Boosting economic growth in any sort of sustained way is really hard. The American public should always be skeptical of policy ideas promising to, say, add a full percentage point to real GDP growth, or anything close. First, a full percentage point is a lot for an economy that has grown at a 2.2 percent annual rate this century.
Second, such policies are hard to find. We spend a lot of time debating issues such as tax reform, greater federal R&D spending, and immigration reform, but from a growth perspective, smart policy may add only a tenth of a point here and there—and even that for a limited time.
Of course, when you are only growing at 2 percent a year, every bit counts, and all those cumulative tenths of a point may add up to something significant.
The difficulty of accelerating an advanced economy is one reason recent advances in artificial intelligence are so exciting. The private sector is developing and deploying a new general-purpose technology with the potential to boost productivity and economic growth significantly.
Many enthusiastic forecasts about AI dabble with moving the decimal point. Instead of 2 percent growth, 20 percent growth. A new working paper from the Anthropic Institute, “Economic Scenarios for Transformative AI,” offers just such a wild scenario, which the authors call “extreme change.” By 2030, AI has touched about 30 percent of all job tasks—think of a job as a bundle of different, discrete tasks—in the economy, or roughly half of everything cognitive workers are doing today. AI applications more than double output on those tasks, and in nine cases out of 10 the AI does the task rather than helping a human do it. It’s a world of 15 percent annual GDP growth and higher average wages but also one of 12 percent overall unemployment, including nearly 18 percent for knowledge workers.
(Keep in mind that this scary jobs scenario assumes, oddly to me, no new tasks will be created for the workers AI displaces—an extreme setting of a parameter versus the milder scenarios that assume one new task for every two that get automated.)
There’s also a midrange “substantial change” scenario, but let’s take a look at the most cautious outlook, the “modest change” scenario. Annual growth runs at 2.4 percent instead of 2 percent, with unemployment just a tick higher than otherwise at 3.9 percent. This really is AI as a “normal” technology.
Now, would that productivity boost justify all the current AI infrastructure spending? Maybe not, but the overall economic impact would hardly be small. There are few policies that would generate that much additional growth, nearly a half point a year, and imagine if we also did a slew of other pro-growth policies as well on taxes, federal science investment, trade, and immigration. The age of AI, or even a modest version of it, could create a foundation that better governance could build on.