Empiricism in Artificial Life

Strong artificial life research is often thought to rely on Alife systems as sources of novel empirical data. It is hoped that by augmenting our observations of natural life, this novel data can help settle empirical questions, and thereby separate fundamental properties of living systems from those aspects that are merely contingent on the idiosyncrasies of terrestrial evolution. Some authors have questioned whether this approach can be pursued soundly in the absence of a prior, agreed-upon definition of life. Here we compare Alife’s position to that of more orthodox empirical tools that nevertheless suffer from strong theory-dependence. Drawing on these examples, we consider what kind of justification might be needed to underwrite artificial life as empirical enquiry. In the title of the first international artificial life conference, held over a decade and a half ago, two streams of Alife research were identified—the synthesis of life-like or living systems versus their simulation. This distinction was perhaps intended to echo that made between strong and weak artificial intelligence, a division that has been readily adopted by the Alife community. While strong AI or Alife is concerned with building bona fide examples of real intelligence or real life, the weak strand of research is concerned with improving our understanding of intelligence and life via the construction of models or replicas of natural systems. While the latter branch of research is understood as an orthodox type of scientific or engineering methodology, the former has often been regarded as more problematic. How can we create genuine instances of intelligence and life without a prior understanding of what constitutes valid membership of either category? In this paper we will not consider what constitutes a definition of life, or what it is to be alive. We will assume that such questions are being pursued by others. Rather, we will be interested in a related question: what would have to be true of the relationship between artificial and natural life for it to be the case that observations of (computational) Alife systems could be legitimately employed to settle empirical questions regarding the nature of life? The root of the problem is hinted at by the suggestion of deception, falsity or unreality that can sometimes be detected in the meanings of both synthetic and simulated. We can clarify this hint by considering two distinct meanings of the term “artificial”. First, the word artificial can be used to describe a manmade example of something natural (hereafter denoted Artificial1). Artificial light, for example, is typically real, actual light that is manufactured rather than generated naturally by the sun, or forest fires, or bioluminescence of some kind—but these are all instances of the same physical phenomenon. By contrast, the word artificial can also be used to describe something that has been designed, perhaps through artifice of some kind, to closely resemble something else (hereafter denoted Artificial2). For example, an artificial lime flavouring, E555, might have been designed to taste like real lime, but is in fact not an instance of the real taste of lime even if it tastes indistinguishable from the real thing. Note that whether or not artificial light or lime can be used to settle empirical questions regarding real light or lime hinges critically upon which definition of artificial is understood to apply. Which meaning of the word artificial is justified in a particular case depends upon what criteria we feel must be met in order for something to count as a true instance of a particular category. In the case of light, we have a physical theory that allows us to lump together sunlight, firelight, torchlight, electric light-bulb light, etc., into one category: real light. Given this, we are in a position to pursue “strong artificial light” research, discovering new ways of generating new kinds of light. Weak artificial light research could investigate ways of producing the appearance of light via constructing models of light (which involve no light themselves) or perhaps fake lighting (e.g., ways of painting a room or a picture to suggest the presence of light). In doing so, we might learn about how light works and why it appears to us in the ways that it does. In the case of lime flavours, we have a less well-specified but widely agreed upon notion that the flavour of real lime fruit cannot be lumped together with some flavouring E555 in a category called “real lime flavour”, no matter how convincingly limey E555 happens to taste (unless perhaps E555 is derived directly and straightforwardly from real limes, or can be proven to be chemically identical to real lime). In this situation, “weak artificial lime” researchers could legitimately concern themselves with producing flavourings that resemble real lime in some way. Through this type of research, we might discover a lot of the chemistry, biology and psychology of what makes things taste like lime. Strong artificial lime research is a little more problematic, however, and we will return to why this is a little later. What is important to notice about this example is not simply that some categories, such as tastes, are subjective, but rather that some are as yet unsupported by some scientific account. The appearance of light is extremely subjective, yet we are able to objectively account for both the perceived variety of different forms of light and their unity through recourse to a physical theory. By contrast, the absence of such a theoretical account of lime flavour prevents us from achieving a similar understanding. In this paper we consider what kind of framework could underwrite strong artificial life as an empirical pursuit, since it is through the promise of generating useful empirical data that strong artificial life typically gains its strongest support.