时当月色Life As It Could Be
Ray sees open-ended evolution as a key element of life, and being able to represent organisms as self-contained programs, without predefined rules (or a minimal number of rules) allows behavior to emerge that might not otherwise have been known or predicted. It is this possibility of exploring life as it could be that holds such promise for the suggestions of this paper, and the speed and flexibility of computer-based experiments allow for creative and efficient exploration many possibilities.
Clearly, exploring possibilities for life could also inform our understanding of terrestrial biology, as well. Steven Levy (1992), in his popular book on A-Life, writes, "By simulating a kind of life different from that which we are familiar, A-Life scientists seek to explore paths that no form of life in the universe has yet taken, to better understand the concepts and limits of life itself".
Noting the potential for A-Life to strike a healthy intellectual balance, Daniel Dennett (1991) writes: “Artificial Life…can be conceived as a sort of philosophy—the creation and testing of elaborate thought experiments, kept honest by requirements that could never be imposed on the naked mind of a human thinker acting alone. In short, A-Life research is the creation of prosthetically controlled thought experiments of indefinite complexity.”
This possibility for studying for life as it could be makes A-Life a powerful tool, and is the critical enabling feature of A-life for the suggestions of this paper.
AN ARTIFICIAL LIFE APPROACH
Characterizing Life
Defining life has been a notoriously difficult and elusive task. Defying definition might at least suggest the possibility that in an important sense, life may be indefinable (Matthews, 1996). Whether or not life is definable could turn out to be an important issue, but a more practical approach is to consider using several or many working characterizations of life without dwelling on whether or not those characterizations are perfect. In fact, as will be discussed later, a perfect definition of life (or whether such a definition is even possible) is not important for the approach being suggested in this paper. What this approach calls for is the abstraction of sufficiently general, fundamental, and minimal characterizations of terrestrial life in the hope that computer experiments based on those characterizations produce life possibility spaces with universally relevant degrees of variation.
Open-ended Evolution and Self-replication
Tom Ray has suggested only two themes for defining life: self-replication and open-ended evolution. Open-ended evolution essentially implies unpredictability which is an essential characteristic on John Maynard-Smith's
3
F3.2-B0.6-0015-02
list of life defining characteristics (Bedau, 1996). For Ray, open-ended evolution of synthetic life should “evolve structures or processes that were not designed in or preconceived by the creator” (Ray, 1996, p. 112). It’s noteworthy that Ray's use of open-ended evolution does not necessarily require natural selection, or any particular mechanism, as the vehicle for that evolution. Even if we were to choose one of the two minimal characterizations suggested by Ray, we're still left with important challenges. Why should evolution of any kind be a necessary characteristic of life? Similarly, why should “self-replication” be a necessary characteristic?
Complexity and Organization
Life is often characterized as being highly complex and organized. But how do we reconcile this when thinking about the origin of life? Are we to think that the initial sparks of life, or the first life-forms, were sufficiently complex? We may not have to because Reggia et al (1993) discovered that self-directed replication could be possible for very simple systems. But if indeed the initial steps in forming life did need to be sufficiently complex systems, perhaps for example, as forms of auto-catalytic networks (Kauffman, 1995), then this suggests the possibility that hitting a certain threshold of organized complexity may make the difference between successfully opposing the Second Law of Thermodynamics or not—on local space and time scales.
Opposing the Second Law
Stuart Kauffman’s more recent work, Investigations, suggests that an "autonomous agent" is "able to reproduce itself and carry out one or more thermodynamic work cycles" (Kauffman, 2000). This is a minimal definition of life that appeals to the phenomenon of opposing the Second Law of Thermodynamics on local space and time scales.
Chris Adami suggests "Life is a property of an ensemble of units that share information coded in a physical substrate and which, in the presence of noise, manages to keep its entropy significantly lower than the maximal entropy of the ensemble, on timescales exceeding the "natural" timescale of decay of the (information-bearing) substrate by many orders of magnitude" (Adami, 1998). The key notion in this definition is keeping entropy significantly lower than the maximal entropy of the ensemble on timescales exceeding the “natural” timescale. It is also significant that this definition does not include self-replication as a necessary characteristic. The definition could be made even more general by removing reference to information, units, and substrate: Life is a property of an ensemble of interactions which, in the presence of noise, manages to keep its entropy significantly lower than the maximal entropy of the ensemble, on timescales exceeding the "natural" timescale of decay of the system by many orders of magnitude.
