Semantic Memories, and other SW developments

Just trying to keep on top of some of the recent SW developments... this
thing appears to be changing subtly, but noticeably.

Seth Russell has recently been discussing "Semantic Memories" and "Clouds",
natural language. While Seth can provide more in depth investigations than
I can, an SM (Semantic Memory) is basically a flat database of the triples
that one has browsed in the URI information space (i.e. the Semantic Web).
It has a lot to do with contexts of URIs. A URI in the context of my own
Semantic Memory may be very different from the one in yours. This is good
for the system to remain decentralised and stable, and seems to me to be an
intuitive next step to the development of the SW.

I quizzed Seth about interchange of Semantic Memories, and he thought that
this may be the way in the SW will grow, e.g. making your SMs available as
XHTML, possibly by embedding different representations of your SMs. This
resolves (for me personally) one of the most troubling aspects of the SW -
that of formats.

When people adopted the WWW, they adopted URIs. Many also adopted HTTP and
HTML, but you can quite easily have an FTP link to some graphic, but that
FTP link is a URI. Everything on the Web is representable as a URI, and
vice versa, the Web consists entirely of URIs (feel free to argue this in
detail, but note that it isn't all that important to the point I am about
to make). Basically, the fact that most of the world communicates in HTML
was a bit of a fluke. I am convinced that the whole world isn't going to
start making their information available in proper "XML RDF", but that
doesn't matter as long as they have a triples store. Even triples aren't
all that essential, as long as information is stored as a mix of URIs and
string literals.

By having our own SMs, we get rid of the need to have a single format for
the Semantic Web: we each choose our own as long as they operate on the
same decentralised principles of the WWW information space, and the basic
SW tenets of triples. Then, any logical and ontological layers that we
apply above that can be done so in isolation, making sure that nothing
interferes with our systems.

This doesn't stop people publishing to the WWW on servers, and indeed it
may encourage it for it assumes that people will download information into
their SMs, and to do that there needs to be information already
available... Web interaction is still going to be vital. But we will also
be able to have our own personal spaces (possibly also in URI space if you
maintain an HTTP server) that we can explicitly trust, and therefore make
solid first party inferences from.

It is clear that the SW is not at all evolving like the WWW did in its
early years. There are some similarities: lack of people "on board", the
nodes and arcs, and so on, but the differences are in the details. The SW
will be more personal, in that contexts of data are *very* important,
whereas documents have almost no contexts whatsoever. Proof and so on has
been proffered by TimBL and others as a potential solution to this, but I
don't think it cuts it: if you are downloading megabytes worth of URIs and
have to make trust assertions about each bit, then that is something you
are going to want the machine to do. Trust is very much a human factor, and
not something that can be decided by a machine. By separating the SW into
small areas of work, it becomes easier to trust certain areas.

The idea is that the SW will be rather fractal and holographic in origin:
everything is connected to everything else, but the further away you get,
the less you trust it, and the less useful and relevant it becomes. That
doesn't mean it is not useful, only that the links you make to get to that
bit of information might mess up your system. By isolating the information
you need, and processing your own SMs, it should be possible to provide a
way round these problems.

Overall, I like the direction in which the WWW has gone, and the direction
in which the SW is going. The emphasis at this point should be on
pragmatism, not abstractions, so I apologise for this rant, but I feel that
it is something that I wanted others to comment on and share at as early a
stage as possible.

Kindest Regards,
Sean B. Palmer
@prefix : <> .
[ :name "Sean B. Palmer" ] :hasHomepage <> .

Received on Monday, 19 February 2001 22:21:08 UTC