Erik Larson

Aug 31, 2016

How is Jarvis AI?

Everything is AI these days, especially face and voice recognition software. Take Jarvis, Facebook CEO Mark Zuckerberg’s latest public relations ploy, complete with stock quips about annoying his wife with gadgetry. The prototype, named after the butler-voiced AI assistant to Tony Stark of blockbuster movie “Iron Man” fame, controls Zuckerberg’s home environment by translating his voice and location into commands, like changing room temperature, or opening a gate or door.

Fine. Sure. This kind of geek tech is cool (I suppose), but how, exactly, is it “AI”? The knee-jerk response I often hear is that AI is the reproduction of behavior or thinking that once required the human brain. Sure, like the desk calculator, as an astute reader pointed out recently in a prior post. (After all, who can multiply large numbers in their head quickly?)

It seems no one really knows what AI is, in the first place. It’s basically undefined, and so can be invoked anywhere there’s a sci-fi feel to some new tech or gadget. It can mean anything.

Futurists like Ray Kurzweil and other AI enthusiasts are quick to point out that AI is a moving target, as more and more of our intellectual lives are increasingly augmented or replaced by computation. This is no doubt true (a truism), but it fails to explain anything. Sure, voice recognition is getting better, and such capabilities were once the sole purview of human minds. Well, so what? The same could be said of the calculator, computers playing checkers or chess, and even in the not-too-distant future driving a car. It’s entirely reasonable that such individual feats by computers simply leave broader philosophical questions about computers getting “minds” and becoming “intelligent” unanswered.

Sure, the iconic Turing Test stipulates that outward behavior is all that matters, but it’s notable that some specific bit of cognition like doing math or playing chess (or recognizing a voice) isn’t the focus of the TT. Instead it’s language use: generic, open-context language use, like in an everyday conversation. That would be evidence of mind, perhaps. But AI today isn’t some generally intelligent activity like conversation but rather a million narrow applications. Success at narrow applications tells us computers are getting more sophisticated; it leaves open whether they are intelligent.

Cut to the chase: what is “intelligence,” anyway? We call something intelligent that has a purpose, forms goals, and has some inner life so that meeting goals satisfies it, and failing to meet them frustrates it. Performing algorithmic feats on multiplication, chess, voice or face recognition, or what have you isn’t the same thing. AI folks love to conflate application progress with the deeper kind of progress that speaks to the question of mind. But it’s all wishful thinking, and sci-fi fuzzies. Zuckerberg’s “Jarvis” is AI, says he. Sure. But AI here is just the buzzword for something “cool.” (It’s also topical, as popular deep learning techniques used by Jarvis are all the rage for voice and image data processing generally.)

Application progress, too, is not so simple. Not everyone shares Zuckerberg’s enthusiasm for computer assistance opening doors or changing room temperatures. Just ask his wife.