Friday, October 31, 2014

Artificial Intelligence and Artificial Stupidity

Tesla CEO and famous technology innovator Elon Musk has repeatedly warned about AI threats. In June, he said on CNBC that he had invested in AI research because “I like to just keep an eye on what's going on with artificial intelligence. I think there is a potential dangerous outcome there.” He went on to invoke The Terminator. In August, he tweeted that “We need to be super careful with AI. Potentially more dangerous than nukes.” And at a recent MIT symposium, Musk dubbed AI an “existential threat” to the human race and a “demon” that foolish scientists and technologists are “summoning.” Musk likened the idea of control over such a force to the delusions of “guy[s] with a pentagram and holy water” who are sure they can control a supernatural force—until it devours them......Adam Elkus in Slate.

Elkus devotes his article to a lame brained attack on Musk's fears. I say lame because Elkus fails to engage Musk on substance or even seem to grasp the nature of the perceived threat. Mostly he just worries that Musk's expressions of concern will hurt funding for AI research - Doh! - and mumbles incoherently about Skynet. He also, and quite absurdly, accuses Musk of being technopathic - a disorder apparently characterized by an unjustified fear of technological change. His article concludes with the following feel-good drivel.

If Musk redirected his energies and helped us all learn how to understand and control intelligent systems, he could ensure that the technological future is one that everyone feels comfortable debating and weighing in on. A man of his creative-engineering talents and business instincts surely could help ensure that we get a Skynet for the John and Sarah Connors of the world, not just the Bay Area tech elites.

By focusing on the science fiction threat of AI, and failing to engage even that, Elkus ignores all the real threats that AI poses. One such threat is taking over many jobs formerly done by humans. It's sometimes assumed that this only happens when the computer gets as smart at its job as the human it replaces. That's far from the case, as anyone who has dealt with phone menu hell knows by personal experience. For the employer, very minimal performance may be perfectly acceptable if the price is right. The phone menu replaces reasonably competent humans who can occasionally exhibit common sense with minimally competent robots with zero common sense, producing a sort of artificial stupidity.

Even Skynet was an example of this kind of artificial stupidity. It's job was to protect the environment of Earth, presumably for the benefit of its inhabitants, but it apparently failed to understand the "presumably" part, but did notice that the big threat to the environment was those self-same inhabitants.

My internet went out for a few days recently, perhaps as a result of a much shorter lived neighborhood power failure. Patient navigation of a phone menu hell - if you can call cursing and snarling, "patient" that is - finally took me to a message that said that the provider knew my service was out, and that their technicians were working on it with no prognosis for a fix.

Of course no internet is a major catastrophe these days. It turns out that my family no longer owns a physical phone book, for example. Fortunately I have this "smart phone" right? Well it seems that I didn't understand my new, one day old, smartphone very well, since it proved equally inept at reaching the internet. It took me perhaps a day to realize that my phone, in its AI wisdom/stupidity passed all calls to the internet thru my still working home WiFi. It was smart enough to realize that this could save me data/money but not smart enough to figure out that this was a bad strategy when my WiFi modem could not connect to the internet..

These kinds artificial stupidity are usually just a minor nuisance, Skynet not withstanding, but they illustrate some of the folly of depending, Elkus fashion, on all of us learning "how to understand and control intelligent systems." Our brains aren't designed for that. A substantial part of my career was spent working with AI systems, and trying to apply them to solve real world problems, and I failed the simple test provided by my "smart" phone. Even a genius like Musk knows that he has problems with controlling AI systems.