EVOLUTION is a wonderful theory, but so far we only have one example of
it鈥攃arbon-based life forms on the third planet out from the Sun. But what
if we could build another example? What if we could build a huge, empty
ecosystem鈥攍et鈥檚 call it Tierra鈥攊nto which we then introduce
creatures that could replicate themselves, compete for resources, mutate to make
new forms, and travel to find niches that might be better? And what if they
evolved quickly, not in aeons and epochs, but in hours, days and months, so that
we could watch?
Such a universe could show us evolution in action. We could run experiments
by introducing new species, or new environments, or by varying the rate of
mutation. We could look at evolution as it happens鈥攏ot backwards through
time, as we are forced do in the real biological world. We might also find that
the pattern of evolution observed in the organic world is only one of a
multitude of possibilities.
Last month, an evolutionary biologist working with a group of computer
scientists created just such a universe. On 16 May, Thomas Ray and his
colleagues on the Tierra Working Group released tiny software creatures into a
universe of 100 computers connected over the Internet. The creatures will roam
this universe, competing with each other for the resources there.
Advertisement
The project leader, Ray, who holds a variety of posts including one at
Japan鈥檚 Advanced Technology Research Center near Osaka, spent years studying
real animals before moving on to silicon worlds. The seeds of the idea for
Tierra, his artificial universe, were sown in the 1970s, when Ray was a graduate
student at Harvard University. He was playing a game of Go with a computer
programmer from the Artificial Intelligence Laboratory at the nearby
Massachusetts Institute of Technology, and his opponent happened to comment:
鈥淒id you know, it is possible to write a self-replicating computer program?鈥�
Science in action
While collecting data in the Costa Rican rainforest, Ray had often lamented
that while he could record the results of evolution, he was never able to see it
in action or research its universal rules. But the Go-playing programmer鈥檚
remark sparked an idea: if a program could be made to replicate, perhaps it
could be made to mutate too. And if it could then be forced to compete as
well鈥攖hat is, to have needs that it must fulfil at the expense of other
programs in order to survive鈥攖hen the program could be made to evolve. He
could build evolution in a silicon bottle.
By early 1990, after teaching himself computer programming, Ray had built a
universe of small creatures that evolved with astonishing diversity, and
developed parasites, immunities and even rudimentary social interaction. (鈥淟ife
and death in a digital world鈥�, 快猫短视频, 22 February 1992, p 36).
All this lived on a single computer.
But Ray soon encountered another hurdle: he wanted to give the creatures an
environment large and diverse enough for evolution to develop complex,
鈥渆mergent鈥� behaviours. Even the largest and fastest massively parallel
processing machines would not provide enough computing power, memory and
diversity for his vision of silicon evolution to take off.
Ray then realised that hundreds or even millions of computers connected
together over the Internet could easily and cheaply provide a universe big
enough, rich enough and random enough for his creatures to evolve. 鈥淥n the Net I
get the size and complexity for free,鈥� says Ray, 鈥渁nd evolution is facing a sort
of `real鈥� world, the Net, rather than some arbitrary concocted simulation game.鈥�
Different computers would offer different ecological niches鈥攅ach as
distinct from the next as temperate estuaries are from tropical rainforests.
Some would be connected to the Internet all the time, exposing their inhabitants
to competition from programs that evolved in other ecosystems. Others would be
connected only intermittently. Each would have its own speed, its own operating
system, its own quirks.
The program itself is simple enough: its main task is to create digital
analogues of variation and competition, the twin drivers of biological
evolution. The Tierra source code, written in the C programming language,
creates a 鈥渧irtual computer鈥� operating within the host computer, so separating
the silicon primordial soup from the other workings of the computer host. The
Tierra program functions as the operating system of this 鈥渧irtual computer鈥�. It
creates a number of digital creatures鈥攖iny, self-replicating machine code
programs, which it introduces into the computer鈥檚 working memory.
Command performance
Each program is 32 instructions long, and each instruction is a five-bit
number. These terse instructions are translated by the Tierra operating system
into commands such as 鈥渄uplicate this code and store the duplicate somewhere in
the memory space that the host is not using鈥� or 鈥渄o nothing鈥�. For each
generation, the operating system runs these tiny programs once, performing the
commands as it comes to them. But at the same time that it runs each program, it
mutates some of them, randomly flipping some of the zeros to ones or vice versa,
or even swapping part of one 鈥渃reature鈥� for part of another.
This creates variation, the first of the two drivers of evolution. Many
mutations make no difference. But others do. The 鈥渄uplicate鈥� command may be
changed to 鈥渄o nothing鈥�, for example. Other mutants may end up with loops of
鈥渏ump forward鈥� and 鈥渏ump backward鈥� commands that run the program repeatedly over
a 鈥渄uplicate鈥� section, causing the digital creatures to make multiple copies of
themselves.
