¿ìè¶ÌÊÓƵ

Tom Ray describes himself as a naturalist in the best Darwinian tradition.
He has spent years pursuing this passion in the rainforest of Costa Rica,
discovering and showing to the world a myriad of nature’s bizarre creations:
vines that transmogrify depending on where they are growing; butterflies
that follow ant birds, that follow army ants, that prey on other insects.
And much more. These days, however, Ray is a naturalist of a different kind,
an observer of the bizarre creations of an alien world, a world where no
human has ever set foot, or ever will.

On 3 January, two years ago, Ray pressed the start button on his computer,
kicking into action a program called Tierra (Spanish for Earth), and waited.
‘I spent a pretty sleepless night, wondering what I’d find the next morning,’
he recalls. What he found was an electronic ecosystem of astonishing diversity
– populations of many different kinds of organisms, all descendants of a
single ancestral organism from which Tierra had come to life the previous
evening. ‘No one expected this would happen, least of all me. From the most
basic of instructions there emerged an astonishing complexity.’

Into Tierra, Ray had introduced what he calls his ancestral organism,
a short set of instructions that simply specified how it would reproduce
itself. As in real life, reproduction was not always perfect and mutations
occurred, producing organisms with new sets of instructions. Equally true
to life, these organisms had to compete for energy and space within Tierra,
and some thrived better than others. But in contrast to evolution’s slow
grind, Ray’s program could compress thousands of generations of life into
a few hours of computer time. The result was that life in Tierra soon looked
uncannily like life on Earth.

There were organisms of different sizes and behaviour: some were parasitic,
some were social, some cheated flagrantly on their colleagues. But it wasn’t
just the components of the Tierran world that seemed to mimic real life;
it was the patterns too, the bursts of change followed by periods of stasis,
and occasional mass extinctions. Plot the records of Tierran life graphically,
and they look just like the history of life on Earth, as recorded in the
fossil record.

Biologists have long fantasised about being able to run the evolutionary
clock over again, to have more than just one world of nature in which to
study the phenomena of evolution. Tierra brings that fantasy closer, although
many refinements are still needed. ‘I plan to introduce sex into the system,
and multicellularity,’ says Ray. ‘Then we’ll be able to run the Cambrian
explosion all over again. That’s my aim. That’s when the interesting biology
really begins.’

The Cambrian explosion, 550 million years ago, was a time of tremendous
evolutionary innovation, when multicellular life became established in a
burst of diversification of basic forms, or phyla. Biologists would like
to know what constraints governed that diversification, how broad the possibilities
really were. Short of turning the clock back and repeating the event, Tierra
and programs like it offer the best chance at present of gaining some real
insights.

Ray’s foray into the world of computing has taken him out of field ecology
and into a sub-discipline of the science of complexity, described grandly
as Artificial Life. So far, most AL practitioners are physicists or computer
experts, intrigued by the incredibly complex systems that sometimes emerge
from the interaction of the simplest of ‘organisms’, more properly called
‘entities’. The shared hope is that fundamental rules will emerge governing
order and evolution in the natural world, rules that will reveal an underlying
simplicity to the great complexity of nature.

Many of the programs AL enthusiasts use look like little more than sophisticated
computer games. The most popular is the Game of Life, where amazingly complex
patterns of so-called cellular automata – arrays of squares whose distribution
changes according to the repeated application of simple rules – evolve from
a simple starting pattern. Of the systems presently being studied, Ray’s
is undoubtedly among the most complex, both in the nature of the ancestral
organism and in the artificial world that evolves from it.

He conceived his plan to create an artificial world – evolution in a
bottle, as it were – more than a decade ago. ‘I can trace the idea back
precisely, to when I was a graduate student at Harvard,’ he says. One evening
Ray was visiting the Harvard Science Center, where the Cambridge Go Club
met regularly. Go, an ancient Chinese game, is exceedingly complex and requires
players to move populations of ‘pebbles’ around a board, the aim being to
trap and destroy the opponent. Because of a certain intellectual affinity,
many of the club’s members were from the Artificial Intelligence lab at
the Massachusetts Institute of Technology.

‘That evening, there was one guy playing by himself, so I sat down and
he explained the game to me,’ says Ray. The lone player explained the game
in very life-like metaphors, such as the strategy of certain groups of pebbles,
the pebbles being surrounded and killed, and so on. This intrigued Ray,
because it had the aura of an artificial world. Then the player casually
asked Ray a question, one that seemed to crystallise in his mind a clear
and powerful goal from a series of barely contemplated ideas already lingering
there. The player said: ‘Did you know it was possible to write a computer
program that can self-replicate?’. ‘The moment he said it I had a flood
of ideas,’ recalls Ray, ‘all the kinds of ideas I’m pursuing now.’ That
was in 1980.

