Individual project proposal: Artificial Life

Proposer: Iain Stewart, room 220a Huxley, e-mail ids@doc.ic.ac.uk

[An artificial Martian!]

The discipline of artificial life is young and still largely unexplored. It is not the same as artificial intelligence, which concentrates on computer models of high-level, "intelligent" thinking processes. Artificial life can be about any computer-based (hardware or software) entities that display some or all of the hallmarks of life: reproduction, growth, cooperation and/or competition, self-defence in an environment, opportunistic expansion into new environments, evolution, adaptation, etc. Artificial life can often give us new insight into the behaviour and evolution of real life. It can also simply be interesting in its own right!

Artificial life has already escaped "the laboratory" in the form of computer viruses. In more controlled environments (simulated worlds) people have set up artificial ecosystems such as Tom Ray's Tierra or the Core War game environment, and looked at how different artificial life-forms cope.

I would be interested in a project that attempts to build and/or evaluate such artificial life systems. I'm especially keen on "naturalistic" simulations, i.e. simulated worlds which are based (even loosely) on the sort of physics-based constraints real creatures face, like limited resources, a space or territory which must be navigated, limited sensory or motor activity and so forth.

You could develop such a simulated world and then either try to hand-design artificial life-forms that do well, or more intriguingly, allow evolution to occur in the population, and see what turns up!

Another possibility is to obtain an existing simulated-world package, and perform an analytic study of what behaviours go on in it. For example: does a greater variety of species make the system more stable? (This used to be confidently asserted about real life in biology textbooks, but people are now not so sure.) What about geographical diversity, or seasons, or just the effect of the model's size? Of course if you develop a simulated world yourself you can also study it in these ways.

Often the really interesting behaviour occurs at the level of whole communities rather than individuals. You could investigate flocking birds (following in the tradition of Craig Reynolds's boids), schools of fish, "packs" of hunting - or hunted! - creatures, and the like, or the behaviour of "superorganisms" like beehives or ant colonies.

A whole planet's worth of life - often called Gaia - is another interesting candidate for artificial life treatment. An exciting aspect of Gaia is the way it can have profound effects on its host planet. (The earth's atmosphere and climate have probably been greatly altered by life.) Does Gaia tend to stabilize a planet's climate, or destabilize it, or can it go either way? Jim Lovelock's Gaia hypothesis is (or was) that of stabilization, but the issue is a controversial one in planetary biology. "Artificial Gaia" simulations such as Daisyworld have been developed, to try and shed light on the problem. You could try your hand at artificial Gaia!

Another possibility: you could simulate the structure of a single creature (or a small number) in greater detail than you could afford to in a large population. Candidates for such greater detail include the brain - as a neural net, for example - and the immune system. Or you could see if an artificial life approach might shed light on the mysteries of morphogenesis. How does what starts off as a tiny embryo manage to follow an often stunningly precise developmental path - how does it control the timing of growth and differentiation, and make and re-make a complex layout of organs and tissues, with exquisite accuracy and reproducibility?

The origin of life remains shrouded in mystery. Artificial life could help shed light on the "prebiotic" phase of evolution, when molecules were just beginning to form structures with some of the attributes of life. For example, in the tile automaton model of Tomoyuki Yamamoto and Kunihiko Kaneko, simple patterns of activity loosely describable as metabolism and self-reproduction emerge from an "artificial chemistry" substrate. You could try building on their work, or creating something similar of your own.

Even the definition of life is a tricky and controversial subject! David Deutsch, in chapter 8 of his book The Fabric of Reality, suggests a new approach based on identifying "embodied knowledge" - such as the implicit knowledge coded over evolutionary time in genes - by what he calls its trans-universe structure: different "universes" (roughly, versions of what's going on in space and time) where for example some life-form has taken a decision in different ways, and reaped the various possible consequences, can end up more similar than simple combinatorial explosion arguments would lead you to think. Sites of life's activity can stand out from their background by virtue of this kind of trans-universe similarity. An ambitious project would be to investigate these or similar concepts of structure or knowledge - by simulation, mathematical analysis, or whatever - and see if they help with sharpening up concepts like "life", "thought", etc. as Deutsch hopes.

These are just sample ideas and I would be most interested in any other artificial life ideas or approaches you might come up with yourself!

Some artificial life links

Here are some web pages on artificial life you might like to look at.

This corner of the web maintained by [photo courtesy of the multimedia lab machines] Iain Stewart <ids@doc.ic.ac.uk>, Department of Computing, Imperial College, London, UK

Annotate the web - add your own comments to this web page or any other for all to see! (Only supported by some browsers, sometimes via an extension or the like.)

(If you're reading this from within IC DoC you can try Crit, an earlier way to annotate the web. Crit Me Now!)