It’s AI’s world, we just live in it.

Or at least that’s the impression you’d get if you went by the increasingly frenzied, sometimes paranoid, headlines that’ve accompanied the meteoric rise of ChatGPT and related applications in 2023.

I’ve scaled up my own AI coverage commensurate with the overall decibel level, but also in an effort to keep myself apprised. Maybe it’s a fad, but it isn’t Bored Ape Yacht Club. Generative AI will draw you a Bored Ape, and assuming you don’t ask for too many of them, it’ll do it for free. Then it’ll write a new Shakespeare play. Once you’ve logged off for the day, it’ll plan Skynet while you sleep — or at least according to the worst fears of sundry luminaries who on Wednesday called for a moratorium on advanced AI development.

I think it’s safe to say the warnings voiced by Elon Musk (and, by way of endorsement, more than a thousand other very smart people) are overblown. Wholly implausible, even.

There’s still considerable disagreement around the concept of AI self-awareness, and it’s not immediately obvious why something which, if prompted, will turn Vermeer’s “Girl with a Pearl Earring” into Vermeer’s “Beaver with a Red Beret” would, unprompted, launch Russia’s nukes, for example. It’s still far more likely that Russia will launch Russia’s nukes. Just ask Alexander Lukashenko.

While a literal AI apocalypse will likely stay confined to increasingly unnecessary (if still mostly tolerable) Terminator sequels, a labor market AI apocalypse isn’t so far-fetched. As discussed here on Tuesday, Goldman estimates that 300 million full-time jobs globally could be vulnerable to replacement. In the same note, though, the bank suggested that over time, AI adoption will likely be a net win for everyone, much like many of history’s other epochal technological shifts.

One way widespread AI adoption could have a positive macroeconomic impact is through increased productivity. Indeed, Goldman posited “a labor productivity boom like those that followed the emergence of earlier general-purpose technologies like the electric motor and personal computer.”

Of course, such booms play out on “long and variable lags” (to employ the exhausted monetary policy cliché), which means we could be waiting around for quite a while. Generally speaking, the productivity booms attributed to the electric motor and advent of personal computing coincided with a 50% adoption rate by US businesses. That threshold was crossed around two decades after the original breakthrough, as illustrated in the figure on the left below.

The figure on the right illustrates the scope of the booms, which Goldman described as “quite substantial.”

Crucially, a world bedeviled by inflation could use a productivity boom. But if the cost is one-fifth of global full-time equivalent employment, and those jobs aren’t recouped, I’m not sure it’s worth it.

Goldman estimated a range for the productivity boost that could play out across the US economy in the event generative AI becomes pervasive. Their baseline estimate is consistent with the boost from “prior transformative technologies,” including electricity and personal computing.

But, as the bank’s Joseph Briggs and Devesh Kodnani wrote, it really depends on “the difficulty level of tasks generative AI can perform,” as well as adoption speed. In the event of a “much more powerful” AI, labor productivity could get a 3pp boost.

While Goldman looked at various combinations of AI power and labor displacement, it seems self-evident that the more powerful the AI, the more workers will be displaced, and the harder it’ll be to reemploy them, or at least in sectors and industries that’ve been disrupted.

Ultimately, Goldman said widespread AI adoption “could eventually drive a 7%, or almost $7 trillion, increase in annual global GDP over a 10-year period.”

In their letter this week, Musk and friends likewise suggested that if humans manage this revolution carefully, our species “can enjoy a flourishing future with AI.” If, on the other hand, we persist in what the same letter called “a dangerous race to ever-larger unpredictable black-box models with emergent capabilities,” we risk foregoing “a long AI summer” by “rush[ing] unprepared into a fall.”