Economics and reasoning with OpenAI o1
OpenAI ● Covered by 5 sources
OpenAI put economist Tyler Cowen in the room with o1, its reasoning-focused model, to poke at tough econ questions. He wanted to see if it actually thinks, not just recites textbook answers.
Based on reporting by OpenAI — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Tyler Cowen has spent decades needling economics students and pundits with questions that don't have clean answers. So it's a fitting move that OpenAI tapped him to stress-test o1, the company's model built to reason step by step rather than blurt out the first plausible-sounding response. The blog post frames this as a kind of oral exam: Cowen throws real economic puzzles at the model and watches how it works through them.
What makes o1 different from earlier chatbots, at least on paper, is that it pauses to lay out intermediate reasoning before committing to a final answer. Economics is a decent proving ground for that kind of behavior because so many questions in the field hinge on tradeoffs, hidden assumptions, and second-order effects that a model can't fake its way past with confident phrasing alone. Cowen's role here is less cheerleader and more examiner, pushing on the kind of nuance that separates a glib answer from one that actually holds up.
OpenAI doesn't frame this as proof that o1 has become an economist. It's closer to a demonstration, aimed at showing how a reasoning-oriented model handles a domain where correctness is fuzzy and context-dependent, unlike a math problem with one right answer. That distinction matters, because a lot of the AI benchmark chatter fixates on things you can score cleanly. Economics resists that.
The broader point buried in this exercise is about trust. If a model can walk a skeptical economist through its logic on a messy question and not just produce a tidy paragraph, that's a different kind of signal than acing a multiple-choice test. Whether o1 actually clears that bar is something Cowen gets to judge in the room, and the rest of us only get the highlight reel OpenAI chose to publish.
My take — AI-written commentary, not fact-checked reporting
Getting a well-known economist to vouch for your model's reasoning is smart marketing, but it's still marketing, and I'd rather see the full transcript than a curated post from the company that built the thing being graded. This is the same pattern I keep flagging: closed labs love showcasing 'reasoning' in domains too fuzzy to fact-check cleanly, and economics fits that bill perfectly.
Read more about this at: OpenAI