Selling the machine: how artificial is AI?
Looking around, you’d think artificial intelligence is about to happen. It’s as if the mass of all the speculation is going to magically bring it into being. In the meantime, corporations want you to think it’s already here.
I’ve engaged in this speculation myself. My novel Biome is all about the creation of a machine intelligence, which I propose will happen just 40 years in the future. As my lead character Ruth Shannon (whose baby this will be) explains in the novel, the correct terminology is MI since “there’s nothing artificial about intelligence.” That’s a point few of the speculators seem to have grasped yet.
So here’s another bit of Maas to add to the accumulating speculation, heading for that black hole moment when there’s a flash of light and the MI is sitting fully formed on our lab bench.
For all I’ve written about MI, I’m not convinced we’re on the brink of it. My suspicion is that machine intelligence is much further away than we’d like to think. In Biome, Shannon suffers more than a decade of failure and is about to give up when she gets help from outside. How likely is something if we can’t do it for ourselves?
You might as well say that there are so many people jumping off window ledges that we must surely be on the brink of learning to fly.
I don’t believe, for example, that we’ll create MI simply by feeding data into a powerful enough computer. I think the Turing Test is a red herring that is so easily faked it’s going to obstruct our recognition of true intelligence when and if it happens.
I suspect we’re engaging in the same kind of wish fulfilment we used to have when the UFO flap made us believe in the imminence of space visitors. The excitement passed. They never came.
A salutory precursor for what’s happening now is the “virtual reality” craze of the 1990s. Like AI, virtual reality had a very long history in the science fiction literature. Writers had been speculating about total immersion in a computerized fake reality since Daniel F. Galouye’s Simulacron-3 in 1963.
In the 1980s, virtual reality began heading toward critical mass. First there was cyberpunk, in which cool streetwise American kids surfed a technological frontier inaccessible (and incomprehensible) to their parents. Then came a deluge of British writers such as Jeff Noon and Paul McAuley who targeted VR fantasies at raved-up techno-hippies.
Meanwhile, corporations began to sell 3D visual systems and head-mounted displays as virtual reality. They weren’t, of course — nobody who put on one of those headsets truly believed they were in a separate reality. Though Douglas Trumbull’s excellent movie Brainstorm suggested practical uses for VR, for example in training pilots and soldiers, for the most part the technology remained locked in video games.
Finally, all these strands converged on a sudden eruption of Hollywood VR blockbusters, most notably The Matrix, with all their attendant media and quasi-intellectual buzz. And then —
The fad passed. Virtual reality never happened.
However, since the term “virtual reality” was hard to define, it was easy to pretend. So corporations sold virtual reality headsets as if The Matrix actually existed. For a while, businesses were even led to believe that their success depended on them having presence on Second Life.
Companies made a lot of money out of a lot of nonsense.
Immersive technology has huge promise, and huge benefits, in everything from medical procedures to driving simulators to therapy for those with mental and behavioral disorders. But it would be stretching the term to suggest that these are true virtual reality.
And the point is, it doesn’t matter that we never actually created a Brainstorm/Matrix form of VR. The term has become so obfuscated that many people probably think we did.
So here we are again, and though cool has moved on, gullibility has not. The corporations that once sold you the cool cachet of fake virtual reality are now trying to sell you the cool cachet of fake AI.
As an example, here’s David Marcus, head of Facebook’s Messenger service, announcing its “personal digital assistant” on 26th August 2015. This digital assistant, Marcus boasted, is “powered by artificial intelligence that’s trained and supervised by people.” They’re calling it “M.” That’s M for Messenger or Machine, I presume, not Marcus.
Wow, that’s thrilling. Somewhere in Facebook’s facilities in Silicon Valley, we’re led to believe, there’s an actual, real life machine intelligence, all bright-eyed and eager to learn. A team of lucky Facebook pioneers are sitting there “training” the MI, presumably by chatting with it, answering its philosophical questions, and easing it past the paradoxes that SF movies from Dark Star onwards have told us cause the self-destruction of any unsuspecting AI.
“Unlike other AI-based services in the market,” Marcus continues gushingly, “M can actually complete tasks on your behalf. It can purchase items, get gifts delivered to your loved ones, book restaurants, travel arrangements, appointments and way more.”
The “other AI-based services on the market” are presumably services like Apple’s Siri, Microsoft’s Cortana and Google’s Now, “intelligent” personal assistants that merge speech recognition with database mining skills to deliver what the undemanding might believe is some kind of human interaction (though it’s hardly on the level of Her).
But wait — what can this terrific AI do for you? Well, take another look at Marcus’s announcement. It can purchase things. Put bluntly, M is merely a way of channeling sales opportunities at its users. Hardly the thing any of us would want a real-life machine intelligence to do, but of course this is far from a real-life machine intelligence.
It’s a computerized sales rep in an age when we’re more likely to follow the suggestions of a computer than those of an actual person. Just a slightly more sophisticated version of Amazon’s algorithms suggesting other items we might like to buy.
The hint to the confidence trick is in Marcus’s elaboration of that confusing phrase “trained and supervised by people.” M’s responses, we learn, are supervised by a bank of Facebook employees — we used to label them call-center operators or customer service representatives — who either oversee M’s responses or write them themselves.
“M is powered by actual people,” as Marcus told Wired. Golly. (And isn’t that exactly what Person Of Interest explored in its episode “Q&A” back in February?)
Marcus admits that thousands of these “trainers” will be employed, all on a contract basis. And who will train the trainers? Presumably whatever companies Facebook partners with to ensure that M pushes their products.
M is a particularly bald-faced misappropriation of the term “artificial intelligence,” but it’s hardly alone, and the obfuscation is going to get a lot worse as innumerable start-ups try to get their own reactive computer systems into the market.
Neither is the fraud likely to be perpetrated only against the easily-impressed users of smartphones and Facebook. A few days ago (in October 2015) we learned that Bob Dylan, hero of college-level intellectuals everywhere, has teamed up with IBM’s Jeopardy!-winning Watson computer (the Deep Blue of game shows) to promote the corporation’s new “Cognitive Business Solutions” division, whose goal is to push the benefits of “artificial intelligence” at businesses.
It’s Second Life all over again.
If you’re like me, you probably can’t wait for this flap to be over with, so that we can go back to working on true machine intelligence, however long it takes.
But just as research into virtual reality all but stalled after its craze had passed, are services such as M going to ensure the retardation, if not death, of genuine research into AI?