AI vs. Demographics

What You Need to Know

One of the key variables determining the strategic outlook is the degree to which artificial intelligence (AI) can deliver productivity gains. Past attempts to forecast productivity have been fraught with problems, so one needs to be humble in ascribing a significant increase in growth to AI. We review the latest attempt to quantify this.

Any boost from AI is not happening in a vacuum. How plausible is it that productivity gains from AI can offset the downward forces on growth from demographics and climate?

To what extent do large productivity gains from AI require mass job displacement? We suggest a preliminary model for this trade-off. This topic raises questions about the labor versus capital share and whether AI may have as large an impact on the distribution of gains as it does on total output.

Some have suggested that a shrinking workforce may not be a problem if AI is about to destroy jobs. However, this notion conflicts with the historical experience of automation. Moreover, the temporal and geographical distribution of these forces is very different.

The interaction of AI and demographics also has broader social implications: we see an evolution in the role of work and the extent to which it can give meaning. The more bullish AI forecasts imply a significant reduction in the workforce.

Our conclusions continue to support the case for US exceptionalism and an overweight to US equities, but not enough growth to shrink debt, hence we continue to believe the role of the dollar will decline.

Additional Contributors: Alla Harmsworth, Robertas Stancikas, and Maureen Hughes

In this paper, we focus on some of the key macro questions that the rapid advance of LLMs prompts. Does a large step up in productivity imply mass joblessness? Or is AI similar to previous rounds of technological advance and automation that ended up creating as many jobs as were destroyed? Is the extra economic growth enough to offset other significant downward forces on growth from a smaller working-age population and an increase in global temperatures by more than two degrees? Also, is this extra growth enough to fend off the need for governments to resort to inflation to deal with the buildup of debt? With 35 to 40 million Americans set to retire over the next decade and less immigration along with an unknown number of jobs likely to be displaced by AI, what is the net impact on the balance of power between labor and capital? It is this set of questions that guides our discussion.

The pairing of AI with demographics may, at first sight, appear odd. One is fast-moving and the other the slowest of economic forces. However, we see them as intimately related in a number of ways. While the hope is that AI might raise productivity, a shrinkage of the working-age population acts to slow growth rates. One of the principal fears of AI is that it leads to mass joblessness, but what does that mean if the working-age population is shrinking anyway? In meetings with investors, some have suggested that a shrinking working-age population may be a good thing if AI is destroying jobs, but is such thinking putting hope over evidence? AI might also change the type of jobs available and the skills required—how does that interact with an aging workforce?

Before we get onto the topic of how the benefits of AI are distributed or what AI means for job enhancement versus job displacement, there is a huge disagreement about what the total productivity gain from AI is likely to be. Daron Acemoglu, winner of the 2024 Nobel prize for economics, has made a high-profile forecast that the annualized increase in US gross domestic product (GDP) from AI will be 1.2% over the next decade, or an increase of just 0.1% percentage points per annum (pppa). A raft of recent papers has come up with a range of higher estimates (Display 1), with most of them forecasting an annual increase in GDP growth rates of 1.0 pppa. Brynjolfsson notably forecasts more than 2.5 pppa.