sabren.net archives
back to sabren.net...
back to the
archive...
entries for 2002/10/08
Sarles
Next summer, I'm going to Sarles, North Dakota for a family reunion.
The census PDF file says that in 2000, Sarles had 25 residents.
Men outnumbered women fourteen to eleven.
There were two people in town that were too young to vote --
one boy and one girl. They were also the only ones too young to
buy drinks. The median age in Sarles was 56 and a half. Four inhabitants were native american. The rest were white.
There are 44 houses in Sarles. Sixteen of them are occupied, one of them by my grandmother. All sixteen are owner-occupied (no rentals).
There is also a public swimming pool, a post office, and a school, which shut down years ago.
Sarles is located in the northwest corner of Cavalier County, about three miles south of some place called Canada, and a mere 1605-mile drive from my front door.
The capital of Cavalier county is Langdon. There's an empty apartment building for sale there. It's selling for twelve grand.
Sarles gets cold. The National Climatic Data Center says North Dakota's average temperature is about 40°F. Mean temperature reaches into the low 70s during the summer, and nears zero in the winter. As I write this, weather.com reports the temperature is 42°F, and that it feels like 36°F. The wind is coming from the North/Northwest at 10 miles per hour.
A long time ago, my grandmother told me you could buy
real estate in Sarles just by paying the back taxes on abandoned
properties. I always thought it would be fun to buy some land in
a place like Sarles, and remake the place: Bring in a T-1 and set up wireless internet across town. Invite smart people to come live
there for practically free. Invite people who like to build things and put up funky houses. Connect the buildings for when it's cold. Have a big greenhouse, get rid of cars, set up a free school.
Try out a bunch of sustainable living concepts.
Or maybe I should just play SimCity.artificial life
For the longest time, I've been thinking of montybot (my hypothetical stock trading system) as an algorithm. The whole premise was to evolve a long string of instructions (procedural approach) or a tree of expressions (functional approach), and that these structures
would more or less be the thing that went off and actually made trades.
Steve Grand makes a good case for not relying too heavily on genetic programming:
[T]rying to speed up evolution in the face of large numbers of degrees of freedom is rather like trying to reach escape velocity by streamlining one’s bicycle—sometimes the numbers are just intractable and another method must be found.
Steve Grand is my hero. He invented Creatures, and wrote
Creation: Life and How to Make It. He says it's
futile to simulate intelligence directly, but intelligence can arise naturally in creatures built from simulated pieces. He often writes that a simulated neuron isn't really a neuron, but a brain made from simulated neurons is a real brain. (His other writings explain this idea far better than I can.)
I agree with him. I'm starting to see montybot not as a script, but as a living creature. Now, one of my earliest concepts for montybot was biological, but back then, I imagined
I would simulate creatures directly. The "creature" would be an object with hard-coded behavior. But, to Steve Grand, that's like building a racecar game by coding things like skidding and spinouts explicitly, rather than modelling the laws of physics and letting these things emerge.
The simple fact is, randomly generated code isn't likely to run.
I had planned to solve this problem with heavy use of the mod operator, which works like mathematical division but only returns
the remainder ("12 mod 5" yields 2). If you have 10 possible instructions (0..9), then any integer mod 10 will map to a valid instruction. Even so, I suspect the code would be dangerously brittle.
But... Using Grand's approach, I would start with neurons and organs and biochemicals, and genes that said how to turn these parts into creatures. The creatures would live in a virtual world, and compete
for resources that correspond to money in the stock market.
Creatures that got too poor would die, while the wealthy creatures
live to breed another day.
Evolution would progress much more slowly, because while code either works or doesn't, the creatures can adapt and learn. And if we have a whole population of creatures, and if they are given the means to communicate, then they might be able to learn from each other.
With this biological model, the forces of evolution, learning, and memetics all work to solve the problem. And why stop there? Why not provide the creatures with tools? Brains aren't usually great at
math, but suppose mathematical expressions were physical objects
in their world? Functions like sums and averages could be carried
around like calculators. Or what about a device that identified candlestick patterns? Externalizing these tools from the intelligence of
the system would allow the creatures to develop their own technology. How cool is that?
I can't wait to get through this cornerhost automation process
so I can start working on this stuff. The biological concept seems
far more promising than just evolving random code... Not to mention a
whole lot more fun!