F3.2-B0.6-0015-02 USING ARTIFICIAL LIFE TO ASSESS THE TYPICA

F3.2-B0.6-0015-02

USING ARTIFICIAL LIFE TO ASSESS THE TYPICALITY OF

TERRESTRIAL LIFE

M. L. Lupisella

NASA Goddard Space Flight Center, Greenbelt Rd, Greenbelt, MD 20771

University of Maryland, College Park, Department of Biology

ABSTRACT

The extent to which extraterrestrial life questions can be confidently addressed rests in large measure on the extent to which terrestrial life is representative of life in general since we will have to draw from terrestrial life knowledge. This paper outlines a long-term research program that could inform the extent to which terrestrial life is representative of life more generally, which might then help inform our level of confidence in applying terrestrial life knowledge to extraterrestrial life issues. The approach involves appealing to the relatively new field of Artificial Life to: (1) use minimal characterizations of life in (2) a large number of open-ended Artificial Life computer experiments to generate "life possibility spaces" (3) the results of which can be examined for their plausibility within the context of relevant constraining knowledge, so that (4) the remaining results can be examined for variability relative to terrestrial life, where low variability might suggest that terrestrial life is typical of life in general, and high variability could be interpreted to suggest that terrestrial life might be atypical. INTRODUCTION

This paper will suggest an approach that in the absence of extraterrestrial life could inform the extent to which terrestrial life is representative of life more generally. This could then inform the level of confidence we might have in applying our knowledge of terrestrial biology and ecology to extraterrestrial life issues such as search and detection strategies as well as the interaction of terrestrial ecosystems with other possible planetary ecosystems. The approach involves appealing to the relatively new field of Artificial Life (A-Life) to: (1) use what might be the most minimal set of life-defining characteristics as the basis for (2) a large number of open-ended Artificial Life computer experiments to generate "life possibility spaces", (3) which can be examined for their plausibility within the context of relevant constraining knowledge, so that (4) the remaining possibility space(s) can be examined for variability relative to terrestrial life, where low variability might suggest that terrestrial life is not an anomaly, but is instead sufficiently representative of life in general. High variability in the possibility space(s) could be interpreted to suggest otherwise.

Definitions of Artificial Life

Chris Langton, the first to use the term, "artificial life", suggests biology has traditionally started from the top and "worked analytically down from there through the hierarchy of biological organization" (Langton, 1996), where ‘analytically’ implies the separation of a whole into sub-elements which can be studied inpidually. This approach seems to have provided a fairly broad picture of the mechanics of life on Earth, but Langton suggests that the dynamics of life have largely gone unexplored because dynamics is concerned with the interactions between parts, which disappear when isolating parts for investigation. Systems with such strong interaction dependency are thought to be non-linear and to require the synthesis of systems to form a coherent whole in order to understand the suite of interactions and how they give rise to overall system behavior. This, according to Langton, is what A-Life attempts to do. This is often accomplished through simulations based on genetic algorithms which are computer

1

F3.2-B0.6-0015-02

algorithms/simulations in which populations of candidate solutions to a problem are stochastically selected, recombined, mutated, and then either eliminated or retained, based on their relative fitness, where fitness is defined by a fitness function against which the effectiveness (i.e. fitness) of any given solution is measured.

Langton's general conception of A-Life also includes "wetware" which involves wet-bench lab techniques and experiments that use real natural life components in a laboratory environment to direct an artificial evolutionary process toward the production of other real natural life elements such as RNA (Taylor and Jefferson, 1991). However, in trying to assess the degree of universality of terrestrial life, wetware may not be the best approach because of its dependency on what are arguably quite specific biochemical configurations and processes.

Margaret Boden, drawing from much of Langton's thinking, emphasizes A-Life as a field which "uses informational concepts and computer modeling to study life in general, and terrestrial life in particular" (Boden, 1996). Thalmann and Thalmann (1994) write: "A-Life refers to all the techniques that try to recreate living organisms and creatures by computer."

Promising Artificial Life Characteristics

This section will outline a number of key features of A-Life, such as dynamic fitness, the emergence of ecological dynamics, selection for self-reproduction, and open-ended evolutionary outcomes, that make it a promising theoretical approach for understanding biology more generally.