Although Ray had a clear vision of what he wanted to do with such a
world, many things conspired to prevent him going forward with it. Not least
was a degree of naivety about computers and programming. ‘I could write
programs at a primitive level,’ he admits, ‘and I think I have a certain
affinity for computing, but my experience was very limited.’ When the lone
Go player tried to explain how to write a program that could self-replicate,
‘either he didn’t explain very well or I simply couldn’t understand what
he said’. With that, the idea of an artificial world was to remain a fantasy
for a while, ‘just one of many I had at the time,’ he says.

Ray graduated from Florida State University in 1976, where he had worked
on projects with Dan Simberloff and Donald Strong, two of the most important
figures in ecology in the US. Strong had taken Ray to Costa Rica, where
he had some research under way. ‘Cheap labour’ is how Ray now jokingly describes
it. Nevertheless, he was hooked. ‘When you are surrounded by rainforest,
you can’t not be affected by it. It’s a powerful feeling, all these fascinating
organisms interacting in the most bizarre and complex ways. That’s why I
became an ecologist.’ Ray even bought a piece of rainforest and established
it as a reserve where he carried out his research.

Graduate school at Harvard was followed by a faculty post at the University
of Delaware, all the time doggedly pursuing whatever phenomenon of nature
interested him, no matter what direction it took. He became deeply immersed
in studying leaf development, the growth patterns of plants. The rationale
was that if you want to understand how a plant behaves as part of an ecosystem
you have to know about its development. The work involved handling a lot
of data, so he used the university’s mainframe computer.

The distance – literal and figurative – that separates users at their
terminals from what is going on in the computer was another impediment to
Ray realising his fantasy. ‘You type things in, and it responds. That’s
it. I didn’t have a concept of the physical representation of the program
in the computer.’ Acquiring that was to be crucial in establishing an artificial
world with the special characteristics that Tierra one day would have.

The breakthrough came in 1988, when Ray bought a laptop computer, the
first personal computer he had owned. Until that time, personal computers
had not been powerful enough for Ray’s data handling and processing. In
any case, he was comfortable using the mainframe. However, now equipped
with a PC, software, and a debugger, he began to become familiar with the
innards of computers in a way he never was previously. ‘The debugger opens
up a window into the computer, lets you watch how the program operates,
lets you see the physical arrangement,’ says Ray.

This insight was crucial, because his aim was to use the computer as
an environment in which his organisms would live. ‘I didn’t want simply
to make a model of biology,’ he explains. ‘I said, here’s an environment.
I want to tailor a life form that is suited to this environment.’ Part of
this environment was to be an energy analogue, which is the amount of time
a replicating algorithm would occupy the central processing unit – the nerve
centre of a computer. Another part is the space the organisms would occupy,
and this was the physical location in memory. These biological analogies
contributed to making Ray’s system unique among artificial life systems.

For a year, Ray immersed himself in programming and computer manuals,
trying to become familiar enough with machine code – the most fundamental
of all programming languages – to create his artificial world. He also dispatched
a computer message, trying to find out what, if anything, like-minded people
were doing. He made contact with Chris Langton, the guru of artificial life
at Los Alamos National Laboratory, New Mexico, and discovered with some
relief that no one had yet achieved what he planned to do. A visit to Los
Alamos National Laboratory was arranged, where Ray met Langton, Doyne Farmer,
Steen Rasmussen, Stephanie Forest, and others involved in complex systems
and artificial life.

Ray came away from the visit with three clear messages from Langton
and his colleagues. First, he must run his program on a virtual computer
– a computer emulated by software within the computer – otherwise there
was a real danger that his organisms would escape and populate other computers,
thus causing havoc, like computer viruses. Second, that the project almost
certainly would not work, because the program would be too sensitive to
the random mutations, and would surely crash. Third, whatever he tried to
do would almost certainly take a long time to work out. This, now, was the
summer of 1989. Langton’s advice about the virtual computer was sound, and
Ray followed it. But, as things transpired, the other two predictions were
wrong.

Guided by analogies from molecular biology, relating to tricks of recognition
and information storage and retrieval, Ray began to design his digital organism.
For a start, in the biological world genetic information is stored in four
nucleotides, grouped in threes. This yields 64 possible instructions, or
codons, which encode a total of 20 amino acids (most amino acids are specified
by more than one codon). In the programming language Ray used, machine code,
there were as many as 100 billion instructions, and vast possibilities for
mutations. ‘My intuition told me that would be a problem,’ he says. ‘I had
no evidence for saying that, just a gut feeling.’ He therefore halved the
original possible instructions, finishing up with just 32, much closer to
the real world.

A second imprint of molecular biology that Ray put into his ancestral
organism was ‘addressing by template’. Usually a program written in the
language of machine code will summon any information it needs by specififying
the exact position, or numeric address, of the information in the computer’s
memory. This is not how biology goes about things. For example, in a cell,
a protein, A, can interact with a second protein, B, only when the two collide
by random diffusion. If the contours of their surfaces are complementary,
the two proteins will then lock together, allowing chemical information
to pass from one to the other. Ray mimicked this trick of nature when writing
the code for Tierra’s ancestral organism.

