HA D.H. LE
“I’ve never loved anyone the way I loved you,” he says.
“Me too. Now we know how,” she responds.
The exchange reads like a movie scene between two lovers, but there is a twist. The man’s lover is a computer operating system named Samantha, and the movie is Her, Spike Jonze’s 2013 film about the romantic relationship between a man and his computer. This idea of love between man and machine is not novel. Spielberg covers the topic in the 2001 film A.I. Artificial Intelligence. Artificially sentient being Jane appears in several books in the Ender series by Orson Scott Card, falling in love with Ender’s stepson and even marrying him later on.
Advances in computing are in a period of rapid development. From Apple’s sassy iPhone “intelligent personal assistant” to Watson, a computer designed to play Jeopardy which has defeated human Jeopardy champions, the possibility of artificial sentience seems to be on the horizon. Yet, while these predictions of a future filled with sentient computers is alluring, computer scientists say modern technology is not quite there yet.
“The presentations of artificial intelligent computers and robots in movies like Her, AI, 2001, and Blade Runner are grossly beyond state of the art,” says Stuart Shieber, Welch Professor of Computer Science in Harvard School of Engineering and Applied Sciences (SEAS). “The computers in these films would easily pass the gold standard for intelligence, the Turing Test.”
The Turing Test is a method to probe the question of whether or not machines can think. Proposed by Alan Turing in 1950, the test involved an “imitation game,” in which an interrogator tried to distinguish between a human and a machine (Shieber 7). In his introduction to his book, The Turing Test: Verbal Behavior as the Hallmark of Intelligence, Shieber explains that instead of defining intelligence as a fixed quality, the test looks at behaviors indicative of thought and performs statistical analyses of those behaviors to achieve a conclusion (7-8). In the original example, Turing proposed that a human “judge” would independently engage two partners in conversation, a human and a computer. If the judge could not distinguish the human from the computer, the computer would have been said to pass the test (Oppy and Dowe). The test is one measure of artificial intelligence and compared to Samantha in Her, current technology fails it.
… it is difficult to name the most important new advancements and research in technology because there have been too many.
While Siri and Watson are masterpieces in computer science and engineering, neither are close to passing the Turing Test. In Watson’s case, though Jeopardy is a natural language-based challenge, its game play does not require spontaneous unpredictable interactions. According to Shieber, the challenge has three constraints: one, the player can pick the questions to answer; two, the structure of the questions is limited; and three, answers are restricted to objective facts and the question topics are also narrow. Watson is capable of working within these constraints, but anything else is impossible. The result is Watson’s knowledge cannot be extended to regular conversations or more subjective or cognitive dilemmas.
“It’s not just that Watson can’t handle this kind of interaction yet,” says Shieber. “Rather, the entire approach to question answering that Watson takes is not amenable to extension to other areas.”
Shieber does not dismiss artificial intelligence’s potential, however. There has been tremendous progress in computer technology, particularly where solutions are more concrete. “Much of the progress has been based on new techniques and breakthroughs in statistical modeling and machine learning,” he said. “You see this progress every day in systems like Siri, Watson, Google Translate, Dragon Naturally Speaking, music recommendation systems, credit card fraud detectors.”
For Shieber, it is difficult to name the most important new advancements and research in technology because there have been too many. One notable system is Harvard’s TERMES system, which demonstrates that a collection of robots can build complex structures without a central command. Still, human-AI romance and the emotional and intelligence capacity required for it remains an area of mystery.
“To my mind, there is no reason in principle that computers could not be thinking things with intelligence, emotion, and the like,” Shieber says. “Whether we are ever actually able to build such a computer is an open question.”
Will artificial intelligence be attainable in our generation? Shieber is doubtful. He described a bet between two prominent engineers, Mitch Kapor and Ray Kurzweil, concerning when artificial intelligence will be attainable. Kurzweil, the lead engineer at Google, is more optimistic, while Kapor, prominent software engineer and founder of the Lotus Development Corporation, is less certain.
“Ray Kurzweil predicted 2029 and has a bet with Mitch Kapor riding on it,” Shieber says. “I’m with Kapor on this one.”
- Oppy, Graham and David Dowe. “The Turing Test.” Stanford Encyclopedia of Philosophy. Stanford: Center for the Study of Language and Information. 26 January 2011. Web. 24 March 2014.
- Shieber, Stuart M. The Turing Test: Verbal Behavior as the Hallmark of Intelligence. Cambridge: Massachusetts Institute of Technology, 2004. Print.
Ha Le is a Brevia contributing writer. She can be reached at email@example.com.