What does a company do when it suspects the world is about to blame it for something? It funds the research.
That’s the uncharitable read on Google’s September announcement that Nobel Laureate Philippe Aghion and Professor Ajay Agrawal are joining its AI & Economy team, with Anu Madgavkar and Daniel Rock coming aboard as new directors. The stated mission is to study AI’s economic impact, with the AI & Economy Research Program focused on global AI adoption and labor market shifts.
The charitable read is also available, and I’ll get to it. But I review AI tools for a living, and the single most common thing I see is a vendor producing its own evidence that its product works. So my reflex when a platform company hires the people who will write the authoritative account of that platform’s effect on jobs is not applause. It’s a raised eyebrow.
Why this hire is not nothing
Let’s be fair about the caliber here. Aghion’s Nobel is not a participation trophy. Agrawal has spent years on the economics of prediction machines, which is the closest thing the field has to a usable framework for what AI actually does to a business. Rock has done the quantitative work on which occupations are exposed to language models. Madgavkar has spent a career on labor market analysis at scale.
This is not a comms hire. If Google wanted friendly numbers, there are far cheaper ways to get them. You commission a white paper from a consultancy, slap a nice chart on it, and ship it with the product launch. You do not recruit a Nobel Laureate, because Nobel Laureates have the one asset that cannot be bought back once spent: a reputation that outlives the employer.
The question nobody in the announcement answers
Here is what I want to know, and what the announcement does not say: who controls publication?
That single detail separates a real research program from an expensive marketing department. If these four can publish a finding that Google’s own products are displacing workers in a specific sector, and publish it without a legal review that softens every sentence, then the program has teeth. If they can’t, then we are looking at the world’s most credentialed content studio.
I have no evidence either way. That’s my point. The press release tells us who was hired and what they’ll study. It does not tell us the governance, the data access, the funding terms, or what happens when the conclusions are inconvenient. Those are the parts that determine whether any of this research is worth citing, and they’re the parts left out.
What this tells you about where AI is actually heading
Set aside the credibility question for a second, because the hire itself is a signal, and it’s a louder one than most product launches.
Companies staff up where they expect pressure. A platform that thought AI’s labor effects were going to be a minor footnote does not build an economics bench. The implicit forecast baked into this announcement is that the labor market conversation is about to become the central AI conversation, and that having your own economists in the room will matter when regulators start asking questions.
For those of us evaluating tools, that’s useful information. It suggests the era of pure capability marketing is closing. The next few years of AI discourse will be less about benchmark scores and more about who gained and who lost, measured in employment data rather than demo videos.
How I’d judge it in twelve months
I’m not writing this off. I’m setting conditions. Three things would move me from skeptical to convinced:
- Findings that hurt. At least one published result that is genuinely bad news for Google’s products or for the adoption story the company sells. If everything that comes out is directionally flattering, that tells you what kind of program it is.
- Data other people can check. Research on global AI adoption is only as good as its methodology being visible. Proprietary internal data with no reproducibility path is an anecdote wearing a lab coat.
- Public disagreement among the four. Real research teams argue in public. A unified house position on something as contested as AI’s labor effects would be the tell.
Independent academics studying this problem now have a well-funded competitor on their turf, one with access to adoption data most of them will never see. That’s a mixed blessing. Better data can produce better answers. Concentrated data can also produce answers nobody can audit.
Google hired four serious people to study a question Google has a direct financial stake in. That tension isn’t disqualifying, but it doesn’t disappear because the names are impressive. Read the output, check the footnotes, and notice what never gets published.
🕒 Published: