More Machine Now Than Man
Sunday, July 24, 2011 at 8:32PM In a 1950 article titled "Computing Machinery and Intelligence," English mathematician and computer scientist Alan Turing asked whether there could be "imaginable digital computers" that could imitate human behaviour convincingly enough that one could not tell the difference between them and real human beings. The hypothetical test he described has since come to be known as the Turing test: a human subject engages in normal conversation with both a computer and a human being in separate rooms, and the computer is said to have passed the test if the subject is unable to tell which converser is the computer and which is the human being.
Turing acknowledged that such machines were still the stuff of science fiction, but he didn't think we were very far off from creating them—he put their arrival at before the end of the century. A couple of generations later, futurist Ray Kurzweil predicted in 1990 that computers would pass the Turing test by 2020 (in 2005 he pushed it out to 2029). Most seem to agree that the ability of a computer to imitate human behaviour is still some way off, for one or the other of two different but related reasons: either the computer's capacity to match human traits (not just intelligence and sophisticated language, of course, but also the other things that make us human, such as emotion, error, inconsistency, and so on) is still relatively low, or human behaviour is somehow too "high," beyond the reach of imitative technologies.
Computers have come a long way since Turing died in 1954 (a tragic story well worth reading, by the way—what an embarrassment was the public mistreatment of the man whose cryptography skills helped crack the German Enigma machine in World War II, and all that mistreatment for nothing more than eccentricity and homosexuality). And while our computers have evolved and worked their way into the centre of our lives, we have evolved alongside them. They have attained much more of the sophistication necessary to imitate our language and thought patterns, and so pass the Turing test, and as we have hunkered down more or less permanently at their keyboards, our language and thought patterns themselves seem also to have changed a little: grown a bit more abbreviated, systematic, formulaic, and telegraph-like, simplified like a machine-language of subroutines. At our brave posts behind those keyboards, free from at least some of the complex and nuanced immediate social pressure that used to partly fill our days, we have nudged human nature a little more toward the mechanical extremities of hostility, partisanship, and stark binaries. Our computers have become increasingly capable of imitating us, and at the same time, to some extent, we have obligingly begun imitating them in little ways.
The talk of tests and machines calls to mind other such diagnostics, such as the fictional Voight-Kampff test from Philip K. Dick's Do Androids Dream of Electric Sheep?, the basis for the 1982 film Blade Runner. The Voight-Kampff is a test for the presence of human empathy, which helps determine whether the subject is a human being or an android. Subjects are asked a series of questions designed to provoke empathetic emotional responses while those administering the test watch for the physiological signs of such responses. The test rests on the presumption that human empathy cannot be faked at this level; if a creature exhibits no signs of empathetic response, it must be a machine.
The trick in Dick's novel, of course, is that no faking is required: only the androids display any genuine human empathy anyway. Dick's postmodern, futuristic world is one in which such human traits are on the way out. The utility of the Voight-Kampff test is challenged partly by advances in android technology, and partly by the fact that the disappearance of human empathy ("flattening of affect") moves the bar down, so that it is a less remarkable achievement to display as much empathy as the average human being. Just about any machine can manage that, so the Voight-Kampff test gets easier to pass.
A computer that passes the Turing test is a victory for silicon and a defeat for flesh and blood, of course, since a passed test means a human failure to detect humanness. If we bring down the bar of human nature and make it easier for the computer to win, just as the "flattening of affect" makes it easier to pass the Voight-Kampff test in Blade Runner, then we lose twice, since we have been beaten by our creations and also by ourselves, having given up what made us marvellous and inimitable in the first place.
I think about tests of our nature like these sometimes when everyone takes to the internet after some tragic event, so naturally they came to mind this past week amidst news of explosions in Norway and the death of Amy Winehouse. These are the moments, after all, when we glimpse ourselves at our most guileless and authentic, if not necessarily at our best.
One could easily have programmed a computer to produce the predictable responses that flooded people's Twitter feeds after these events—the hollow tributes, the jingoism, the geopolitical oversimplifications, the cynical cracks about the 27 Club, the tasteless too-soon humour, the excoriations of mourners of the latter tragedy for ignoring the greater enormity of the former. These tweeters had all clearly passed the reverse Turing test at least once when they filled in CAPTCHAs while signing up for their Twitter accounts; it wasn't at all clear where they stood with the regular Turing test, or whether they might rise out of the mechanical chatter of the site's equally predictable and formulaic spambot white-noise. Above all, and most disconcertingly, there was that "flattening of affect," the lowered bar of empathy against which all machines are suddenly judged as people, and all people suddenly machines. Rather than warmly demonstrating empathy—understanding and sharing the feelings of others—so many simply carried out the usual cold, pre-programmed scripts: Detached Internet Cynicism 1.0, Celebrity Judgment 2.0, Hairtearing & Teethgnashing 3.0, Bitter Recrimination Lite, Panic 2000.
That our computers might one day effortlessly pass the Turing test is an ambitious goal for artificial-intelligence research; if it happens, here's hoping it's because our machines managed to become more human, and not the other way around.
Reader Comments (2)
"The human failure to detect humanness...." What a sparse and precise way of expressing what we will lose and are losing via our own melding, via little fingertaps, with the technology that has, in some ways, come to stand in for the human connection involving warmth, empathy, and curiousity about the world rather than the need to speak from our own egocentric (and often uninformed and unexamined) viewpoint. Thank you for this. You're helping me to understand the wired world without completely becoming a Luddite.
Warmth, empathy, and curiosity about the world: that's it exactly, especially the last one. It's worrisome when people are cold and detached, but they can sometimes be these things temporarily; it's a disaster, though, when they're incurious, because it means cold and detachment are staying for good.