Erin Lubin/The New York Times
By JOHN MARKOFF
Published: December 28, 2013 258 Comments
PALO ALTO, Calif. — Computers have entered the age when they are able to
learn from their own mistakes, a development that is about to turn the
digital world on its head.
The first commercial version of the new kind of computer chip is
scheduled to be released in 2014. Not only can it automate tasks that
now require painstaking programming — for example, moving a robot’s arm
smoothly and efficiently — but it can also sidestep and even tolerate
errors, potentially making the term “computer crash” obsolete.
The new computing approach, already in use by some large technology
companies, is based on the biological nervous system, specifically on
how neurons react to stimuli and connect with other neurons to interpret
information. It allows computers to absorb new information while
carrying out a task, and adjust what they do based on the changing
signals.
In coming years, the approach will make possible a new generation of
artificial intelligence systems that will perform some functions that
humans do with ease: see, speak, listen, navigate, manipulate and
control. That can hold enormous consequences for tasks like facial and
speech recognition, navigation and planning, which are still in
elementary stages and rely heavily on human programming.
Designers say the computing style can clear the way for robots that can
safely walk and drive in the physical world, though a thinking or
conscious computer, a staple of science fiction, is still far off on the
digital horizon.
“We’re moving from engineering computing systems to something that has
many of the characteristics of biological computing,” said Larry Smarr, an astrophysicist who directs the California Institute for Telecommunications and Information Technology, one of many research centers devoted to developing these new kinds of computer circuits.
Conventional computers are limited by what they have been programmed to
do. Computer vision systems, for example, only “recognize” objects that
can be identified by the statistics-oriented algorithms programmed into
them. An algorithm is like a recipe, a set of step-by-step instructions
to perform a calculation.
But last year, Google researchers were able to get a machine-learning
algorithm, known as a neural network, to perform an identification task
without supervision. The network scanned a database of 10 million
images, and in doing so trained itself to recognize cats.
In June, the company said
it had used those neural network techniques to develop a new search
service to help customers find specific photos more accurately.
The new approach, used in both hardware and software, is being driven by the explosion of scientific knowledge about the brain. Kwabena Boahen, a computer scientist who leads Stanford’s Brains in Silicon research program, said that is also its limitation, as scientists are far from fully understanding how brains function.
“We have no clue,” he said. “I’m an engineer, and I build things. There
are these highfalutin theories, but give me one that will let me build
something.”
Until now, the design of computers was dictated by ideas originated by the mathematician John von Neumann
about 65 years ago. Microprocessors perform operations at lightning
speed, following instructions programmed using long strings of 1s and
0s. They generally store that information separately in what is known,
colloquially, as memory, either in the processor itself, in adjacent
storage chips or in higher capacity magnetic disk drives.
The data — for instance, temperatures for a climate model or letters for
word processing — are shuttled in and out of the processor’s short-term
memory while the computer carries out the programmed action. The result
is then moved to its main memory.
The new processors consist of electronic components that can be
connected by wires that mimic biological synapses. Because they are
based on large groups of neuron-like elements, they are known as
neuromorphic processors, a term credited to the California Institute of
Technology physicist Carver Mead, who pioneered the concept in the late 1980s.
They are not “programmed.” Rather the connections between the circuits
are “weighted” according to correlations in data that the processor has
already “learned.” Those weights are then altered as data flows in to
the chip, causing them to change their values and to “spike.” That
generates a signal that travels to other components and, in reaction,
changes the neural network, in essence programming the next actions much
the same way that information alters human thoughts and actions.
“Instead of bringing data to computation as we do today, we can now bring computation to data,” said Dharmendra Modha,
an I.B.M. computer scientist who leads the company’s cognitive
computing research effort. “Sensors become the computer, and it opens up
a new way to use computer chips that can be everywhere.”
The new computers, which are still based on silicon chips, will not
replace today’s computers, but will augment them, at least for now. Many
computer designers see them as coprocessors, meaning they can work in
tandem with other circuits that can be embedded in smartphones and in
the giant centralized computers that make up the cloud. Modern computers
already consist of a variety of coprocessors that perform specialized
tasks, like producing graphics on your cellphone and converting visual,
audio and other data for your laptop.
One great advantage of the new approach is its ability to tolerate
glitches. Traditional computers are precise, but they cannot work around
the failure of even a single transistor. With the biological designs,
the algorithms are ever changing, allowing the system to continuously
adapt and work around failures to complete tasks.
Traditional computers are also remarkably energy inefficient, especially
when compared to actual brains, which the new neurons are built to
mimic.
I.B.M. announced last year that it had built a supercomputer simulation
of the brain that encompassed roughly 10 billion neurons — more than 10
percent of a human brain. It ran about 1,500 times more slowly than an
actual brain. Further, it required several megawatts of power, compared
with just 20 watts of power used by the biological brain.
Running the program, known as Compass, which attempts to simulate a
brain, at the speed of a human brain would require a flow of electricity
in a conventional computer that is equivalent to what is needed to
power both San Francisco and New York, Dr. Modha said.
I.B.M. and Qualcomm, as well as the Stanford research team, have already
designed neuromorphic processors, and Qualcomm has said that it is
coming out in 2014 with a commercial version, which is expected to be
used largely for further development. Moreover, many universities are
now focused on this new style of computing. This fall the National
Science Foundation financed the Center for Brains, Minds and Machines, a new research center based at the Massachusetts Institute of Technology, with Harvard and Cornell.
The largest class on campus this fall at Stanford was a graduate level
machine-learning course covering both statistical and biological
approaches, taught by the computer scientist Andrew Ng. More than 760 students enrolled. “That reflects the zeitgeist,” said Terry Sejnowski,
a computational neuroscientist at the Salk Institute, who pioneered
early biologically inspired algorithms. “Everyone knows there is
something big happening, and they’re trying find out what it is.”
No comments:
Post a Comment