Most of us have pondered this question at least at some level during our lives. But can we quantify in any sense what life is and what properties we attach to it? Recent work, by several groups around the world, in modelling artificial life systems is trying to do just this.
At what degree of sophistication do we decide that something is alive? There is plenty of discussion in biology about the status of real life-forms from viruses to bacteria, cells and above. What are the properties that make something "alive"? Some definitions might be an ability to autonomously reproduce and evolve in some sense.
We can further complicate the debate by thinking about "intelligent" life of course but leaving that aspect apart - is a simulated entity that exists solely in a computer-generated world "alive"? In terms of the statistical mechanics and information theoretical properties of such an entity, is it really different from a carbon based "real" life-form?
Philosophically I suppose we all have our in-built prejudices and maybe most of us do not like the idea that there is nothing special about us as biological life forms. Whatever one's individual prejudices, there are some interesting ideas to be considered - quantitatively by simulating and analysing the emergent properties of "artificial life" entities.
We have been working on a simulation environment for artificial life forms. As discussed in a recent talk on the thermodynamics of artificial life, the goal of this work is to get a handle on exactly how life models explore their phase space. The model can be said to have a very large "space" of possible states. We can initialise it in various random or semi-organised ways. There are some "attractors" or areas of the model space that it always seems to end up in. The whole investigation becomes more than just a game when we systematically start cataloguing those special attractors and relating them to particular animal behaviours.
The idea of computer-generated life-forms dates back to John Conway's famous "Game of Life". This is a well-known cellular automaton program and in it a number of interesting patterns emerge from a simple set of rules for the "bits" that live on a square lattice. Our model is more sophisticated than Conway's - but it seems it does not have to be much more sophisticated to yield interesting behaviours.
It is now widely believed that anything we might reasonably call "life" is sufficiently complex that it never reaches what we normally consider an equilibrium in terms of statistical properties. Consequently there has been discussion internationally on whether we need a "4th Law of Thermodynamics" to explain or at least characterise life systems. At any rate, we have found some quite remarkable emergent properties in a very simple model.
We consider a set of rule-based animals that life in a simulated world. They can consume food, move around and most importantly breed. Sex of course in the simulated world is so much less messy than in the biological world! What has turned out to be most interesting in our model are the effects of having two separate species interact as predator and prey. We model incredibly simplified "rabbits" and "foxes" living in a flat world and find that although the model has no special patterns built into it by us, some quite distinctive collective behaviours emerge as animals start to interact in large numbers. It turns out that spiral wave-fronts emerge as quite beautiful and consistently repeatable patterns as one group of animals flanks another to attack or escape predators. These patterns appear to be analogous to the flanking-front behaviours common in spontaneous troop movements during a battle.
A great deal of study is underway around the world on genetic programming methods and evolutionary computing. At one level these studies have the analogous goal of "breeding a better rabbit" or a "better fox" from some starting (guessed) rules. This is a useful approach to many problems, but the artificial life approach focuses on the collective properties of the microscopic entities.
Although the properties of a model rabbit or fox are rather simple and at present are controlled by us, what about the collective herds or swarms that spontaneously form in the model. Could these be said to be emergent life forms in some sense? We do not program a particular swarm - it just comes into being as a consequence of the microscopic interactions of individual animals. The compelling analogy of course is the relationship between simple cells and higher-level life-forms in "real biological life".
We do not of course consider our rabbits and foxes to be alive - they are just programs of machine rules that we set up - but seeing a collective swarm form migrate across the simulated world and consume another swarm makes one wonder about what makes these emergent collectives different from real life forms.
It is philosophically appealing - to me at least, that there are quantifiable mechanisms and procedures that we can investigate and which might give us a handle on life itself, without recourse to mysticism. It has been speculated that we will have powerful enough computers in 2020 to simulate a similar numbers of neuronal components as the human brain has. Perhaps we might therefore expect to be able to generate emergent artificial intelligences by then. How many microscopic entities do we need to simulate before we can say we have an artificial life-form?
One can imagine some amusing future events. A notice will come around warning us not to switch certain computers off at night as it would "kill the artificial life forms living in them". We might have to apply to the University Ethics committee before we undertake certain programming projects as it might "have consequences for artificial life-forms." We might not be allowed to work on genetically modified artificial life-forms in New Zealand. Perhaps our next supercomputer will have to have biohazard warning stickers on it.
Isn't "Life" interesting :-)
Ken Hawick, March 2005.