explains some mystifying experimental fact or when a sculptor suddenly grasps how to complete a work of art. Both experience an “aha!” moment that tells them they’ve found what they want, even if they can’t explain where it came from.
Wigderson regards this analogy as more than a clever metaphor. If the brain is a computer—a fantastically complex computer, but a computer nonetheless—then leaps of intuition, creativity, and aesthetic judgments must ultimately be the results of reasonably efficient computations. Understanding these mysterious processes in algorithmic terms remains a far-off goal, but computer scientists, neuroscientists, and evolutionary biologists are beginning to use the notions of computational complexity and intractability to understand how humans and other animals process information and make decisions.
For example, Adi Livnat, a Miller Postdoctoral Fellow at the University of California at Berkeley, and Nicholas Pippenger, a mathematics professor at Harvey Mudd College, have developed a model that uses a sort of inverse of the Nash equilibrium to define conflict in neural systems trying to deal with warring impulses. An animal that has to endure electric shocks in order to obtain food, for instance, must balance desire for food against fear of physical harm. Regarding these two urges as separate agents within the animal’s brain, conflict arises, Livnat says, because each agent tells the other, “I want you to do something else.”
Evolutionary theorists have argued that the brain ought to develop systems to resolve warring impulses in an orderly manner, yet the fact that humans and other animals can be rendered helpless by indecision suggests this isn’t so. Livnat and Pippenger mathematically demonstrated that conflicting agents in the brain, each independently pursuing its own goal, can nevertheless contrive to produce a beneficial compromise. Crucial to their model was that computing resources were limited: in that case, it turns out that internal conflict can produce an adequate solution where a perfect solution is beyond the system’s computational capacity.
In general, however, understanding neural systems from a computational perspective is a formidable task. As
Luis von Ahn, a professor of computer science at Carnegie Mellon University, points out, many of the tasks that computers can’t do are so easy for humans that we don’t think of them as requiring mental effort—recognizing the letter A no matter what style it is written in, for example, or being asked to say whether a picture has a cat in it.
As an empirical way to explore the types of computation that are involved in processes such as image perception, researchers have built neural networks or other adaptable computer systems that can “learn” from a set of training examples. While these efforts can produce improvements in task performance, understanding that improvement in direct algorithmic terms is difficult. Training tends to produce systems that are “completely incomprehensible” in algorithmic terms, says von Ahn. Knowing their structure doesn’t mean you know what they’re doing.
It’s far from clear, adds neuroscientist Larry Abbott of Columbia University, that knowing how computers cope with a task like pattern recognition would necessarily help in understanding how the brain does it. Abbott doesn’t doubt that the human brain works in an essentially algorithmic way, but “the hardware is very different.” Compared to a computer, he says, the brain is “slow and unreliable,” yet it seems far more internally active in that it constantly tries out different ways of processing information rather than sticking to a preordained routine.
Although some philosophers and other researchers disagree, computer scientists and neuroscientists almost universally believe the brain does what does by performing extraordinarily elaborate computations. Millions of years of evolution have trained neural systems in a more or less ad hoc fashion to perform numerous tasks. Neuroscientists are trying to tease out the essential features of these systems by comparing them to computational models and algorithms. The day might come, says Abbott, when it is possible to understand an organism’s behavior in terms of its neural circuitry.
David Lindley is a science writer and author based in alexandria, va. richard m. Karp and Christos papadimitriou, both of the university of California at berkeley, contributed to the development of this article.
For many teenagers, texting is replacing talking on cell phones, according to a new online poll of 2,089 u.s. teenagers conducted by ctiA, the international wireless telecommunications association, and harris interactive. teens say they spend nearly as much time talking as they do texting, and prefer texting as it “is all about multitasking, speed, privacy, and control,” says Joseph Porus, a harris interactive vice president. how pervasive is texting? of the respondents, 42% say they can text blindfolded and 47% say that without texting their social life would deteriorate or simply end.
teens’ suggestions for future cell phones include phones in the form of sunglasses and jewelry; a “dream device” that is software based and would enable “the user’s fingerprint to turn anything” into a mobile device; and a LEgo-like design so a user can “build just what he wants for every occasion.”
As for landline phones, 40% of the teens say a cell phone is the only phone they will ever need.
Elvira Fortunato, a professor
of materials science at the
new university of Lisbon,
has developed a paper-
based transistor that might
be suitable for disposable
electronics, such as rFid tags
and smart labels, New Scientist
reports. Fortunato built the
transistor by covering both
sides of a sheet of paper with
metal oxides before applying
aluminum contacts. the
transistor acts as a flexible
substrate and as an integral
part of a semiconductor by
helping to amplify the electrical
current that passes through
the transistor. the transistors
are vulnerable to tears or
wetness, but both problems
can be overcome by laminating
the device.
References:
Archives