AI as a design problem

a few weeks ago on slashdot
i experimentally re-tested the waters
of the semantic web community
that i'd mostly given up on ten years ago

and found them, still,

prescriptivist

trying to solve the problem
by spanking non-conformists
until they bought into a
semi-arbitrary
set of syntactic rules

which keeps them busy
and gives them someone to blame
but advances AI not a whit




a step above prescriptivism
is descriptivism
which is embodied by Lenat's Cyc
and perhaps also by 'The Sims'

trying to describe human behavior
with a definite constrained vocabulary
(Sim1 covets Furniture2)

but i fear the catch-22 of descriptivism
is that there isn't any platonic 'form', there,
waiting to be described
(as there more-or-less was
for physics, chemistry, astronomy,
and even biology)

but instead a mathematically intractable
neural 'chaos'
an everchanging hierarchy
of human moods and motives

crudely reflected in 'ontologies' like the classic
Yahoo top menu
(currently:
Arts&Humanities, Business&Economy, Computers&Internet,
Education, Entertainment, Government, Health, News&Media,
Recreation&Sports, Reference, Regional, Science,
Social Science, Society)

designed entirely
to meet the instantaneous needs
of (yesterday's) web surfers




which is why i'm exploring heraldic barcodes
that try to map the top levels
of this semantic hierarchy
(for a specific subdomain)
onto an intuitive graphical equivalent

whose strengths and weaknesses
not only declare themselves openly
but also
hint at how to improve them
in graphical tweaks
that better fit the human users

A generalised more-like-this button

the longterm trend
in browser evolution
has to include
generalisation
of functions
across the widest possible range
of data/content/objects

so the 'subscribe' button
that looks for an rss feed

ought to be generalised as a
more-like-this button

that offers 'subscriptions'
to every possible relevant feed:

- watch for more by this author?
- watch for more on this site?
- watch for more on this topic?

- watch GoogleNews for relevant string/patterns?
- narrow GoogleNews feed to particular news sources?

and these feeds should have killfiles as well

so when your 'artificial intelligence' pattern
briefly floods you with gossip items about
'AI' star Haley Joel Osment
you can flip on a temporary filter

using a form/syntax
that works exactly the same
for all your more-like-this feeds

Web2.0 on the Tree of Life

as a human branch
browses from website to website
webpage to webpage

storytopics are activated
in its radial ontology

and it may be moved to
bookmark the page
or blog it for others

or create a new content-item
responding to it
to be posted on the same site
or a different one




the content being browsed
can be thought of as
a stream of data-items

being evaluated
approved
rejected
acted upon

the branch's blog
can be thought of
as a filtered subset of that stream
plus newly added content

as, similarly
the moderator of a website
filters the stream of new content
submitted by other readers




so as browsing software evolves
it must become easier and easier
to apply the same actions
to every sort of content

sites pages sections-of-pages
images soundfiles movies

sets of images
sets of pages

conversational threads

all equally bookmarkable
or bloggable

effortlessly shared with a definite
or undefined audience

with relevant past discussions
anywhere on the Net
automatically tracked

and current or new discussions
automatically opened to those who qualify

Documented emotion on the Tree of Life

for each story
in our radial ontology

and for each player in those stories

and for each emotion
experienced by those players

assign a distinct emotion-color

with similar emotions
getting similar colors




we assume that most human branches
will have experienced some shades
of every flavor of emotion

but we have almost no evidence
of who what where and when




so if an artifact survives
that 'documents' a past emotion

let us light
an appropriately colored star
at that time and place

(a dimmed cloud
if these are
as usual
uncertain)

with unambiguous documents
shining in pure colors
and ambiguous
impure




and first let's map
the innovators of documentation

who purified an emotion-color
in a surviving artifact-document
for the first time

these stars will cluster strongly
around individual artistic geniuses

and cluster more weakly
across eras of more general genius

and the palette of colors
will gradually fill out
as overlooked
or taboo emotions are added




but even as more and more artifacts
survive
documenting more emotions
more clearly

we still have to doubt
the authenticity
of each particular depiction
that claims to pertain
to a specific branch
at a specific point

for publication usually
assumes a public
false
face




the advent of photography
meant that occasional
true faces
were captured at specific
times and places

and the advent of flickr
means that dozens of these are now published
every second
[rwwl]

Search-engines on the Tree of Life

one class of stories
all web-surfers partake of

is searching for info

and the usual startingpoint
is google or another search-engine

and one or more typed-in query-strings

which should eventually lead to
the info sought




many searchers may pursue the same info
perhaps by different routes

or related info

which could all be aggregated into a FAQ

with the best search-engine
then being defined
as the one that leads you quickest
to the proper FAQ




wikipedia
is filling the FAQ-role
for ever-increasing numbers of queries

but their policy unfortunately
still eschews
extensive linking of
available web resources

which to my way of thinking
is the natural successor to
google-style page-hit-lists




for example

i held off researching
snapshot indexing-schemes
while i was formulating my
heraldic barcode approach

because past experience suggested
the available web resources
would be more frustrating and confusing
than enlightening

but this morning
i felt i'd progressed enough
to dip a toe into google

trying various combos of
snapshot photographs
ontology classifying metadata etc

and sorely regretting that wikipedia
doesn't yet have
an annotated history on this topic

because what i got was all dead ends

not a single proposal that overlapped mine

which doesn't remotely imply
that these don't exist

only that keyword search is inefficient
for tracking them down




one especially frustrating thing
is that google still doesn't
give any hint of the document types
of the page-hits

article or abstract
project proposal or specific ontology

and this topic is especially messy
because academics are busy poisoning the well
generating clouds of ink
meant to create an illusion of expertise

instead of humbly acknowledging
that the hardest question, as always
involves indexing human behavior...