Artificial Intelligence: It Still Stinks

Update 24 June 09: This Rodney Brooks article in the April 09 issue of Cosmos presents an expert viewpoint which I consider to be one of the most honest and insightful about AI, from an AI expert. Brooks has long been a champion of the "bottom-up" approach to creating lifelike behaviour in robots, and seems to understand the importance of "embodying" an intelligence, complete with emotions or quasi-emotions.

Forbes has published an Artificial Intelligence Report (via Instapundit) online that presents an array of viewpoints on the present and future of AI. From Kurzweil-like true believers in the coming of superhuman intelligent machines to skeptical wanna-believers like David Gelernter, over 20 insiders and quasi-insiders provide their insights into the likely future of AI as they see it.

Unfortunately, not even the insights of Ben Goertzel, Hugo de Garis, Judea Pearl, and other experts, provide any evidence that AI researchers have confronted -- much less overcome -- the monumental conceptual obstacles that have prevented AI research from "reaching first base" in the attempt at devising a human-level intelligent machine.

Despite the painfully obvious differences between human consciousness and algorithmic approaches (even algorithms informed by Bayesian networks) to solving problems, today's students and tomorrow's researchers appear to be stuck in mental conceptualisations of intelligence that were already outmoded 20 years ago. Until students are relieved of such dysfunctional shackles, it is likely that they will continue repeating the mistakes of their predecessors in AI research.

Herbert Gelernter reveals that IBM is planning to pit its "Blue Gene" project against human contestants on the Jeopardy game show. Such a test of general knowledge would certainly hit at the sense of human supremacy where it might hurt the most, should the computer win. But it is difficult to know where the human contribution begins and ends in such projects as "Blue Gene" and "Deep Blue." These projects are not actually artificial intelligences so much as highly tuned human-machine team efforts. Such a "neo-Turing Test" is an improvement over the original, but if IBM does not allow a thorough look "behind the curtain", the wizardliness of Blue Gene might not inspire any more confidence than Oz deserved.

Machine "intelligence" has made inroads in many areas, providing useful assistance to human experts in science, medicine, economics, commerce, transportation, the military, and increasingly in government. But these are all special-purpose problem solvers and aids to decision making. Not broad spectrum all-purpose intelligences in their own right, capable of learning and acting on their own initiative.

Human intelligence is the only "higher" intelligence that we know of. Until AI researchers confront the core problem of human intelligence -- embodied consciousness -- they will continue to beat around the bush for many more decades, and end up little ahead for all the effort.