story :: FEBRUARY 09, 2005
‘This is not a simulation’
MSU researchers build a world and watch
By LAWRENCE COSENTINO
|MSU graduate student Jeff Clune puts Avidians
through their paces.
|Clune and graduate student Gabe Yedid discuss some new findings.
You’d think that Michigan, as the
only state in the union with an opposable thumb, would be more advanced
than the lower states when it comes to studying the evolution of life.
And you’d be right. For the past several years, a dozen or so
faculty and graduate students at Michigan State University have spearheaded
an exciting and strange field of study with mind-boggling implications,
not just for evolutionary science, but for 21st-century technology as
Their work doesn’t deal with dogs, goldfish, moths or even bacteria,
to name but a few beasts that evolve new forms in a short enough period
of time for humans to watch. Rather, they work with completely digital
organisms — madly self-replicating packets of information resembling
computer viruses — that go through tens of thousands of generations
in the time it takes an apple to fall.
As they beget and beget and beget, these tiny creatures adapt to their
surroundings, sprouting the digital equivalent of fins, claws, beaks,
Brad Pitt-firm pectorals — whatever it takes to transmit their
digital code to another generation.
|Contrastingly shaded computer creatures struggle for dominance.
(Graphic courtesy of MSU)
By the end of a “run” —
a controlled orgy of digital replication — “they are able
to perform complex functions they couldn’t do before,” explains
Robert Pennock, philosophy professor at MSU’s Lyman Briggs School
“And,” he adds with grave emphasis, “the code which
allows them to do it isn’t something we programmed in. It’s
something that evolved.”
(“Evolved, evolved, evolved,” goes the sci-fi echo.)
This new field of study, called “evolutionary computation,”
has spread to universities throughout the nation, but MSU is firmly
placed as the field’s guiding light and nerve center. Digital
evolution research by Pennock and his colleagues has graced the pages
of Nature, an ultra-prestigious science journal. The MSU team’s
work also made the cover of the February 2005 issue of Discover magazine.
Much of the work at MSU is done by means of a software program called
Avida, which enables scientists to compress millions of years of evolutionary
history into the running time of “Troy” (or, hopefully,
“We can’t rewind the earth and start it again,” says
Jeff Clune, a mischievous-eyed philosophy major and graduate student
who came to MSU to be in the thick of the ferment. “But we can
do the equivalent with Avida.”
So who should care if computer giraffes evolve computer necks to munch
computer leaves on computer trees? Anyone who cares about the unimaginably
vast possibilities of life, Clune says
“Natural selection has the ability to produce extremely interesting
solutions to problems,” he explains. “Anything in nature
you think is ingenious — the fact that hawks can fly, that whales
can swim into the depths of the ocean, that cheetahs can run as fast
as they do — is a product of natural selection.”
But it took millions of years to produce a paw, an antenna, or a stinger.
MSU’s digital organisms are able to make comparable developmental
leaps in days, minutes, even seconds.
Clune sat me down in front of his G-4 laptop at the Digital Evolution
Laboratory, in MSU’s Engineering Building on Wilson Road, to demonstrate
a “run.” Around us, half a dozen graduate students worked
intently on various Avida-related experiments.
A red dot — the lone forebear of a world in the making —
appeared in the center of an empty black square. In a second or two,
another red dot appeared, due west of the first. Suddenly, whole clouds
of multicolored, digitally saw-toothed blobs clouded the screen. Clune
explained that each color represents a group of genetically identical
organisms, jostling for dominance in what amounts to a digital petri
This is the part where digital evolution becomes a spectator sport.
“The turquoise people are doing pretty well,” said Gabe
Yedid, a serious, poker-faced graduate student in zoology, who came
over to look.
“This patch is making a run for it,” said Clune, pointing
at a pink mini-France with Napoleonic ambitions to match. A tiny peninsula
of yellow marked a lone holdout for its own dying color, which apparently
wasn’t making the evolutionary grade. I imagined the last pterodactyl,
mammoth, or dodo, closing its eyes and quietly calling it an epoch.
Just what goes on behind these compelling mini-dramas? “The code
for each digital organism randomly varies,” Pennock explains.
“Mutations happen at random in the instruction set. If the random
variation makes the organism a little bit better than another in some
way, then it will survive to replicate.”
Avida’s computer organisms don’t fulfill every definition
of life. They “metabolize” numbers into extra computer time,
for example, instead of eating Coney Islands and playing touch football.
But the processes they undergo — replicating themselves, producing
offspring that vary randomly, and changing over time when some of the
little guys survive and others don’t — are real, not simulated.
“This is not a simulation of natural selection,” Pennock
says. “This is an instance of it.”
The model hits closer to home than some people might think. Even a “real”
organism — take Brad Pitt again — can be thought of as a
program of genetic instructions using the nifty equipment it’s
collected over the eons (in this case, a hunky actor coating) to lure
mates and spool out more Brad Pitt programs.
The tasks the Avidians perform aren’t as spectacular as beaver
dams or elephant stampedes, but the results are still eye-opening if
you know what you’re looking at. In one run, Clune says, the critters
were asked to do a logic operation: Look at some pairs of numbers and
spit out a “one” if either (but not both) of the numbers
in the pair was a one. Clune says the professor who hand-coded the task
swore up and down that no Avidian could accomplish the feat in fewer
than 19 steps.
“All of a sudden one day, we looked in and saw they’d cut
it down to 11,” marvels Clune. “They had taken advantage
of a shortcut that none of us had even contemplated.” Even Charles
Ofria, director of the lab and the author of the Avida software, hadn’t
foreseen the solution.
