Darwin's Dangerous Idea - Evolution and the Meaning of Life

174 PRIMING DARWIN'S PUMP The Laws of the Game of Life 175
have a greater claim to having been designed? Perhaps, but there is no line to
be drawn between merely ordered things and designed things. The engineer
starts with some objets trouves, found objects with properties that can be
harnessed in larger constructions, but the differences between a designed and
manufactured nail, a sawn plank, and a naturally occurring slab of slate are
not "principled." Seagull wings are great lifters, hemoglobin macromolecules
are superb transporting machines, glucose molecules are nifty energypackets,
and carbon atoms are outstanding all-purpose stickum-binders.
The second point is that Life is an excellent illustration of the power— and
an attendant weakness—of computer simulations addressed to scientific
questions. It used to be that the only way to persuade oneself of very abstract
generalizations was to prove them rigorously from the fundamental principles
or axioms of whatever theory one had: mathematics, physics, chemistry,
economics. Earlier in this century, it was beginning to become clear that
many of the theoretical calculations one would like to make in these sciences
were simply beyond human capacity—"intractable." Then the computer came
along to provide a new way of addressing such questions: massive
simulations. Simulation of the weather is the example familiar to all of us
from watching television meteorologists, but computer simulation is also
revolutionizing how science is conducted in many other fields, probably the
most important epistemological advance in scientific method since the
invention of accurate timekeeping devices. In evolutionary theory, the new
discipline of Artificial Life has recently sprung up to provide a name and an
umbrella to cover a veritable Gold Rush of researchers at different levels,
from the submolecular to the ecological. Even among those researchers who
have not taken up the banner of Artificial Life, however, there is general
acknowledgment that most of their theoretical research on evolution—most
of the recent work discussed in this book, for instance—would have been
simply unthinkable without computer simulations to test (to confirm or
disconfirm) the intuitions of the theoreticians. Indeed, as we have seen, the
very idea of evolution as an algorithmic process could not be properly
formulated and evaluated until it was possible to test huge, complicated
algorithmic models in place of the wildly oversimple models of earlier
theorists.
Now, some scientific problems are not amenable to solution-bysimulation,
and others are probably only amenable to solution-by-simulation,
but in between there are problems that can in principle be addressed in two
different ways, reminiscent of the two different ways of solving the train
problem given to von Neumann—a "deep" way via theory, and a "shallow"
way via brute-force simulation and inspection. It would be a shame if the
many undeniable attractions of simulated worlds drowned out our aspirations
to understand these phenomena in the deep ways of
theory. I spoke with Conway once about the creation of the Game of Life,
and he lamented the fact that explorations of the Life world were now almost
exclusively by "empirical" methods—setting up all the variations of interest
on a computer and letting her rip to see what happens. Not only did this
usually shield one from even the opportunity of devising a strict proof of
what one found, but, he noted, people using computer simulations are
typically insufficiently patient; they try out combinations and watch them for
fifteen or twenty minutes, and if nothing of interest has happened, they
abandon them, marking them as avenues already explored and found barren.
This myopic style of exploration risks closing off important avenues of
research prematurely. It is an occupational hazard of all computer simulators,
and it is simply their high-tech version of the philosopher's fundamental
foible: mistaking a failure of imagination for an insight into necessity. A
prosthetically enhanced imagination is still liable to failure, especially if it is
not used with sufficient rigor.
But now it is time for the my main point. When Conway and his students
first set out to create a two-dimensional world in which interesting things
would happen, they found that nothing seemed to work. It took more than a
year for this industrious and ingenious group of intelligent searchers to find
the simple Life Physics rule in the Vast space of possible simple rules. All
the obvious variations turned out to be hopeless. To get some sense of this,
try altering the "constants" for birth and death—change the birth rule from
three to four, for instance—and see what happens. The worlds these
variations govern either freeze up solid in no time or evaporate into nothingness
in no time. Conway and his students wanted a world in which growth
was possible, but not too explosive; in which "things"—higher-order patterns
of cells—could move, and change, but also retain their identity over time.
And of course it had to be a world in which structures could "do things" of
interest (like eat or make tracks or repel things). Of all the imaginable twodimensional
worlds, so far as Conway knows, there is only one that meets
these desiderata: the Life world. In any event, the variations that have been
checked in subsequent years have never come close to measuring up to
Conway's in terms of interest, simplicity, fecundity, elegance. The Life world
might indeed be the best of all possible (two-dimensional ) worlds.
Now suppose that some self-reproducing Universal Turing machines in the
Life world were to have a conversation with each other about the world as
they found it, with its wonderfully simple physics—expressible in a single
sentence and covering all eventualities. 8 They would be committing a log-
8. John McCarthy has for years been exploring the theoretical question of the minimal
Life-world configuration that can learn the physics of its own world, and has tried to enlist