Harnessing Artificial Intelligence

I’ve always loved movie robots, even the bad ones. But as the machines around us become unnervingly smarter, it’s hard not to worry that artificial intelligence with malign intent—some version of the Terminator or HAL—will eventually be unleashed on the real world. I’m not alone in this bleak thought, of course. Stephen Hawking warned that “the development of full artificial intelligence could spell the end of the human race.” And when the Oxford philosopher Toby Ord examined a host of existential threats—asteroids, nuclear war, climate change—for his bracing book The Precipice, he ranked “unaligned artificial intelligence” the most probable of all. Plunging into the latest crop of AI books, I was somewhat comforted by experts’ consensus that “the singularity”—the point at which AI surpasses human intelligence—is not imminent; guesses about the timing range from a few decades to centuries from now. But we shouldn’t relax just yet. These authors contend that anxiety about a future of robot overlords is distracting us from the immediate danger of “narrow” AI, which powers everything from smart speakers to self-driving cars. To some, this AI seems clever but benign—a digital idiot savant that tackles tasks such as getting from point A to point B without hitting anything. But as Michael Kanaan writes in T-Minus AI, it can be coldly efficient at bad stuff too. As the U.S. Air Force’s cochair of artificial intelligence, Kanaan is keenly aware that some nations aggressively deploy AI to advance antidemocratic agendas. The Chinese government, for example, is building an AI-powered “social credit system” that tracks online and real-world activity to calculate individuals’ “social trustworthiness” and reward or punish people accordingly. Loitering, littering, and spreading “unacceptable” information can lower one’s score—and make it harder to buy a plane ticket, get a loan, or qualify for a job. Darker still, China now uses facial recognition to identify members of the mostly Muslim Uighur minority and round them up for detention and “reeducation.” (Of course, Western companies and governments aren’t beneath bending democratic ideals to monitor consumers and citizens either.) Kanaan is therefore calling for an urgent conversation about the implications of unconstrained AI and how we can put up guardrails to preserve our liberty and security. William Davidow and Michael Malone’s The Autonomous Revolution and Daniel Susskind’s A World Without Work are similarly direct about the risks posed by existing AI. They argue that its rise is not just another technological transformation that, like the Industrial Revolution, will create a wealth of new jobs even as it eliminates others. Rather, they suggest, we’re at a historic inflection point, confronted with a unique, intelligent technology that is outperforming us on an expanding array of tasks and could eventually replace much of human work. Davidow and Malone, a tech industry executive and a journalist-entrepreneur, say that instead of generating productivity improvements that increase GDP and prosperity (as previous advances have), AI may boost productivity while driving down GDP. Among the potential results: an abundance of cheaper stuff, yes, but also declining wages, growing economic inequality, less work all around, and the emergence of “millions of ‘ZEVs’—people of Zero Economic Value—individuals you would not hire even if they worked for free.” Susskind, an economist and a former policy adviser to the British government, likewise sees us humans retreating to an ever-shrinking set of activities as AI encroaches. Truck drivers plus GPS are a powerful combination; but when AI becomes both pilot and navigator, 3.5 million U.S. truck drivers will need to find new work. Even jobs that depend on our most human traits—empathy, judgment, and creativity—are probably not immune. Machines may never have a doctor’s “gut feeling” about a diagnosis, or follow their muse as a composer might, but they will surely learn to mimic those human acts so well that we can’t tell their imitation from the real thing. These authors do see a way forward in a world with less work and growing inequality, though. They envision new regulatory bodies (Susskind proposes a Political Power Oversight Authority, for instance, to rein in companies with too much clout) and a heavier hand from government. Among Davidow and Malone’s proposals, should free-market solutions fall short, are (brace yourself) “systems for redistributing wealth, such as higher taxes, free universal health care, and universal basic income.” Having become convinced that no job is safe, I was eager to dig into Kevin Scott’s upbeat Reprogramming the American Dream. Scott, Microsoft’s chief technology officer, expects an explosion of new jobs for rural workers, particularly in manufacturing. “AI will ultimately be about human empowerment, not displacement,” he assures us. Indeed, he describes the many types of blue-collar jobs that AI, robotics, and automation are creating. One entrepreneur, for instance, launched a computer-driven precision plastics fabrication business in which someone with “a machinist diploma…and a little on-the-job training can land a well-paying job in a small town that was once counted out.” Scott also points, without apparent irony, to a whole new class of jobs supporting AI itself: data labeling to train AI, human-run temp agencies for robots, and construction jobs building giant data centers like Microsoft’s 500-acre facility in Virginia. Scott’s is a hopeful view that largely avoids the elephant in the room—whether these new jobs can fully replace the ones that are lost. These authors may diverge on the particulars, but they share a qualified optimism about what’s to come. The key, they largely agree, is to establish rules and systems now to avert the worst outcomes in the future. As Kanaan puts it, we need to implement AI “only in ways consistent with fundamental human dignities…and only for purposes consistent with democratic ideals, liberties, and laws.” What do you suppose an artificial intelligence would think of this plan? When Kanaan asked a version of the machine-learning application GPT-2 to riff on the idea, it instantly delivered a pitch-perfect response: “Our job is now to convince the public in particular that using AI to achieve these aims is a necessary and desirable part of our society…. But in the end, the future demands we make moral decisions as we begin to build a world that is truly safe and sustainable, one where humans and AI can truly coexist together.” Although the machine doesn’t know in the human sense what it has written, one can’t help perceiving at least a facsimile of self-interest—and a clue about what the future holds.

Harnessing Artificial Intelligence

Partner Center