[SPOILER ALERT] R2D2, HAL 9000, the Terminator . . . at Next IT, we love seeing A.I. embodied in the movies. (Yes, we were pretty stoked about “Her.”) It’s fascinating to compare Hollywood’s portrayal of A.I. with technology that’s currently available, and to see how future visions differ.
Thanks to the growing awareness of artificial intelligence brought-on by the likes of Siri, Watson and Cortana, we’ve been out to the movies a lot lately. The most recent being “Ex Machina,” the story of a seductive robot with bad intentions and the poor soul whose heart she crushes.
In a super-secret location, the CEO of a big search-engine company asks a lucky employee, Caleb, to perform a Turing Test on a new (and, as it turns out, lovely) android. As the employee evaluates Ava, he falls for her and helps plot her escape from the evil CEO, who plans to wipe her memory at the completion of the experiment. Things don’t go so well for Caleb: as it turns out, Ava was just manipulating him to plot her own, solitary, escape.
What’s fascinating about “Ex Machina” is the way in which Ava gains the smarts to pull one over on both her creator and her evaluator. She’s been programmed to be self-evolving, learning to emulate a person by consuming the search-engine company’s vast database of interactions and inputs. And she’s picked up on the fact that how people ask a question yields many more insights than what they’re asking about. As psychologist and author James Pennebaker has found in his research of linguistics, it’s the little things that can give so much away.
When building the virtual assistants of today, the little things make all the difference. With the aid of some sophisticated software, language-model creators are able to see the many variations on a theme: the different ways people can ask the same question, and the small permutations that can change the meaning of a question or statement entirely. Keyword identification won’t cut it – human language is far too complex. (Which, incidentally, is one of traditional web-search’s big problems.) In Next IT’s classic example of understanding gone wrong, a homonym produces a Turing Test fail:
User: Can I bring my fly rod on the plane?
Virtual Assistant: Where would you like to fly today?
The creators of “Ex Machina” undoubtedly interviewed a natural-language specialist or two in the creation of the film, and it shows. A.I. of the future will certainly use language with finesse. But, if recent releases are any indication, it may be best to hold-off on any romantic intentions.