BOOK REVIEW DESK
Is This Chip Educable?
By
CARL ZIMMER
Published: March 10, 2002,
Sunday
ARTICLE TOOLS
DIGITAL BIOLOGY
How Nature Is Transforming
Our Technology and Our Lives.
By Peter J. Bentley.
Illustrated. 272 pp. New York:
Simon & Schuster. $25.
Biologists tolerate a level of mystery
in their work that would drive your average engineer or computer programmer
crazy. They've put together a complete rough draft of the human genome but they
have little understanding of how those 40,000 or so genes work together to make
a human. They've mapped every muscle and nerve in a fly's wings, yet still
struggle to explain how it keeps from crashing into a wall. No engineer would
build a DVD player without knowing what every circuit was for; no programmer
would let a computer write its own code. Or at least that's how things used to
be. As Peter J. Bentley demonstrates in ''Digital Biology,'' the cool, rational
temple of technology is becoming infested with biology's weedy enigmas.
Microchips, for example, can now evolve.
Bentley describes how Adrian Thompson, a British engineer, came up with a few
dozen random arrangements of transistors and programmed a computer to test how
well they did various jobs, like distinguishing between high-pitched and
low-pitched tones. The first generation of chips always performed miserably,
but some of them a little less miserably than the rest. The computer saved the
less miserable designs and combined them into hybrids. In the process, it also
sprinkled a few random changes into the designs, mutations if you will. A few
offspring could distinguish between the tones slightly better than their
parents -- and they produced a third generation. By mimicking evolution for a
few thousand rounds, the computer produced chips that did their job exquisitely
well. But Thompson doesn't quite know how they work. To understand them, he
resorts to measuring the temperature of parts of the chips, like a neurologist
using an M.R.I. scanner to probe a brain.
People have been exploring digital
biology since the 1970's, and Bentley's book is not the first history. Its
predecessors include Steven Levy's ''Artificial Life'' (1993), Kevin Kelly's
''Out of Control'' (1994) and ''Emergence,'' by Steven Johnson, published last
year. In some ways, ''Digital Biology'' suffers by comparison. Some of
Bentley's case studies have been written about before, and he doesn't try hard
to explain how digital biology may transform culture. He promises it will
change our lives, but backs up the claim only with lists of coming appliances:
washing machines with chaos-theory-driven spin cycles! Carpet-cleaning
microrobots! But more stuff tends to clutter life, not change it. To observe
people leading an utterly conventional lifestyle, just watch ''The Jetsons.''
Yet Bentley has an important advantage
over previous chroniclers: he is a digital biologist himself. (He teaches at
University College, London, and specializes in evolutionary computing.) Digital
biology is at a crucial point in its history; it is quickly changing from thought
experiments into a real science, and Bentley is part of the experience. His
book is fascinating because it gives us a sense of what it's like to be
overwhelmed the way Bentley is these days -- he calls it ''riding a tornado.''
He has also become a discerning student of biology. He demonstrates a good
sense of what biologists know about how life works and what they don't. And he
shows how biology is essential to the work he does. The strategy ants use to
follow scent trails becomes a method for laying out networks of cellphone
towers. The way embryos develop becomes a method for turning a small program
into a complex one without any intervention from a programmer.
Bentley is interested in more than just
building the next algorithm. He wants to understand the deep meaning of digital
biology -- what common principle ties together projects as disparate as
computer immune systems, neural networks and virtual ant colonies. He believes
complexity can emerge spontaneously in any system in which many parts interact
according to certain rules. The rules can be simple, but it's crucial that each
part, be it a neuron or a chunk of programming code, can affect the behavior of
other parts, creating a complex pattern of feedbacks.
The rules of a system actually matter
more than the stuff the system is made of. In fact, the stuff matters so little
that Bentley sees no real difference between digital biology and biology
outside of a computer. To him, there is nothing artificial about artificial
life: ''The first person to hold a conversation with an alien intelligence will
not be an astronaut, it will be a computer scientist or computational
neuroscientist, talking to an evolved digital neural network.''
In a sense Bentley is right, but in a
sense that is nearly meaningless. Computers don't replicate nature; they
replicate what we think we know about nature. And biologists are the first to
tell you their models of the brain, the immune system or the network of
proteins in a cell are pretty crude. Computer programs modeled after these
models are even cruder. The most complex digital ''brain'' consists of a few
thousand simulated neurons -- a far cry from the human brain, which consists of
100 billion neurons, each of which is connected to thousands of its neighbors
and uses dozens of neurotransmitters to communicate with them. To treat them as
the same thing is a bit like treating four notes played in a thousand
combinations as the same thing as Mahler's Ninth Symphony. They share some things,
but not the things that really matter.
A biological concept doesn't even have
to be true to make for good software. Bentley describes how computer scientists
invented a way to destroy computer viruses based on a 1970's model of the
immune system. But immunologists now consider the model a failure. The fact
that computer programmers can turn a failed biological idea into a powerful
program is proof that life and machinery are not interchangeable. Instead, they
both draw their strength from a common source -- the murky depths of
complexity.
The relationship works both ways. Just
as computers can be lifelike, biologists realize that in a lot of ways life
acts like a computer. Bentley doesn't spend much time on this, but it is an
astonishing development in biology. Neuroscientists build neural networks to
understand how different parts of the brain work; researchers who study insect
navigation build robots to test their ideas; biochemists now treat genes as if
they were lines of code in a piece of software. Thinking of life as a computer
doesn't drain the majesty from life. In fact, its grandeur only deepens. And
that kind of insight is worth more than all the carpet-cleaning robots in the
world.
Carl Zimmer writes
a column about evolution for Natural History. His most recent book is
''Evolution: The Triumph of an Idea.''
Published: 03 - 10 - 2002 , Late
Edition - Final , Section 7 , Column 1 , Page 25