Wednesday, January 31, 2007

Welcome to Hamath

Wikipedia says of "Hamath":

Hama is a city that is located on the Orontes in Syria, north of the city of Homs.
That's how we open our introduction to the Hamath blogger, which has nothing to do with the city "located on the Orontes in Syria." Simply, Hamath is the juxtaposition of HAM and Mathematics, and HAM is an acronym for Henry, Abel and Mike.

Once upon a time HAM used to be HAM Computers. We promised the friendly citizens of Tustin cheap, reliable and skillful computer repair/construction.

But now HAM combines with our love of Mathematics -- because that is how the three of us were bonded together originally -- to form Hamath, a blog of science, technology, and interesting tidbits we find interesting. It is as much for readers as it is for us; through Hamath we hope to learn and appreciate the ever-accelerating frontier of science, and bring that frontier to you.

We hope you find our blog posts interesting. Also, this blog isn't just a one-way channel of information; we think of it as two avenues of conversation between you and us. So feel free to add your thoughts and comments, to build a forum of discussion.

Project Hoshimi

Project Hoshimi looks extremely interesting; it's a programming competition to develop AI code to control a bunch of nano-bots in the quest to cure a human body. Of course this is all virtual, these nano-bots are only video-game entities -- but I would like to say that the AI code is very real. The informational session for Project Hoshimi at USC is at 5:00 tonight, and I think I'm going to attend (and hopefully take part in this competition).

In addition, Project Hoshimi is apparently an international competition, with finals in Korea and prizes of $8,000. So if you think you've got the smarts and hacks to create an intelligent nano-bot immune system, join in!

I haven't taken a deep look at the details of Project Hoshimi, but I'm pretty sure the coding will involve techniques from flocking-AI. In other words, the approach for this will be trying to get some overall intelligent behavior from an aggregate of relatively stupid constituents. For example, have you ever noticed a flock of birds flying around? As a whole, the flock seems to be fairly intelligent. It doesn't "run" into telephone poles, the birds seem to know how to separate and rejoin around the pole. But when you look at each individual bird, they're pretty dumb. They only know two things: don't fly into things, and try to maintain a constant distance from all your other bird-mates. When you put a lot of these agents together, you get flocking behavior.

Flocking agents are pretty low on the ladder of AI complexity, but it does serve as a basis for complex intelligence as emergent from the large sum of simple agents. I have a belief that real intelligence (human-like intelligence) can be constructed from smaller parts. What's complex is not the constituents themselves, but the interactions between them. In physical terms, we already know that's how the brain functions: it's just millions and billions of neurons that only know one thing: fire if the signal reaches a certain threshold.

I don't propose that we try to emulate the brain by building the electronic equivalent of neurons, but we should examine, in more detail, emergent properties from interacting agents. A rigorous system of mathematics to describe such dynamic networks would be indispensible towards constructing "true AI."

Follow up on Hoshimi: After attending the informational meeting I am really excited about Project Hoshimi. Apparently it is sponsored by Microsoft (which explains why it only works with .NET languages), but people with no coding experience can take part too -- there's three modes, Beginner, Intermediate, and Expert Mode. The first two require no coding experience, and give you a really flexible and powerful user interface to control the AI of the nanobots. The Expert mode gives you the tools to write your own AI code, which is really cool too. So I encourage all of you to take part at least in Round 1 of the Hoshimi competition, you have until sometime in March to pass a really simple, simple test.