Once the host computer鈥檚 memory is full, the virtual operating system
releases a 鈥渞eaper鈥� that kills off the oldest creatures and the ones that
generate error messages; to make room for newborn programs. This creates
competition: if a program does not run well, it dies.
Through the Tierra operating system, the human operators have Godlike
control. They can define how fast the creatures will mutate and specify how much
time they get on the central processor to reproduce. This last power is
analogous to setting the richness of resources in a biological system鈥攊s
it a balmy tropical isle or a harsh mountain top? Or they can specify how much
of the memory of their host they get to inhabit, to mimic different-sized
territories: is it the Amazon basin, or an oasis in the Sahara? The operating
system also provides comprehensive information on what is happening in the
evolutionary 鈥渟oup鈥�, recording every birth and death, every code sequence in
every new creature, every successful 鈥済enome鈥� and every interaction between the
creatures.
Ray hopes that Tierra will answer a number of questions that have remained
unanswered up to now. In the real world, species rely on an ecological niche
created by differences as slight as a change in the colour of tree bark. But
why? Tierra offers the chance of answering such questions because it can be
repeatedly reset to a particular starting state and then allowed to evolve under
slightly different conditions. And there are plenty of other questions. Under
what conditions do species compete to exclude one another rather than coexist?
What regulates the ratios of hosts and parasites in a population? And is Earth鈥檚
鈥減unctuated equilibrium鈥� the normal rhythm of evolution?
No one yet knows how the digital creatures will respond to the variety of
ecosystems that will face them. It may turn out that some species do well in
high-speed environments just as certain plants thrive in the sun. Some may
thrive on constant connection to the Net, like sharks that can take anything
that comes their way. Others, like Galapagos tortoises, may need the relative
isolation of an intermittent Net connection.
To allow the creatures to seek out congenial homes, Ray has given them access
to a modified version of the Internet Ping command. Ping is normally used to
discover whether an address is valid, and whether a computer with that address
is connected to the Net. The modified command, Tping, allows Tierra鈥檚 digital
creatures to ask questions about the environment on a particular computer. Among
these questions are:
What is the speed of the computer鈥檚 processor unit? A faster processor
means there are likely to be more spare processing cycles that the creature can
use.
How much RAM does the computer have? More memory means more resources.
How many other Tierra creatures already live there? If there are only a
few, there will be less competition.
What operating system is running on the computer?
Finding a niche
Ray and his colleagues do not know how the creatures will use this
information, but it may be that they evolve a capacity to relate information
gleaned with Tping to their ability to survive. If so, a niche could develop for
creatures that migrate short distances rapidly, but can tolerate slow processors
and take advantage of the Unix operating system?
The high-powered band of volunteers that makes up the Tierra Working Group
draws its members from the Santa Fe Institute in New Mexico, supercomputing
centres in the US and private companies. They have adapted Tierra to run on a
variety of operating systems, from Unix to Windows 95 and Macintosh.
The Working Group seems confident that the creatures will not evolve into The
Code That Ate The Internet, since they are impotent without the 鈥渧irtual
computer鈥� formed by the Tierra program, which cannot migrate. The creatures have
no access to Internet addresses. They migrate only to Tierra addresses that are
slightly different from each node鈥檚 normal Internet address. Ray stresses that
any creatures that happen to migrate to a computer that is not running the
Tierra program will not reproduce, and will simply disappear as soon as the
computer鈥檚 operating system decides to use that spot in RAM for something
else.
Over the next few months the working group hopes to open Tierra on more
sites, so widening the evolutionary pool. About a year from now, Ray hopes to
expand the experiment further by allowing anyone with an Internet connection to
download and install Tierra.
Despite the working group鈥檚 assurances, not everyone is happy about the idea
of an evolving silicon universe on the Internet. Tsutomu Shimomura, the security
expert who tracked down computer hacker Kevin Mitnik, asked Ray at a recent
conference whether someone could deliberately re-engineer the Tierra creatures
to carry viruses. Ray was not able to reassure Shimomura that this was
impossible. But he has set Manor Askenazi of the Santa Fe Institute the task of
testing for security weak points.
But for evolutionary biologists, security is secondary to the question of
whether Tierra will reveal useful information about evolution. Marcus Feldman, a
mathematical population biologist at Stanford University in California, remains
sceptical. Tierra is 鈥渙kay, as long as we don鈥檛 call it biology鈥�, he says.
鈥淓volution is more than a programming problem.鈥� He argues that Tierra does not
draw firmly enough on our understanding of evolutionary theory and population
genetics.
Yet it may be too early for such dismissals. The next five years will begin
to show whether the Tierra approach remains an interesting freak show and
teaching aid, or whether it will generate new insights into some of the
intractable problems of biological evolution, such as the emergence of sex,
culture, and complex ecologies.