Re: [-empyre-] Archives, metadata and searching

["Stephen Jones" <sjones@culture.com.au>]

Simon, and all the other metaticians

The problem with metadata is the same as the problem with robotics and
traditional AI in general. You cannot predict from the top what
categories of things you will want to allow to be searched for. Or if
you look at the word allow, who is making the decision as to what to
allow?

In robotics, if the robot is not programmed to deal with the contents of
its environment then it will get lost or broken or something. If the
environment is a real world environment then it is going to be changing
all the time and the number of categories by which it makes decision or
classifies things is going to very rapidly exceed the number of options
that have been provided.

The content of the site has to be its own metadata and the search
function has to be someting more like google than some predetermined
subset of the language.

We function as classifier engines and make up our classifcations on the
fly, according to our needs of the moment. Ultimately the only way this
is going to work is to emulate human processes by developing neural
network systems of sufficient depth. They also need to be able to
produce new layers and new connections as they experience the need for
those new connections. This is Hebb's "The Organisation of Behaviour". I
accept that this is a quantum leap in technology (requiring
self-organising electronics), but more importantly it also requires a
major rethink of all these kinds of activities in which the top-down
approach is still paramount.

We are not preset, we do everything we are able to do by learning how to
(in some sense of learning or another) or by seeking out the answer, or
the best approach or whatever (all just sophisticated forms of
learning). There cannot be a "genetic code" for metadata just as there
cannot be a preset code for all the things that one might want a
shopping robot to be able to do or all the ideas that one might want to
express in language. Sadly metadata is inherently a very constrictive
political device, the main result of which is to strait-jacket thinking,
much like fundamentalism does.

Meanwhile, I have a nasty suspicion that simple text-search is still
probably the most effective way to find anything, though perhaps with a
considerable increase in the use of boolean functions. The better
possibility is to use one of those security search systems that reveal
relationships between individuals and/or groups by, for example, looking
for the flow of money in and out of bank accounts.

I need a system that allows me to track the relationships between
individuals that enabled the developement of early electronic arts in
the world. I need to to be able to see what influences people had, what
they produced, who and what their productions influenced and so on. Good
records kept in electronic form can be searched, so long as the
characters are ascii encoded. Bad metadata decisions end up obscuring
things. It's like a lost book in the library. It might not have been
stolen, it might simply have not been put back on the right shelf, but
it is still just as lost.

Of course I am mainly talking about text, but classifier engines can
look at pictures and recognise faces and they can listen to sound and
distinguish a fugue from a conversation. How do I know that? because
that is what we can do. As I said, it may well require a quantum leap in
tech but it is not intrinsically impossible, though I suppose that it
does bring up the question of whather a machine can be conscious and to
paraphrase something John Searle once said (TASOC 1996) Can there be a
machine that thinks? Yes there can, because we are machines that think.

The main issue here is that we need to completely rethink how we
understand deterministic processes and break our philosophical thinking
out of the Cartesian mould, because it's getting green and smelly and it
is not at all like a good stilton. All the things that minds do are done
with strictly deterministic technologies, it's just that there are a
myriad of bits and pieces doing it and the complexity of the
organisation is orders of magnitiude above our currrent imagination of
what machines are, even if the machine we think about is the latest
computer chip.

I know I'm talking about AI and you're all just trying to figure out
what shelf to put that video on but it may well be that I don't think
like you and that you're classifications are completely alien to me.

Anyway, sorry a bit of a rave, but it is a very interesting problem and
I do think it will take some sort of radical solution that does not
involve presets. I keep thinking of the simulation problem. The best
description of the universe is itself.

cheers

Stephen Jones