时当月色The opposite might also be implied in that if novel alternative biologies do not readily emerge, or occur in small numbers with limited variation, then perhaps we might have increased confidence in applying terrestrial biology knowledge to extraterrestrial biology issues. Central to these assessments is the notion of similarity (or dissimilarity) which would have to better defined, preferably quantitatively, in order to assess the extent to which certain outcomes are similar or dissimilar to terrestrial life and to each other. Komosinski et al. (2001) have begun to try to quantify dissimilarity between artificial creatures, so it would appear possible to assess similarity between artificial and real life-forms, qualitatively as well as quantitatively. Such an assessment might be measured at many levels such as molecular, biochemical, and structural deviations from DNA, a more primitive RNA molecule such as 16sRNA, or an RNA-like molecule thought to be a plausible theoretical precursor. If enough experiments were run, degrees of reproducibility may help quantify the probability of outlier results, and patterns among different simulation results might emerge, perhaps then pointing to possible methods for quantifying similarity, making the task of measuring similarity less difficult than it might first appears.
The possibility spaces would have to be analyzed to see if qualitative or quantitative categorizing could be done to play a role in assessing the variability of the simulation results. For example, there may be 100 different possibilities that arise from 100 different simulations, but some, or many, of these may be sufficiently similar so as
5
F3.2-B0.6-0015-02
to be grouped together into one category or subcategories. A distribution of possibilities would help assess the overall variability of the possibility spaces if such categorization seemed reasonable. For example, a terrestrial life category would be an obvious choice, leaving the variability of the remaining possibility space to analyze. These assessments will not likely be easy to make, and would of course not be determinative, but hopefully informative at least at a macro level.
ANALYSIS
Characterizing Life
Perhaps the most difficult problem with this approach is appropriately characterizing life. Some of these difficulties have been touched on, but there are potentially deeper concerns.
The first concern is a kind of circularity problem based on the fact that a characterization of life must be derived from characteristics that are indicative of terrestrial biology. Any such basic conception of life would be based on our terrestrial knowledge, and hence, arguably, our terrestrial biases, since we have no other physical data. Yet we are trying to transcend terrestrial biases. This “one data point” problem—the one data point being all terrestrial life which is part of the same phylogenetic tree—is a serious epistemological challenge. However, as indicated above, it does seem possible to establish quite fundamental characterizations of life and use them as the foundations on which to run A-Life computer experiments. If the characterizations are sufficiently minimal, terrestrial biases will be somewhat mitigated by producing broad possibility spaces. It is imperfect for sure, but the power of A-Life simulations to generate possibilities may be sufficient to make the assessments this paper suggests.
A deeper challenge is that life may be intrinsically, or at least epistemologically, indefinable. Defining life could be a deep metaphysical and/or epistemological problem in the sense that (1) there may be no clear boundary between life and non-life, (2) we don't know enough to make a clear distinction, or (3) that while there may be a universal characterization, we have our own insurmountable epistemological limitations that would prevent us from obtaining a complete and indisputable characterization of life.
If life is indefinable, then claims of universality are suspect. But, since the approach being suggested in this paper is about asking whether potentially realizable life possibilities are sufficiently varied, the universality claim has primarily to do with whether that degree of variance in the possibility space is universally relevant—that is, that it is possible for the universe to generate that level of variety—not whether the specific results of each experiment outcome are universal. This is one of the main reasons for trying to establish the most minimal characterizations of life and appropriately applying constraining knowledge on the possibility spaces—to avoid unduly restricting results while still maintaining the relevance of the possibility spaces and any associated variance.
One specific practical way to deal with this problem in terms of the approach suggested in this paper is to use many minimal characterizations of life in many different series of experiments, each series or life possibility space being based on its own minimal life characterization, to generate many life possibility spaces, each of which could be analyzed in isolation and relative to each other.
A-Life Computer Experiments vs. Reality
A justified and pervasive concern about models and computer simulation/experiment results is their relation to physical reality. Because the A-Life simulations suggested in this paper are of an open-ended nature, we might think we should be especially concerned about the extent to which the simulation results, i.e. the life possibility spaces, are consistent with physical reality. However, as mentioned in the previous section, it is not the claim of this paper that the simulation results will represent actual or future natural biological phenomena. The suggestion here is only that if the computer experiments produce a sufficiently varied possibility space, that universe may also be capable of generating such variation, and that as a result, we should take seriously the possibility that terrestrial life may be only one of many kinds of life, perhaps calling into question the confidence we might have in applying terrestrial biological knowledge to extraterrestrial life issues.
This suggestion will be supported based primarily on the extent to which this approach can produce optimal life possibility spaces, where ‘optimal’ means that the possibilities are physically plausible and that the emergent variation is indicative of potentially realizable variation in the universe. That is, the experiments may produce results that may or may not ever occur in the universe, but that could occur (and whether they do in reality could depend entirely on random factors). Similarly there might be other possibilities and actual natural realizations not represented in the possibility spaces. It is also possible that many or all of the results might never be naturally
6
F3.2-B0.6-0015-02
realized in the universe because of many factors, including random factors—a cornerstone of evolution as we understand it.
Is this Approach Feasible?
There are many “ifs”: (1) If appropriately minimal characterizations of life can be established, and (2) if chemistry can be properly programmed in an open-ended evolutionary simulation environment and (3) if experiments based on those characterizations can be run to produce a optimal life possibility spaces, and (4) if we can properly analyze that possibility space for variance, and (5) if we have some idea of the threshold variance that would signal calling into question the confidence we might have that terrestrial life is sufficiently representative of life in general, then we could make at least a first order assessment as to the extent to which terrestrial biological research can be confidently applied to extraterrestrial life issues. Even if all of the above are possible, there is no logical necessity that the artificially produced variance will necessarily be indicative of the kind of variance the universe is capable of, and this will be a difficult open question for broader community to bear in mind. However, for the purposes of a first order assessment in the absence of a physical example of extraterrestrial life, this approach may be worth considering.
The characteristics of Artificial Life covered previously in this paper suggest that Artificial Life has the promise to produce artificially generated physically relevant alternative biologies1, where presently we only have one data point, namely terrestrial biology—a situation we may be in for a long time. Indeed, part of the motivation for the approach in this paper is to ensure we take proper precautions in maintaining the scientific integrity of that “second data point”. As is often the case in science, a broad theoretical framework can facilitate a better understanding of terrestrial life, and perhaps more importantly, if other kinds of life are possible, a better understanding of life in general.
This paper has tried to show that while the Artificial Life methodology presented here presents significant challenges and would likely be a long-term research program, there appear to be no fundamental theoretical barriers preventing the potential of this approach from being realized.2 Despite the theoretical and speculative nature of this kind of research program, the continued absence of biology’s “second data point” suggests that such a research program should be considered to help inform issues regarding possible extraterrestrial life. ACKNOWLEDGEMENTS
Many thanks to Fred Suppe, Jim Reggia, John Andersen, and Benton Clark for their guidance and comments. REFERENCES
Adami, C., Introduction to Artificial Life, Springer-Verlag, Santa Clara, 1998.
Axelrod, R., The Evolution of Cooperation, Basic Books, New York, 1984.
Bedau, M., The nature of life, in The Philosophy of Artificial Lif e, edited by M. Boden, Oxford University Press, Oxford, pp. 332-357, 1996.
Boden, M., The intellectual context of artificial life, in The Philosophy of Artificial Life, edited by M. Boden, Oxford University Press, Oxford, pp. 1-35, 1996.
