Semantic Symposium at NYU

Since I was in Manhattan at the time, I attended this really cool event, which featured lots of really smart people basically knocking their heads against the (intractable) problem of AI reasoning based on doing entailment from semantic knowledge, obtained from texts.

It was fascinating to see the attacks that everyone was using to try to slay the dragon. Just about every approach you could imagine was presented. Everyone with something concrete was susceptible to questions - how do you deal with temporality, what about metaphor, etc - to which they would say, "I haven't thought about that yet", or even better, "well we haven't considered these issues yet, but we are about to field test out system with a large company and so we will soon know if it works".

The emperor truly has no clothes.

Still, for someone with lesser goals (a tool to support human reasoning) the mechanics that were presented were fascinating.

Currently Travelling

My blog posts have transmogrified into travel entries as I am currently circumnavigating the globe. Here for more details...

A Vision of Life

There's an article on CodeProject that shows a genetic algorithm tuning a neural network to simulate "ants" finding food. The ants start out kind of dumb, just milling around aimlessly (or more accurately, randomly). But some ants randomly have better food collection strategies (small direction changes, fast velocity) that means they collect more food than the other ants. At the end of each "generation" the low scoring ants are unceremoniously removed from the gene pool and the high scoring ants spread their successful food gathering strategies to their ancestors.

It's a grim view of life really. Are we all nothing more than the bearers of a successful food gathering strategy that has been passed down through the generations? Apart from the random droppings of food in the simulation and the ants' disturbing mechanical persistence at finding more and more of it, there is only darkness.

Semantics and Search

Right now I am working on different projects at Dog Blue Software that revolve around natural language processing and applying it to Internet search. Natural language processing looks at how all the words on a page are used in sentences and giving the computer some idea about what all those words mean (the semantics).

Working with information from the Internet is challenging on a number of levels. There is the amount of data. There are the limitations of the interface. There are the multiple languages and idioms.

Some people talk about "The Semantic Web" as some kind of saviour to the problem of sifting through large amounts of unstructured data. Clay Shirky neatly critiques the limitations of such a concept, even if it were to coalesce into being.

Rather than a simplistic solution - "we'll just ask everyone to become an Internet librarian" - or Google's needle in the haystack solution - "we'll give you ten needles that everybody already knows about, you can forget about all the other ones" - the solution to navigating large information spaces will probably require a level of sophistication bordering on artificial intelligence.

But until then, letting the computer find the meaning of text rather than its representation is an idea that seems to have promise.