14may2002

Internet OS


http://conferences.oreillynet.com/cs/et2002/view/e_sess/2396


human scale storage: now thinking about storing everything that happens to me,
and maybe even carry it
task vs program?
query vs hierarchy?

--> almost at the point where voice is as good as typing, he says. hmm.

ooh, there's a dll with windows to do text parsing! things like NP1 and NOUN2 --
deep object, modifiers. extractable! i wonder how programs can do this?

they've got their own version of "http://www.cogsci.princeton.edu/~wn/" called
mindnet. "quack" "is caused by" a "duck", would be one link in there for
example. see:
http://www.google.com/search?hl=en&q=microsoft+mindnet
and
http://www.comp.lancs.ac.uk/computing/research/ucrel/public/1035.html
longer paper:
http://research.microsoft.com/scripts/pubs/view.asp?TR_ID=MSR-TR-98-23
basic link:
http://research.microsoft.com/nlp/tell_me_about_yourself.asp

--> askMSR: automatic question answerer from microsoft. prototype, research
thing. type in a nat lang question. it actually plucks out answers, rather than
chunks of text from resources. eg "who is head of ms research", 6 answers with
"rich rashid 26%" being the top.

(okay, so they're talking loads about linguistic analysis, patterns of text, etc
to sort mail. all this stuff would be *really* good if it wasn't bundled up with
the applications and the OS. why don't they sit in like procmail?)

some interesting stuff about how to alert you to things. there are loads of
inputs (mail, etc). by collecting contextual information (webcam: are you in
conversation, mic, etc), can judge how to prioritise input for you, and change
the output?

(it's a reverse scaling problem. works brilliants at 100% effectiveness, and
progressively less good with less effectiveness. so we'll never get there. i'd
never use it  at 50% because all my stuff would get lost. classic problem of how
to get to utopia)

* continuous execution: event driven vs explicit command
event driven is like standing information requests or filters on always changing
data --> lifestreams [not that he said that]

basic idea is to: create a document + everything about it, collaboratively.
instead of just using word and creating a doc. --> [opendoc]

traditionally
os are location based: file, folder, service

but
people more naturally think in associations, queries
- data ref'd by query (ie contents) not location
- and some other stuff i didn't catch

so he's saying: let's make everything a search:
what have i seen?
what is related?
how are the things i'm working on changing?

(i think this is a distraction. yes, this is useful ---- but we need to have
multiple ways of referencing things, right? surface structure and deep
structure. stuff on my desk has a location AND associations, in the physical
world and my head. both, surely?? and THAT'S why location stuff works so well --
i store the associations myself)

so now he's mentioning the os being sql, so you can query it by contents easily

now he's saying it should keep copies of everything you see

HANG ON (i say) why does anything change at all? why not keep histories of
everything?

how do you do ui the way people think?
deep history
dynamic org -- views
associations

demo ui which is a text box at the top of the page -- you type and it helps you
type, names, etc. so everything has a way of searching, etc. syntax? wow. cui.
also, calender etc, very insider the os. working on a document, and associated
documents kind of float on the right hand side, and tooltips over recognised
people names float with info. "ultimate smart tags"

conversational user interfaces mentioned here:
http://research.microsoft.com/adapt/
http://research.microsoft.com/adapt/conversation/

ways of modelling the people who are associated with you
kind of querying the lifestream says jones

as a paradigm: he's saying that web services are query driven, so this fits very
well

keywords: event driven execution, data focus, self-describing data (as opposed
to implicit data definition, which is unix apparently), self running programs
[that's demons, from unix surely?], differences for people: 3d, linguistics,
maps, all allowed interfaces.


=====

thought about ZUIs:
what we saw [zoomable ui] was a reconception of the text document, nothing
higher than that. the hypertext/transpublishing bit is kind of being built into
that. but the www comes much much later.