This is another example of a very general and minimal characterization of life. Exactly how low the entropy needs to be and how many orders of magnitude longer in time the structure and/or function is maintained are important details that will not be addressed here. The primary point is that this characterization serves as an example of the kind of thinking that could be helpful for the purposes of the approach suggested in this paper. Using Artificial Life Computer Experiments to Generate Life Possibility Spaces
Artificial life simulations vary widely, which is part of the appeal of the field. Given the characterization of life suggested above, data such as natural relaxation times for ensembles (e.g. collection of molecular structures) would have to be a hard-coded in a computer experiment environment, as would basic chemistry.
Some measure of structure and/or function maintenance would also be required, perhaps as a kind of dynamic fitness function. The preferred aim would be the largest number of experiments possible based on a minimal characterization of life. Preferably, the experiments should be run for the computational equivalent of natural evolutionary times in order to allow for sufficient time to generate possibilities.
Knowing when enough experiments have been performed would be difficult, but we might approximate or derive such a number by considering the number of significant evolutionary events, or key evolutionary junctures, that may have occurred on earth. For example, the number of key evolutionary pergences in the earth’s history might be used, either directly or more indirectly, to obtain the number of new experiments required.
If this method seems problematic, we might consider that experiments could be run until sufficiently perse possibilities were generated. If, after running a reasonable number of experiments, much variation was generated, this might signal that enough experiments had been run to satisfy this approach and hence suggest that terrestrial
4
F3.2-B0.6-0015-02
life may not be indicative of life in general. If, however, this approach continually required an usual number of experiments before something sufficiently new was seen, this could signal that the approach will not likely yield a sufficiently perse possibility space, perhaps either because there were problems with the experiments (e.g. not sufficiently breaking out of terrestrial biology boundary conditions), or perhaps more importantly, because terrestrial life is indeed indicative of life in general.
The number of experiments may also be constrained by time and computational power. Just how constrained the number of experiments would be could turn out to be an important practical limitation of this approach. But again, if the number of runs is statistically sufficient, it may not be that important, and with today’s computational power, many comprehensive experiments should be possible.
Applying Constraining Knowledge to the Possibility Space
This approached is based on the requirement that the computer experiment design and results be properly constrained, so that anomalous results do not mislead assessments of the potential variation (Sullins, 2001). A-Life may be able to produce alternative life possibilities, but so can the human imagination—as the abundance of science fiction wonderfully demonstrates. We also know that while imagination can lead us in fruitful directions, unchecked imagination can also lead us astray. Part of what this approach is striving for is the production of possibilities consistent with physical laws and whatever “laws” of biology are thought to be relevant.
In theory, if the rules and initial conditions are set up properly, then all the possibilities generated should be viewed as plausible alternatives. However, mistakes could be made, sufficient knowledge might not be applied to the rules, or a condition could arise as part of the experiment that has no plausible physical basis. Some results, therefore, might have to be dismissed, but perhaps with unusual caution given that this approach is trying to create alternative biologies.
Significance of Remaining Variance and Practical Relevance
The approach outlined in this paper suggests that if a high number of significantly varied alternative life possibilities emerge from properly constrained open-ended evolutionary computer simulations, then we might consider that terrestrial life is not necessarily sufficiently representative of life in general, because such a varied range of alternative life possibilities could be physically realized the universe. Research based on terrestrial biology as applied to extraterrestrial life issues should then be understood in this context, namely that terrestrial biology may not be typical, perhaps giving rise to caution regarding the confidence we have in applying that terrestrial biology knowledge to extraterrestrial life issues. This could affect planetary protection issues such as forward contamination associated with a human mission to Mars (Lupisella, 1999, 2000). It may also suggest the need for more novel, more efficient, and more varied search and detection strategies such as a cooperative robotics approach which might use many varied single function detection agents that can cover a large area, and that can cooperate to apply the fullest range of functionality needed at any given location of interest that is detected by any other robot(s) (Lupisella, 2002). High variability may also suggest the need for rigorous longer-term contingency planning given that detecting and positively identifying an alternative life form could prove much more elusive and require much caution.
