The Information - James Gleick [173]
Von Neumann’s error was discovered by the scientist who became Bennett’s mentor at IBM, Rolf Landauer, an exile from Nazi Germany.♦ Landauer devoted his career to establishing the physical basis of information. “Information Is Physical” was the title of one famous paper, meant to remind the community that computation requires physical objects and obeys the laws of physics. Lest anyone forget, he titled a later essay—his last, it turned out—“Information Is Inevitably Physical.” Whether a bit is a mark on a stone tablet or a hole in a punched card or a particle with spin up or down, he insisted that it could not exist without some embodiment. Landauer tried in 1961 to prove von Neumann’s formula for the cost of information processing and discovered that he could not. On the contrary, it seemed that most logical operations have no entropy cost at all. When a bit flips from zero to one, or vice-versa, the information is preserved. The process is reversible. Entropy is unchanged; no heat needs to be dissipated. Only an irreversible operation, he argued, increases entropy.
Landauer and Bennett were a double act: a straight and narrow old IBM type and a scruffy hippie (in Bennett’s view, anyway).♦ The younger man pursued Landauer’s principle by analyzing every kind of computer he could imagine, real and abstract, from Turing machines and messenger RNA to “ballistic” computers, carrying signals via something like billiard balls. He confirmed that a great deal of computation can be done with no energy cost at all. In every case, Bennett found, heat dissipation occurs only when information is erased. Erasure is the irreversible logical operation. When the head on a Turing machine erases one square of the tape, or when an electronic computer clears a capacitor, a bit is lost, and then heat must be dissipated. In Szilárd’s thought experiment, the demon does not incur an entropy cost when it observes or chooses a molecule. The payback comes at the moment of clearing the record, when the demon erases one observation to make room for the next.
Forgetting takes work.
“You might say this is the revenge of information theory on quantum mechanics,”♦ Bennett says. Sometimes a successful idea in one field can impede progress in another. In this case the successful idea was the uncertainty principle, which brought home the central role played by the measurement process itself. One can no longer talk simply about “looking” at a molecule; the observer needs to employ photons, and the photons must be more energetic than the thermal background, and complications ensue. In quantum mechanics the act of observation has consequences of its own, whether performed by a laboratory scientist or by Maxwell’s demon. Nature is sensitive to our experiments.
“The quantum theory of radiation helped people come to the incorrect conclusion that computing had an irreducible thermodynamic cost per step,” Bennett says. “In the other case, the success of Shannon’s theory of information processing led people to abstract away all of the physics from information processing and think of it as a totally mathematical thing.” As communications engineers and chip designers came closer and closer to atomic levels, they worried increasingly about quantum limitations interfering with their clean, classical ability to distinguish zero and one states. But now they looked again—and this, finally, is where quantum information science is born. Bennett and others began to think differently: that quantum effects, rather than being a nuisance, might be turned to advantage.
Wedged like a hope chest against a wall of his office at IBM’s research laboratory in the wooded hills of Westchester is a light-sealed device called Aunt Martha (short for Aunt Martha’s coffin). Bennett and his research assistant John Smolin jury-rigged it in 1988 and 1989 with