“I think the point of the forecast isn't to get it right. It's nice if you do. The point of the forecast is to force the committee to make clear where they differ from one another, what they're watching, and why, and explain why they got it wrong.”
The Spillover · October 2026
Posen is answering the criticism that the Fed is a poor forecaster, and whether AI could do the job better. He says the economy is too complex for any model to simply get the prediction right, and that the forecast's real job is accountability.
Transcript
Is that some sense that you could be embarrassed or held to account if you did something that contradicted your previous statements? Whereas I think Chair Warsh would be perfectly comfortable saying, ha ha, you thought something I said at a Hoover Institute conference in 2025 binds me. It doesn't. You lose. I mean, you wouldn't quite say it the right way. When the
facts change, I change my mind. Change my mind. That would be the nice
way he'd say it. So I do think that if you take the moral hazard view seriously, and in my view, excessively seriously, you might deliberately try to indulge in that, make markets not be too assured that their little models actually predict. Again, goes with what Sebastian was saying earlier. I think in the end, that is still small potatoes, to use a technical term, versus the two big issues in what you said, which is: could the Fed get things right more often? And could the Fed be vulnerable to AI-led disruption? So, on the first, I guess there, as much as I believe AI is going to transform everything, certainly the think tank business we're all in. I do think the national economy is so complex and so large, and it goes to the hayeking, Hayek's critique of you can't know everything that the sort of ambition of the central planner or the Enlightenment project. That if I had all the equations for everything in the world, I could press a button and I'd know where everything was going, that it's really not ever going to turn out that way. And I would go further, and this again is my point of view and probably not widely shared. I think one of the things Chair Warsh and others has attacked is they say, well, the Fed's a lousy forecaster, or they're not even as good a forecaster as the markets. I don't think that's the point of the forecast. I think the point of the forecast isn't to get it right. It's nice if you do. The point of the forecast is to force the committee to make clear where they differ from one another, what they're watching, and why, and explain why they got it wrong. What's the new information or the new insight? That's to me is the point of the forecast. And so, again, AI may help with that, but I don't think it's not like, oh, it gets the prediction right and this all goes away. Right.
You know, my friends in AI Replace would seize on one thing you just said and flip it. And the thing that they would seize on is the notion that when you have a very complex system, you can't have equations. It's precisely where there is huge complexity, enormous amounts of data, a massive combinatorial space, something like protein folding or the game of go earlier on, or modeling all of text. This is actually where AI has the advantage. And it leads to a thought experiment, which I'm actually genuinely agnostic about at the moment. I'm thinking about it. Could you replace the FOMC with an AI?
Would you rather have a group of humans around a table or a model?
Allowing that both are flawed.
More from The Spillover
