Artificial Life, or “A-Life” as it calls itself, arrives as the newest wave of computer-driven science threatening to produce as deep as upheaval in biology as that effected by fractal geometry and chaos theory in the physical sciences. For the most part, A-life’s objects are virtual creatures: patterns of activity that occur inside a computer, represented for us by bug-like, bird-like, plant-like assemblages of pixels which undergo electronic versions of growth, procreation, death, birth and evolution. But real robots that mimic living creatures are also included.
To the uninitiated, Artificial Life sounds like Artificial Intelligence, and given the gulf between the latter’s original claims and its pathetic delivery, might not unnaturally produce a sinking feeling. This would be a mistaken response. It seems clear, after reading Steven Levy’s survey, that the pioneers of this fledgling science have learned a lesson from AI’s trumpetings; they are cannier in the claims they make and, having suffered a good deal of establishment indifference and ridicule, more realistic in how they promote their subject. Along with this they are also more worldly – several are entrepreneurs, rich from exploiting their work – and less likely to be gulled by their own rhetoric. More important, much of their work evinces a wholesale methodological opposition to the prevailing AI paradigm, which is dominated by an exact, serial-processing logic operating according to a top-down, global analysis of its object. Overthrowing this paradigm results in bottom-up models of biological activity, which achieve global behavior via local rules and the parallel processing of information.
A rich source of such models is the class of mechanisms know as cellular automata (CA). To take a basic example, imagine an extended chessboard and suppose each square – thought of as a simple machine with a finite number of states – can register 0 or 1 or be blank. Suppose, further, that at each time-tick the state of a square is a determined locally – entirely by the state of its four contiguous neighbors according to some simple rules. The resulting system would be two-dimensional CA whose development can exhibit an extraordinary and surprising range of global behaviours. Though invoked originally by von Neumann as a way of demonstrating the theoretical possibility of a self-reproducing machine – and interesting to A-lifers for this very reason – CA have become a central tool in replacing the seriality seen to be von Neumann’s legacy to computer science.
The work of the biologist Thomas Ray is a striking illustration. Ray’s aim is to simulate natural selection. To this end he has defined an artificial environment, a two-dimensional CA called Tierra, into which he releases creatures – bits of code less than a hundred instructions long – with a set of simple survival and reproduction rules governing their struggle for computer-time and memory. By building in sources of mutation, and having their fitness rewarded over runs of many thousands of generations, his creatures evolve a whole array of biologically recognizable characteristics, from parasitism to various forms of predatory activity, unthought of and unplanned by Ray.
From a different direction, the computer scientist Rodney Brooks asks: how can a bee, operating at barely a kilohertz, with little onboard computing power, navigate fast around an unseen noisy environment, feed itself, find its way home, communicate and reproduce, whereas the best robots produced by current AI, connected to powerful external computers running at several megahertz, only just manage to traverse, very slowly, a pre-planned, sterile environment? Brooks’s answer is to build robots as distributed (parallel) assemblages of locally controlled mechanisms, whose architecture structures their interconnection from the bottom up. The creatures, which have no guiding plan, no map of their surroundings and no memory longer than a few milliseconds, exhibit global behavior, for minutes or hours on end, that we would describe as purposeful, or for that matter “alive”, were they insects.
For the cyberneticist John Holland, simulations of nature, in particular models of evolutionary fitness, suggest the idea of genetic algorithms. Starting from a rule no better than a random guess at the answer to some task, putting difficulties in its way and chopping up and rejoining the relatively successful portions, he was able to breed an algorithm that was just short of the optimal solution to the task, found – not easily – by conventional mathematical research.
A-life has a vision. In its weak form this is to contribute significantly to the biological sciences, producing new ways of understanding the nature of nature and its mechanisms (such as illuminating the concept of emergence, whereby unpredictable, complex and lifelike behavior arises from simple rules). Only a prejudiced observer would deny the feasibility of this. The strong version sees the definition of life de-natured – cut free from carbon-based and water-based forms to include all manner of manufactured organisms. It is of course the latter, conjuring the familiar terrors of golems, Frankenstein monsters, the supersession of human beings, runaway technology, domination by robots, and so on, that makes the headlines. Could A-life creatures, chunks of code enjoying the benefits of electro-evolution, get so complex that consciousness might be one of their emergent properties? Could such creatures come to the recognition that they are inside a computer? If they did, could they develop the technology to escape? In the past decade reputable physicists and computer scientists have taken seriously the possibility that the entire physical universe in which we find ourselves is a computer. Such a computer is presumably one that we can’t – this side of God – escape from. Would our hypothetically conscious chunks of code, electronically savvy in unpredictable ways, be similarly boxed in?
Levy’s book is hugely informative, covering a lot of ground: from Wolfram’s classification of cellular automata and Hillis’s exploration of punctuated evolution on his connection machine to the creation of electronic plant life, the emergence of flocking mechanisms in birdoids, trail-following antoids, camouflage, electronic sex and cannibalism, and much else. Based on interviews with the leading figures and told pretty much from their viewpoints, Artificial Life tends to be choppy, thematically repetitive, and lacking in critical distance from the ideas, people and movements it reports on. It does none the less convey the excitement and importance of the game. A-life has formidable implications for the way we think of ourselves as purposeful, sentient beings. The effect is bracing and unnerving.
TLS April 15, 1994