OpenAI Wants PhDs to Prove Its Models Don't Wreck the Economy

OpenAI is finally getting serious about a question that should have been asked louder years ago: what do models like GPT-3 and Codex actually do to people’s livelihoods? The company put out a call for PhD-level researchers to study the economic impacts of its large language models, framing it as essential for both smarter deployment decisions and shaping public policy. This isn’t just academic curiosity. They’re openly acknowledging that the benefits of AI, like the broader technological gains of the last 40 years, could easily pool at the top unless someone actively measures and mitigates the damage.

The research agenda they’ve outlined is pragmatic, targeting three specific levers: AI deployment policies, system design choices, and real-world public policy interventions. OpenAI is dangling access to its latest releases—Codex, DALL-E 2, and ChatGPT—as tractable case studies. It’s a smart move, essentially admitting that in-house research won’t cut it and that external, rigorous measurement is non-negotiable. They’re even trying to build a bridge between researchers and the companies already building products on their APIs, aiming to get data directly from the front lines where workers and consumers interact with these tools.

There’s a clear tension in their pitch. While they cite evidence suggesting no immediate risk of widespread technological unemployment, they immediately follow with the grim reality that unemployment shocks—even temporary ones—wreak havoc on individual wellbeing, and that inequality amplifies societal fractures. It reads like a company trying to get ahead of a regulation curve they see barreling toward them. The call isn’t just for papers; they’re offering API credits, red teaming invites, and tightly tailored partnerships, treating this economic analysis with the same urgency as a safety evaluation.

For a firm racing to build artificial general intelligence, this move feels less like charity and more like an insurance policy. My skepticism is reserved for the execution. Connecting researchers with firms that have a vested interest in a positive narrative could get messy. The results will only be as good as the access and the independence granted. If OpenAI genuinely lets researchers follow the data wherever it leads—even if it points to uncomfortable conclusions about its own products—this could establish a vital feedback loop. If not, it’s just another well-funded whitewash.