AI Optimism: Reasons for Hope in the Science of Artificial Intelligence

Article excerpt

SKEPTICS HAVE LONG HAD A LOVE/HATE RELATIONSHIP with Artificial Intelligence (AI). On the one hand, we love science and technology and hold them up as our best hope for humanity's future, while our favorite science fiction films and television series teem with intelligent computers, androids, and robots. On the other hand, premature claims made in the name of Artificial Intelligence sound overblown and beyond current science and technology, so they become targets for skeptical debunking. (One possible explanation for this apparent dichotomy is that if the story is set far enough in the future we tend to be more willing to suspend disbelief in the interest of the plot, whereas if the story is set in the present or near future, our critical faculties for what is possible now or in the near future override our desire to suspend disbelief.) In Skeptic Vol. 12, No. 2, 2006, for example, Peter Kassan's "AI Gone Awry" affords readers a pessimistic assessment of the state of the AI dream and challenges computationalists who hold that "mechanical" AI is the inevitable outcome of Moore's law and the exponential growth of computational computing power, which they believe will be achieved in the short run. The problem may be more difficult than that--as Skeptic publisher Michael Shermer has been known to joke in his lectures, "AI is five years away ... and always will be."

Although skeptics have made many good points in challenging AI optimists, I want to revisit the subject and remind readers that humans once fantasized flying by attaching feathers or bird-like wing structures to their arms, but ended up creating wing/foil shapes that redirected air-pressure for lift. So, indeed, human (machine-based) flight did not simulate the flapping of birds' wings, but "humans traveling by air" became the result. More impressively, no one bothered fantasizing the possibility of discovering "clocks" running in reverse-time to measure the age of organic artifacts, but Willard Libby, leveraging applied chemistry and mathematics, computed the half-life of Carbon-14 present in organic material. Naturally, engineers who imagine building mechanical brains modeled on human ones may confront the possible design hindrances and computational overloading that Kassan identifies in his Skeptic article, but human scientific development has not advanced sufficiently for Kassan to declare his pessimistic view certified. Furthermore, innovators today don't need to construct a "live" (strong) AI device, as they can already simulate mechanically the illusion of (weak) human intelligence. Please allow me to unpack what that means.

Mechanical Strong AI? (1)

In the traditional AI approach, researchers distinguish between constructing machines that completely simulate the human brain (mechanical strong AI), or ones that simulate "intelligent" behaviors (mechanical weak AI).

Kassan suggests that in order to achieve "successful" AI billions of mechanical neurons that process information in microseconds will be needed. That may not be so. While cognitive and testing experts benchmark human intelligence by speed of ideation and response, this metric remains prejudicial. Mechanical brains do not need to match the synaptic speed and processing power of human neural networks in order to exhibit an "illusion" of intelligence--a premise that drives this article. Also, designers simulate slower intelligence with slower mechanical synapses. AI pessimists should not ignore the processing speed and power currently proposed by proponents of DNA computing or of molecular computers (15 x [10.sup.15] molecular devices in the space of a spoonful of water!), (2) nor should they dismiss possible quantum computers. (3) Consider also chess computers, such as Houdini 3, Stockfish DD, Komodo TCEC, Critter 1.6a, (4) which consult memory and render "weighted" decisions much more rapidly than their human counterparts. Their chipsets and decision-tree softwares are cross-associable to numerous other applications, a second premise driving this article. …