All News
federal-reservekevin-warshai-economicsmonetary-policyproductivitysuper-intelligencelabor-market

Warsh put three AI futures in front of the Fed. None of them came with a number

The Fed chairman's first big AI speech sketched three scenarios for the US economy and put no numbers on any. What the bank would actually do is unstated.

Vlad MakarovVlad Makarovreviewed and published
7 min read
Warsh put three AI futures in front of the Fed. None of them came with a number

Kevin Warsh's first substantial speech about artificial intelligence, delivered on 16 November to a Federal Reserve conference in San Francisco, contains three scenarios for the American economy and not a single number for any of them. No growth rate. No unemployment rate. No probability on any branch, no timeline for when a branch might arrive, no threshold at which the central bank would move. The text, published as Opening Remarks at noon Eastern, is remarkable for how openly it names its own uncertainty and how completely it declines to size it. A central bank's working currency is magnitude, and the whole speech declines to issue any.

Three futures, described in adjectives

The scenarios come in the speech's own order. In the first, super intelligence, which Warsh abbreviates SI, "leads to an increase in productivity and economic gains that are shared broadly across labor and the owners of capital". Real GDP growth rises persistently above its long-run average, and the United States stays close to full employment. That is the good outcome, and it is described entirely in direction: up, broadly, persistently.

In the second, the same technology produces substantial labor market disruption, "as some jobs are displaced, rather than complemented, by technology". Growth still rises above the long-term average. But widespread automation, working through ordinary labor market frictions, could push the jobless rate above levels consistent with full employment. Read that sentence twice, because it is the one a monetary policy committee would have to vote on: output accelerating while employment deteriorates is the precise configuration in which the Fed's two mandates point in opposite directions. The speech offers no view on which one wins.

The third branch is four words long and gets the shortest treatment. "Or SI could disappoint." After heavy capital investment and a stock market that has already paid for the promise, "expected returns could turn down and equity valuations could fall, leading to an economic slowdown and a more notable rise in unemployment. In time, the economy would recover, but without the productivity gains."

Then the caveats, which are unusually explicit. The Board's staff "has begun to evaluate various contingencies, applying stylized, model-based frameworks to generate scenarios". And: "These are illustrative scenarios, not forecasts. Moreover, many other outcomes—probably including one that will prove most accurate—could be drawn." Warsh is telling the audience, plainly, that the list is not exhaustive and that the true path is probably not on it. That is intellectual honesty. It is also, from the institution that sets the price of money, a striking thing to publish.

The third path is a market argument already in progress

Of the three, only the disappointment scenario has a live, public controversy attached to it, and it is not a monetary policy controversy. It is the argument about whether the revenue financing the AI buildout is real. The circular-financing case made in the Wall Street Journal is that vendor investments, customer commitments and capacity deals have become sufficiently interlocked that reported demand is partly a function of the capital being raised to serve it. If that reading is right, the correction does not arrive as a productivity surprise. It arrives through valuations, then through hiring, then through spending.

The speech does not adjudicate that debate, and it does not have to. What it does do, almost in passing, is undercut the assumption that the first scenario is the default. Warsh notes that "by some estimates, research efforts have increased more than 20 times since the 1930s", and then quotes Nicholas Bloom, Chad Jones and their coauthors: "Research productivity has fallen by an even larger amount." The case for super intelligence supercharging growth rests on it doing something the internal combustion engine and electricity "did not themselves automate": the creation of ideas. Warsh says SI "could supercharge idea generation", changing "the method—and the speed—of innovation". Could. The word is doing a lot of work in a paragraph that also cites a ninety-year decline in the productivity of research itself.

What the Fed would do, which the speech does not say

Here is the gap that a reader with a rate decision to anticipate will notice first. Three scenarios, each with a different implication for policy, and no mapping from scenario to instrument.

If growth runs persistently above trend with full employment, the question is whether a central bank treats that as an inflation risk to lean against or as a supply expansion to look through. It is the difference between tightening into a productivity boom and letting it run. If displacement dominates and unemployment climbs while output rises, the Fed faces a genuine conflict and has to choose which mandate binds, which is a judgment the Federal Open Market Committee has historically made in private and explained afterwards. If the investment cycle breaks, the problem becomes a demand shock transmitted through asset prices, which is the most familiar of the three and the one the existing toolkit handles best.

Warsh addresses none of this directly. He says the widespread adoption of the technology "will have significant impact on the economy, with important implications for monetary policy down the road", that "the big questions are about magnitude, scope, and timing", and that the Fed's role is "neither advocacy nor opposition". All three statements are defensible. Together they amount to a promise to think about it later.

The one concrete passage is a warning: "SI empowers all, including bad actors, so we are thinking especially of critical infrastructures and their potential vulnerabilities." That is a financial stability sentence placed near the end of an economics speech, and it may be the most operational line in the document.

A phrase with a political owner

There is a second layer, and Reuters reported it without editorialising. Warsh uses "super intelligence", which the wire described as "the language preferred by President Donald Trump". The chairman also volunteered that he is "not exactly impartial about American leadership", and asked, rhetorically, "can you be a superpower without super intelligence?"

Both things can be true at once: a central banker can hold a view about national competitiveness and still run an institution that is neither advocate nor opponent of a technology. But the two claims pull in different directions, and the speech does not reconcile them. The setting sharpened the effect. The San Francisco event was invitation-only, attended by Governor Lisa Cook, venture capitalists, money managers, philanthropists and chief executives, and Reuters found that "much of the conversation remained speculative". Panelists debated whether AI could let the United States outpace other economies and worried that students might grow too dependent on it. Three days earlier, the same conference had been framed around an American public that has grown wary of the industry. A room of builders, told by a regulator that he is rooting for the country, is not a room primed for scepticism.

What would settle it

One document would move this from rhetoric to economics: the staff scenario work itself, published with its parameters. The Fed already releases Summary of Economic Projections with ranges and a Tealbook with explicit alternatives. A scenario exercise that names three states of the world and publishes none of the assumptions behind them cannot be checked, replicated or argued with, which means it cannot yet inform anyone's expectations about rates.

Or, failing that, a staff paper. Warsh's closing line is that "our job is to prepare for whatever comes next". Preparation that cannot be measured is difficult to distinguish from observation, and a central bank that has correctly identified the shape of a shock but not its size has done the easy half. The hard half, the half that shows up in the dot plot and the statement, is still unwritten.

Related Articles

Scroll down

to load the next article