Algorithmic Monoculture: Why Hiring Tools Reproduce Bias

Claire Monroe

Welcome to the show everybody! I'm Claire Monroe, here with Edwin Carrington. Edwin, I want to start with a phrase I can't stop thinking about: algorithmic monoculture. It comes from a 2022 paper out of Stanford by Rishi Bommasani, Percy Liang and their colleagues, titled Picking on the Same Person. Their argument is that when lots of organizations rely on the same model, or models trained on the same data, the same individuals end up rejected by all of them. They call it outcome homogenization.

Edwin Carrington

It's an important idea, Claire, and it changes how we should think about bias in hiring. For most of history, a biased hiring manager was a local problem. He was wrong, and he cost a good candidate one job. That candidate walked across the street, met a different manager with different blind spots, and got hired. The errors were scattered. Scattered errors are survivable.

Claire Monroe

Because every door had a different lock.

Edwin Carrington

Exactly. Now imagine every door has the same lock. If a thousand companies screen through the same tool, and that tool has learned to undervalue a certain kind of résumé, that person isn't rejected once. They're rejected a thousand times, by the same logic, in the same week. And no single employer ever sees the pattern, because each of them only sees their own funnel.

Claire Monroe

And the Stanford team found something even more uncomfortable. When models share training data, the effect is strongest for specific individuals, even more than for whole demographic groups. So it's not only that a group is disadvantaged on average. It's that particular people get locked out of the entire market.

Edwin Carrington

That's the part leaders miss. They ask, is this tool accurate? That's the wrong question on its own. The better question is, whose mistakes am I inheriting, and how many other companies are inheriting the same ones?

Claire Monroe

There's a second paper that backs this up. Jon Kleinberg and Manish Raghavan at Cornell published a study in the Proceedings of the National Academy of Sciences in 2021. Their finding was that even when an algorithm is more accurate than any individual human evaluator, having everyone use the same one can make overall hiring decisions worse.

Edwin Carrington

More accurate for each company, worse for the whole system. That is the monoculture in one sentence. Independence of judgment is a hidden asset, and most organizations are trading it away without noticing.

Claire Monroe

So let's talk about how these tools learn to pick on the same person in the first place. The study that stopped me in my tracks came from the University of Washington in October 2024. Kyra Wilson and Aylin Caliskan tested three large language models on more than five hundred real résumés against more than five hundred real job descriptions. Over three million comparisons in total. The models favored white-associated names eighty-five percent of the time. Female-associated names were favored only eleven percent of the time. And the models never favored Black male-associated names over white male-associated names. Never, Edwin. Not once.

Edwin Carrington

Eighty-five percent. And notice what changed between those résumés. Not the experience. Not the skills. The name. That tells you the model isn't evaluating the work. It's evaluating a proxy.

Claire Monroe

And this isn't new. Back in 2018, Reuters reported that Amazon scrapped an experimental recruiting tool because it had taught itself to downgrade résumés that included the word women's, as in women's chess club captain. It learned from ten years of résumés that came mostly from men.

Edwin Carrington

That's the mechanism, Claire, and it's worth saying plainly. A screening model trained on your past hires learns who you hired. It does not learn who performed. Those are two very different datasets. If your history carries a bias, the model doesn't correct it. It automates it, at a speed and volume no human recruiter could ever match.

Claire Monroe

So the bias gets in through the training data, and then it gets amplified through consistency.

Edwin Carrington

And then it spreads through concentration. When the market converges on a handful of vendors, one vendor's blind spot becomes an industry's blind spot. The résumé filter that quietly penalizes a gap year or a particular zip code isn't one company's problem anymore. It's a gate across an entire profession.

Claire Monroe

Which is scary, because most companies buy these tools to reduce bias. That's the pitch.

Edwin Carrington

The pitch assumes the tool is neutral. No tool is neutral. Every filter encodes a definition of what good looks like. The only question is whose definition, and whether anyone has checked it against actual performance.

Claire Monroe

Okay, so what do leaders actually do? Because I don't think the answer is to throw out structure and go back to gut feel.

Edwin Carrington

No, gut feel is its own monoculture, the monoculture of one person's instincts. The answer is to change what you measure and where you measure it. Two moves. First, measure against your own standard, not the market's. Second, measure early, before the résumé screen has already narrowed the pool on proxy signals.

Claire Monroe

Say more about your own standard.

Edwin Carrington

Most screening tools compare candidates to a generic picture of a good hire, built from someone else's data. OAD works the other way. It measures behavioral drive, meaning how someone is wired to operate under role-specific pressure, and compares candidates against a Top Performer Benchmark built from your own best performer. Every candidate is scored on the same five dimensions: Drive, Influence, Resilience, Structure and Leadership. So your filter reflects what success looks like inside your organization, in that role. That's the opposite of a monoculture.

Claire Monroe

Because my top performer isn't my competitor's top performer.

Edwin Carrington

Precisely. And the benchmark measures how people operate, not what their name is or which school is on the résumé. OAD has been EEOC compliant since 2001, with no adverse impact on any protected class, and zero successful legal challenges in thirty-eight years. But I want to be careful here, Claire. It is built as one structured input in the hiring decision. Never the sole basis for a hire. A human stays accountable for the decision.

Claire Monroe

I like that, because it gives leaders a practical checklist for any vendor, not just OAD. What data was this trained on? Who else in my industry uses it? Can I see pass-through rates by demographic group at every stage of my funnel? And can you explain why a candidate was rejected?

Edwin Carrington

Those four questions will tell you more than any sales demo. And if a vendor can't answer them, you have your answer. You would never let one supplier set the price of every input in your business. Don't let one algorithm set the definition of talent for your whole industry.

Claire Monroe

So if you want to see what your own standard looks like, start with the person you already trust most on your team. OAD's free Top Performer Profile takes seven minutes, you get results in twenty-four hours, and there's no credit card. It's a clear look at what a validated, structured input looks like next to whatever you're running today. That's O-A-D dot A-I.

Edwin Carrington

The best companies have never won by thinking like everyone else, Claire. Hiring is no different. Build a filter that is yours.

Claire Monroe

Perfect place to leave it. Thanks for the wisdom, Edwin.

Edwin Carrington

Always a pleasure, Claire.