Dynamic Fitness

Formalizing the selection criteria via a program, which itself is allowed to evolve by the co-evolutionary process noted below, can get us close to natural selection by eliminating the a priori, externally imposed selective criteria usually involved in simulation programs. Danny Hillis used selective processes, and more importantly, from work based on the co-evolution of hosts and parasites, allowed for evolving evaluation functions (fitness functions) to efficiently find optimal sorting circuit designs. Instead of having the sorting networks tested against a fixed set of fitness evaluations (i.e. in this case, sorting problems) the sorting problems were allowed to change over time in response to the sorting networks. This prevented the sorting networks from getting stuck on local fitness maxima.

Essentially, these coupled populations, co-evolving via Darwinian selection, can bootstrap each other up the evolutionary ladder far more efficiently than they can alone. Indeed, Hillis’ (1991) evolving “computational selective agent”, or fitness tests in the form of evolving sorting problems, managed to generate a better design than other well-designed sorting networks. This example demonstrates that fitness can be simulated as a relative, changing quantity, which depends on the details of the system's evolving selective criteria at any given time—just as we see in nature. It also shows the power of using a co-evolutionary approach, and more generally, suggests that intentional design efforts by humans to create optimal systems, including alternative biochemistry, could fall short compared to the efficiency and creativity of automated computational open-ended evolutionary selective processes. Ecological Dynamics

Computational ecologies get A-Life closer to nature's complex evolutionary dynamics by incorporating many species of organisms co-evolving to form ecological webs. A specific example involves A-Life work based on the game theoretic model, the Iterated Prisoner's Dilemma (Lindgren and Nordhal, 1991). Over the long run, an inpidual’s score is maximized by cooperating, and this cooperative pattern has been shown to emerge via ordinary Darwinian mechanisms such as assuming that inpiduals want to maximize their immediate pay-off (Hamilton 1981; Axelrod, 1984). The evolutionary and ecological relevance is apparent when we note that strategies were allowed to evolve via an open-ended process by basing the decision on whether or not to cooperate on varying history lengths of previous interactions. The emergence of cooperation supports the suggestion that A-Life experiments are approximating evolution since cooperation has evolved via natural selection on earth.

Selection for Self-Reproduction

Tom Ray took the key step in removing all externally imposed selection criteria when he created his Tierra simulation system (Ray, 1996). In this approach, self-reproducing programs compete for computer processing time and memory space where the selective criteria of the programs is the success of self-reproduction. The programs copy themselves, and those which do it best survive and flourish. The fitness function is contained within the basic function of the replication of the organism itself—as in natural selection.

2

F3.2-B0.6-0015-02

With noisy replication, i.e. mutations, offspring that behave differently can be generated randomly, and these variants, combined with selection for reproduction, have resulted in quite complex and directly biologically relevant “organisms” and phenomena. For example, a parasite program evolved which exploits the ancestor program (the only program written by Ray) by using the ancestor replication loop to replicate itself. This allowed parasites to copy faster since they do not have to replicate the replication loop. But as we see in nature with viruses and cells, the parasites cannot take over the population to the point of driving the ancestor host to extinction, so a coexistence balance results. Another mutant organism resulted which is immune to the parasite because the immune organism makes it impossible for the parasite to use its self-replicating code and is also able to replicate twice as fast, driving the parasite to extinction.

As a tropical biologist, Ray has recognized other biological phenomena resulting in Tierra such as punctuated equilibria, competitive exclusion, symbiotic relationships and cheaters. Ray also reports the evolution of novel self-examination where organisms without an ending template evolved and were still able to calculate their size by using a mid-point of their genome, subtracting it from their beginning template and multiplying it by two. This indicates the evolutionary power of his approach.

Ray notes how much of the evolution of his system is driven not just by physical selective constraints of computer processing time and memory (the analog of the non-biological physical environment) but also by interactions with and adaptations to the biotic environment (e.g. other organisms) which is considered to be the primary force for persification of organisms. Ray takes this as an encouraging sign that his evolutionary system is behaving consistently with nature.

搜索“diyifanwen.net”或“第一范文网”即可找到本站免费阅读全部范文。收藏本站方便下次阅读，第一范文网，提供最新外语学习F3.2-B0.6-0015-02 USING ARTIFICIAL LIFE TO ASSESS THE TYPICA全文阅读和word下载服务。