As the one-or-not-one quiz shows, the Avida researchers don’t
just flip a switch, sit back with a pale ale and watch the micro-zoo
go nuts; they set up specific experiments to test what the zoo can do.
Why, for example, are some organisms nice to each other, when they might
just as well get an evolutionary leg up by killing and eating them?
That’s the phenomenon of altruism, or cooperation, which is Clune’s
specialty. He points out that kin altruism (be nice to your family because
they have your genes) and reciprocal altruism (clean the parasites out
of my mouth and I’ll eat your enemies) account for a lot of niceness
in nature, but not all of it. Some animals, among them prairie dogs,
will cry out when a predator comes on the scene, at the cost of drawing
the predator’s attention to itself — even if non-kin prairie
dogs benefit from the heads-up.
Pennock says Avidians sometimes behave the same way. “Under certain
conditions, they give away a lot of their computer time and evolve quite
a bit of this altruistic behavior, and under other conditions they stay
more selfish,” he says. Pennock and Clune are entertaining the
still-controversial hypothesis that one-for-all-and-all-for-one groups
of otherwise unrelated organisms simply emerge at random, like so many
ad hoc Masonic lodges. They hypothesize further that these groups, accidentally
packed with altruistic members, tend to leave the mutually destructive,
dog-eat-dog pile-ups behind in the evolutionary race, passing on their
goody-goody genes to the next generation.
That’s only one of the avenues of study the Avida team is testing.
Another graduate student is plumbing the evolutionary advantages of
sex, which seems at first blush to be an unnecessary energy drain (no
snickering from the back row). Still another student is looking at parasites.
“It turns out that parasites evolve and glom on to other programs,”
Pennock says. “It’s the opposite of altruism.” He
lets loose a big, honking scientist’s laugh at the notion.
And there’s much more work to do. Matt Rupp, a doctoral student
in computer science, is trying to generate conditions under which Avidians
begin to replicate completely by accident. If he succeeds — and
he’s getting there — it will eliminate the need to program
replication into the first ancestor, blunting one of the arguments leveled
by creationists at research on the origin of life. “It removes
one more variable if humans didn’t ‘touch’ these organisms
[even at the beginning],” Rupp explains.
Perhaps because Pennock’s first book (“Tower of Babel: The
Evidence against the New Creationism”) made him their nemesis,
he and the rest of the team face the ire of creationists all the time.
Many have even downloaded the Avida programs in hopes of finding a meaningful
flaw (they haven’t). After the Discover article came out last
month, the magazine sent him a pile of adverse mail.
“Usually these objections are based on misunderstandings of the
work we do here,” Pennock says. A common cavil is that the computer
was designed by a human, so an intelligent designer must be directing
the evolution that goes on inside. But that’s like accusing Isaac
Newton of plucking an apple from a tree and holding on to it all the
“Nobody disputes that humans built the computer,” says Pennock
sourly, “but the intelligence that went into building a computer
had nothing to do with the organisms and what they were able to do at
the end of a run.”
When discussion turns to the practical applications of digital evolution,
Pennock re-animates. “There’s a whole range,” he says.
“Engineers are finding that if you let evolution do it, you can
come up with better design solutions.”
Even Hollywood is in on the act. “Many of the battle scenes of
the movie “Troy” used virtual soldiers and stunt doubles
whose software motion controllers were evolved,” Pennock notes.
(Seems we just can’t get away from Brad Pitt.)
He weighs his speech much more carefully, however, when it comes to
heavier stuff. Conceding that applications are already in development
in the auto industry, his diction lengthens into SUV-sized pauses. “Much
of this stuff … is still … held … pretty tightly …
because it’s giving them a competitive advantage,” he says
(ever the evolutionary scientist). “So they really don’t
want to reveal too much about how they’re using it.”
Practical application is also the subject of Pennock’s next book,
which might explain the sudden tightness of lip.
One application Pennock is eager to mention is an undergraduate-friendly
version of Avida that will enable younger students to enjoy lab sessions
in evolutionary science. “It’s very hard to test evolutionary
hypotheses using regular organisms, because they’re so slow,”
he explains. The software will be tested in Pennock’s Lyman Briggs
classes next year, and then distributed nationally.
Pennock and Clune are also developing a high-school version, raising
the exciting prospect that “amateur” researchers might add
to the world’s store of knowledge on evolution someday, just as
an army of backyard astronomers are discovering comets and galaxies
For his part, Clune is eager to expound on Avida’s implications
for the future. “This stuff is going to change the face of modern
technology,” he proclaims.
“It excels at doing things humans don’t do well, like dealing
with complex systems with a lot of interacting parts. It could optimize
how streetlights should be timed, where you should build roads, how
companies ship their products, how universities schedule their final
Clune says that instead of figuring out the solution to a knotty design
problem, working through every painful step and frustrating dead end,
engineers may someday be able to simply “grow” one.
“The cool part is, it solves problems without human input.”
And then there is the philosophical dimension. “What counts as
life?” asks Pennock. “This turns out to be an important
question, even for things like the Mars program. If you’re sending
something up looking for life on other planets, you better have thought
carefully about what life is, so you can recognize it when you find
it. That’s a philosophical question.”
But it’s money that usually talks, even in science. Pennock, who
has been fighting troglodytic opponents of evolution for decades, adds
sardonically that digital evolution may even help convince Americans
who still don’ t accept Darwin.
“It works!” he laughs, as if dangling a finely wrought pocket
watch before an infant. “See, you can make money with it!”
to respond? Send letters to firstname.lastname@example.org.
our Letters policy.