you mean exactly?
Use redundant information. If someone is really close.. list the close
clasher in some files with enough detail to tell them apart. That should
serve as warning to others. The George Foremans and twins should be
listed together, their identities pried apart with force.
> Going beyond that, I just think it would be very useful to have a truly 
> unambiguous way of referring to someone.  It seems like a fundamental 
> building block of any system that is likely to become automated in some, 
> as yet undefined, fashion.  This is why we have bank account numbers 
> rather than just bank account names for example.
Yup passport numbers, nodeid, social security number, case numbers.
Used a lot. They are not perfect either and are not self-sufficient.
You need public key at some point anyway... That can double as
identifier. Has good entropy and usable bits.Better than any uuid guid.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
Url : http://lists.usefulinc.com/pipermail/foaf-dev/attachments/20040107/2cae03f9/attachment-0001.pgp