2014年1月13日月曜日

Fiat Lux

I'll write about artificial life and such, starting from bonsai.
bonsai r6

I've always been thinking, existing evolutionary systems are too simple in their environments (or physics). Granted, super simple and abstract rules can result in complex and interesting phenomena (class 4 in Stephen Wolfram's terms).

The problem is, such game-theoretic phenomena are hard to notice and often completely useless outside that specific system. What I want to see is phenomena or morphology worthy of bio-mimicry.

Now the real challenge arises; how to achive such system. Ideal solution would be to create atomic scale system and pour in massive human and computation resource. (i.e. creation of autoverse in Permutation City) In real life, it's not feasible to create such system and wait until whole ecosystem of living things emerges; we need to pick a specific kind of life and tune everything to make them survive.

Plants seem good choice for a few reasons:

  1. No brain (fast iteration)
  2. Human-scale (obeys familiar physics)


In the first few iterations of bonsai, I was actually targetting recreation of embryogenesis at maybe somewhat reduced scale and even watched some online lectures on cell biology and plant biology (also, bonsai is written in C++ at that time). But it became gradually apparent that tuning such system so that something remotely similar to what we call plant to sustain is extremely difficult and fruitless. Simply put, we don't have enough real-world data to tune many parameters. Also, it'd be extremely fragile (from lack of computation resource), so no interetsting evolution will occur.

This is why I switched to more light-weight, artificial design approach. In bonsai's current state (r6), the following aspects are simulated.

  • oragan-level morphogenesis based on some coding system (genome)
  • downward lighting and photosynthesis
  • asexual replication with genome mutation
  • pseudo mechanical simulation (just a bunch of sanity checks on shape)
It starts off with 9 identical genome, and for most of the time, they die away. However, some plants do look good (to my eyes) and survive for fair amount of time, so I'm thinking total population is too small. Next revision will have re-written light simulator with accelerated ray tracing.