The language of machine code contains just two digital ‘letters’ (1
and 0), out of which Ray had to construct all the instructions and information
the organism would need to recognise various bits of itself. He put a short
code of four instructions, in the pattern 1111, at the head of his creature,
and another group of four in the pattern 1110 at its tail. ‘Between these
two instructions, I filled in a program that would start by looking for
the pattern complementary to 0000 to find its head and record its location,
then look for the pattern complementary to 0001 to find its tail and record
its location; and then calculate the size,’ he explains. The program in
between the head and tail codes contains instructions for replicating the
organism and finding a nearby location for the ‘daughter’ organism. Moreover,
addressing by template, as opposed to by position, also allowed organisms
to find neighbours with which they might interact.

Ray was convinced that Langton would be proved wrong in his prediction
that the program would be too fragile to accommodate mutations and would
crash. Nature copes with mutations, so a computer program should too, he
thought. But he believed Langton was probably right in thinking that writing
the program would be a mammoth task, involving a long and painstaking search
for the necessary components. In the end, though, it all came together unexpectedly
quickly. The ancestral organism, all 80 instructions of it, took just six
months to create.

There were no specifications about how the organism might mutate, or
what kind of descendants it might produce. The ancestor would simply be
susceptible to errors in replication and to random flips in the code, analogous
to mutations caused by cosmic rays, for instance. By creating an organism
of this simplicity, with no specification of how it might evolve, only that
it would, Ray hoped his system would synthesise life, not merely simulate
life. By now it was 3 January 1990, and the program was ready to run.

In that first run, Tierra essentially produced a glimpse of the entire
landscape of possibilities in its digital world. From the single 80-instruction
ancestral individual injected into the digital ‘soup’ of the computer, tens
of thousands of descendants had been produced over thousands of generations.
New species emerged, some bigger than the ancestor, some smaller, and with
the passage of Tierran time, the digital ecosystem fluctuated in composition,
sometimes being dominated by a few types, sometimes displaying a wide diversity.

The rapid emergence of parasites was a surprise. These creatures, a
mere 45 instructions long, had lost the coding for replication and survived
by hijacking the replication code of a nearby organism. They didn’t directly
harm their hosts, but did deprive them of valuable energy and space. In
environments where space was in short supply, the parasites flourished;
but when their hosts become scarce and the parasites were no longer able
to find replication codes to co-opt, they soon died – just as in real life.

Moreover, just as real organisms acquire immunity to parasites, so,
too, did a descendant of the ancestral organism. The copying instructions
of this organism, 79 instructions long in total, had somehow become resistant
to the advances of the 45-instruction parasite. In the presence of this
immune creature, the parasite was driven to extinction.

Some Tierran creatures went further. After being attacked by parasites,
they cannily hijacked their attackers’ code and used it to promote their
own replication. Some of these so-called hyperparasites, in turn, evolved
into social creatures, organisms made from 61 instructions which could reproduce
only by helping each other. But, as in any system built on trust, this one
was vulnerable to cheats. The cheats came in the shape of creatures (27
instructions long) that positioned themselves among groups of cooperating
hyperparasites, appropriating the instruction pointer as it is passed between
them. All of this was the product of a few basic constraints and the selection
of organisms produced by purely random mutations.

The menagerie of creatures in Tierra is fascinating enough in itself.
Yet the patterns of their interaction, and of the ecosystem as a whole over
time, are all the more arresting because of the way they echo patterns of
life on Earth. Many of the phenomena that ecologists study, such as the
abilities of certain species to out-compete others, the dominant roles played
by certain predators, and the evolution of altruism and cooperation in ecological
communities can be seen in Tierra. And the long periods of stasis interrupted
by a burst of change is uncannily reminiscent of the pattern of punctuated
equilibrium over which evolutionary biologists have debated for so long.

Plot the history of life in Tierra over long periods, and occasional
mass extinctions can be seen – ‘and that’s in the absence of asteroids,’
comments Ray. Crashes of diversity – mass extinctions – appear to be an
inescapable property of the evolving system, not necessarily something imposed
from the outside.

This similarity between terrestrial biology and the Tierran world is
giving artificial life researchers confidence that they are onto something
very fundamental. Even in systems much simpler than Ray’s, for example one
developed by Kristen Lindgren of Chalmers University, Goteborg, similar
patterns are seen, albeit without the richness of the biology of Tierra.
‘You get the sense of something deep as an organising force,’ says Ray.
‘I’ve always had the idea that would be true, ever since high school, the
feeling that there is some process on the level of physics that leads to
increasing complexity.’

With the digital world of Tierra at this fingertips, Ray has gone beyond
being simply a naturalist of this new and strange domain. He has become
an experimentalist in that world, changing the rate of mutation if he chooses;
changing the competitive edge conferred by, say, being small or being large;
throwing an assembly of organisms together, just to see what will transpire;
ultimately, rerunning the Cambrian explosion. ‘It is,’ as he says, ‘like
playing God.’