Pacing paradox: Has AI become death, destroyer of worlds?

BRET STEPHENS

Creators who anguish over the consequences of their creations are nothing new. Alfred Nobel and dynamite, J Robert Oppenheimer and the atomic bomb, even Tim Berners-Lee and the World Wide Web: Regrets, they had more than a few.

Now it’s the turn of the parents and nursemaids of the artificial intelligence revolution. Last week, Jacob Coxon, a 27-year-old researcher at Anthropic, quit his job with a warning on social media that the leading AI companies were “gambling with our lives” and that “the people building AI earnestly believe it could kill us all by the end of the decade.”

That was followed by a lengthy open letter from Dario Amodei, Anthropic’s CEO, in which he warned that within six months to a year, a swarm of AI agents, acting autonomously and improving continually, “could be capable of taking over the entire internet,” the potential cost being measured in “hundreds of billions of dollars in damage.” He called on AI companies to slow down to “align and safeguard their models,” as well as for our country to try to reach agreements with China that could mitigate the more acute risks connected to AI.

The letter was endorsed by Elon Musk and OpenAI’s Sam Altman, among others. Then our modest and self-effacing president weighed in.

“The only control or ‘guardrails’ that AI needs is a STRONG AND SMART (High IQ!) PRESIDENT, and the U.S.A. has that, in spades!” Donald Trump wrote on social media, before taking a shot at Amodei for “pretending to be a ‘perfect little angel.’” The president thinks the safety concerns are a “hoax” and that AI regulation should be light, lest China steal a march on America’s slender AI lead. Agreeing with him is Nvidia’s Jensen Huang, who dismissed Coxon’s warnings as “not grounded in science.”

At the bottom of the argument are two large unanswered questions. First, is AI principally a disrupter or a destroyer? Second, if it’s a destroyer, just what kind is it?

Until last week, I suspect most people, including AI sceptics like me, saw it as a technology that looked a lot like previous technologies of mass disruption: steam engines, aeroplanes, cars, PCs and so on. If that’s the right paradigm, then the right course is to discount the fears and seize the economic opportunity. For all the anxiety about AI causing near-universal unemployment, nothing like that has come to pass. “There is no evidence that AI adoption has negatively impacted firms’ job postings,” notes a recent paper from Stanford’s Institute for Economic Policy Research. “Among firms that adopted enterprise AI, employment grew by 10% in the two years following adoption.”

That may change in a few years, but it defies centuries of economic history. Technologies that save labour and enhance productivity transform the nature of employment while tending to increase its stock. The lighter the hand of government, the faster companies can innovate, adapt and create value as well as new jobs.

But what if the disruption paradigm is, if not wrong, then dangerously inadequate? What if we aren’t dealing with the digital equivalent of the Model T but something more akin to splitting the atom or the research that, many suspect, helped create a highly infectious virus that may have leaked from a Chinese lab?

The last two presidents to deal with similar dilemmas were Barack Obama, who in 2014 imposed a funding pause on certain types of gain-of-function research so that the risks could be better understood, and Harry Truman, who in 1946 created the Atomic Energy Commission that transferred control of the country’s nascent nuclear complex from military to civilian hands.

Both analogies are instructive but imperfect. Obama’s instincts were surely right to take a harder look at dangerous research. But some of the action shifted to China, under conditions that may have been even less safe.

Where does that leave us?

When slowing down isn’t a realistic possibility, sometimes the right answer is to speed up — if only to accelerate our understanding of the means by which we can use AI against the risks we create through it, or that it creates by itself. One consolation is that this isn’t the first time great scientific minds have warned that they’ve created the instruments of humanity’s destruction.

Less consoling is that past reprieves are no guarantee of future salvation.