Driverless cars are not the only area where less of substance is being communicated than a year or two ago. This combined field, which is now hard to split into its two parts and within the decade may in most observers’ minds be fully merged, has over the past seven months only got four articles, beyond technical updates, on my desk. What do they have to say?
The oldest, Salon’s November 23rd “Robots are making us less human, too: “Certain things essential to the democratic fabric erode,”” didn’t follow its title, but provided ideas worthy of discussion. It was an interview of television director Maxim Pozdorovkin, whose work “The Truth About Killer Robots” ran on HBO November 26th. The show considered “the ethical dimension of using robots in our everyday life,” with concerns based on the worst these automata have given us, from spectacular industrial accidents to one employer after their implementation going from 3,200 workers to 800, and on to workers’ general losses, as in Kurt Vonnegut’s Player Piano, of “self-satisfaction” and “a sense of becoming better.” Pozdorovkin also claimed the shrinking likelihood of doing tasks such as banking and hotel check-ins with humans “undercuts the social fabric.” These are real if sometimes here overstated problems – how can we best deal with them?
An intriguing characteristic of good artificial intelligence systems is that they throw away what we consider best practices and invent their own, which are often superior. That was the underlying main point of “One Giant Step for a Chess-Playing Machine” (Steven Strogatz, The New York Times, December 26th.) The piece discussed AlphaZero, described as “a deep-learning algorithm” as opposed to “the world’s strongest chess programs” which had a larger base of technical knowledge but lacked insight into unprogrammable “basic principles.” AlphaZero, after it “played against itself millions of times and learned from its mistakes,” “crushed… the reigning computer world champion” in a hundred-game match in which it, though having 72 draws, won the remaining 28 and was thus undefeated. That could only have been because it was developing its own ways of winning. Of course, chess is only one opportunity for such self-directed thinking machines, of which AlphaZero constituted “humankind’s first glimpse of an awesome new kind of intelligence”; Strogatz mentioned “the great unsolved problems of science and medicine, such as cancer and consciousness; the riddles of the immune system, the mysteries of the genome.” Scary, and we may be there before we know it.
On the diametrically opposite side, we had “’AI could send us back to the stone age’: In conversation with the End Of The World” (Olivia Tambini, TechRadar, January 31). This piece is really a summary of previously published work, some in this blog, about the dangers of “general intelligence” capability, such as AlphaZero above freed to solve the world’s problems. We again, though with different words, got the Terminator autonomous goal-seeking problem, along with the results of an interview with author and philosopher Nick Bostrom, concerns similar to those named in the landmark now-19-years-old Bill Joy Wired paper “Why the Future Doesn’t Need Us,” the black-box nature of high-level machine knowledge development, and the need to somehow program “benevolence.” Nothing especially new here, but one of the world’s most suitable topics for repeating, re-repeating, re-re-repeating, and so on.
Last was Joanie Courtney’s April 5th Fox Business “The robots are here: New, unheard-of job titles signal growing occupations in digital age.” There are not enough of those I have always said, and nothing here, especially Courtney’s efforts to blame employers’ insufficiently paying practices on workers, changed my view. It is true that we should “encourage among future workers… the ability to keep learning and adapting,” though that is not a skill but a meta-skill, and that there is a gap between ordinary people becoming computer-theory experts and them responding to eras ending with the folded arms of 1980s autoworkers, but the “simulation training” Courtney advocates needs to be much more than that.
Next week, on to autonomous vehicles and my annual projections.