Posts
Stanford exposed bias in AI hiring tools. But we’re still arguing about the wrong end of the problem.
See other posts from by Bryan Adams
CEO
Stanford’s latest AI hiring research exposed a serious bias problem. But it also points to a deeper issue: the industry has spent three years building technology around rejection, not trust.
The algorithmic monoculture is real. The more uncomfortable finding is that the industry has spent three years building a machine that is brilliant at separating average candidates from unqualified ones, and structurally blind to the great ones it was supposed to find all along.
By now you have probably seen the Stanford study, or at least the headline version of it. The largest independent analysis of AI hiring ever conducted, four million applications across 156 large employers, and a finding that lands exactly where you would fear it would. When many companies lean on the same screening algorithm, the same people get rejected repeatedly, and certain groups carry far more of those rejections than chance could ever explain. The researchers call it algorithmic monoculture, and it is a genuinely important piece of work that everyone in talent acquisition should read.
But I want to talk about the part of it that almost nobody is talking about, because the bias finding, as serious as it is, has crowded out a quieter and in some ways more uncomfortable question. Not "is the machine fair", but "what did we build the machine to do in the first place, and why on earth did we point it there?"
We pointed nearly all of it at filtering. And I think filtering, as the central obsession of modern hiring, is a far newer and far stranger habit than we like to admit.
The obsession is about three years old.
I wrote recently about why your best candidates have quietly stopped applying, and how a rising application count is the very thing hiding their absence from you. The mechanism, borrowed from a 1970 paper on used cars, was that once effort becomes unreadable, the sellers of genuine quality leave the market first. I stand by every word of it. But sitting underneath that argument there was an even simpler observation that I did not press hard enough at the time, and it is this.
Nobody was obsessed with filtering five years ago.
The fixation is recent, and it is a direct reflex to volume. When LinkedIn started clocking eleven thousand applications a minute and AI made a tailored CV free, the panic that followed was real and understandable, and the whole industry dropped into the same defensive crouch at almost the same moment. We bought screeners. We bought fraud detection. We bought faster ways to reject at greater scale, and we told ourselves we were getting more sophisticated about talent when the truth is we were just getting more frightened of volume.
And in the hurry, we made a category error that has cost us more than any bias headline ever will. We took a volume problem, treated it as a quality-detection problem, and then built the wrong instrument to solve it.
What a filter is
Here is the thing about a filter that we conveniently forget while we are busy buying them.
A filter is a threshold instrument. Its entire job is to find a line and sort the world into what falls below it and what clears it. That makes a filter genuinely excellent at one task, and one task only, which is separating the unqualified from the adequate, the below-the-bar from the at-the-bar. It is a median-detection machine, calibrated by design to the average.
Now ask that machine to find greatness, and watch it fail, because greatness does not live at the threshold. It lives well above it, in territory where a pass-or-fail gate has nothing useful left to say, since everyone above the line reads to the filter as simply "passed". Worse still, your most exceptional candidates are very often the ones who do not match the template at all. The career changer. The unconventional path. The person whose best qualification is a way of thinking that no job description has ever known how to ask for. The filter does not see those people as great. It sees them as a mismatch, and it removes them with the same quiet efficiency it uses on everyone else.
So, look honestly at what we have built. We have spent three years and a remarkable amount of money constructing a machine that is superb at telling average candidates apart from unqualified ones, and we have been calling that talent acquisition. It is not. It is rejection, optimized. And it was never, at any point, capable of finding the people we most wanted, because finding the great was never something a filter could do.
You cannot do this from a crouch
There is a second cost, and no amount of technology will fix it, because it sits in the posture rather than the tooling.
Every screening-first system is built on a premise of suspicion. The candidate is a risk to be verified, a claim to be doubted, an artefact that might be fake. I understand precisely how we got here, and I am not pretending the fraud, and the AI-generated noise are imaginary. But I would ask you to sit with one question, because I genuinely cannot find a conclusive answer to it.
How do you offer a wonderful experience to your best possible candidate from a starting position of cynicism that was designed to catch out your worst one?
You cannot. Experience is not a layer you bolt on at the end. It is the downstream expression of a mindset, and if the mindset is suspicion, the experience will smell of it no matter how much you spend making the careers site beautiful. The excellent candidate, the one with three other options and no patience for being processed, can sense a defensive crouch from the very first click, and they do the most rational thing available to them, which is to take their attention somewhere it is welcome.
The challenge we have refused to name
Hiring is not one problem, and I would never insult your intelligence by pretending it is. It is a tangle of them: sourcing, screening, filtering, personalizing, converting, closing, all of it in pursuit of the best talent a market will give you. But inside that tangle, the work of finding great people holds two challenges we obsess over, and one opportunity we have barely touched.
The two challenges are the ones you already fund. Filter out the people who genuinely cannot do the job. Persuade and convert the people who can. Fine.
The opportunity, sitting in plain sight with almost nothing invested against it, is this. To trust and empower excellent candidates to make a smart decision for themselves.
Think about who your best candidate is. They are more self-aware than your model, better informed about their own ambitions than any assessment, and more honest with themselves, when handed the truth to work with, than your funnel has ever permitted them to be. They are, in other words, the single most accurate and lowest-cost judge of their own fit you will ever have access to. And we have built an entire apparatus whose purpose is to overrule them before they get a word in.
The proposition
So here is the idea I want to leave you with, and it is honestly a different bet from the one the industry is making.
Stop treating the candidate's judgment as something to be replaced and start treating it as what should be recruited.
Give your best people the truth about the work. The real texture of the role, the conditions, the trade-offs, the kind of person who thrives in it and the kind who quietly cracks. Then hand them the wheel. Let them match themselves against that reality in their own time, in private, with no process performed at them, and let them decide whether they are right for you. Let them own that conclusion and let them choose whether it travels with their application or stays theirs entirely.
This is the whole reason we built Happydance the way we did, with the matching living on the careers site and in the candidate's hands rather than the recruiter's, upstream of the application rather than after it. Not as another gate, but as an act of trust that asks the candidate to invest real effort precisely because the effort is finally being spent on something that represents them. It is the practical edge of the argument behind Give & Get, and of the book I am writing now, Sell The Truth. The companies who tell the truth and trust people with it will out-hire the companies who sell a fantasy and then police the response to it.
The great ones were never going to clear your bar to prove themselves to you. They were always going to decide whether you were worth their time. A filter cannot win that decision, because a filter was built to answer a completely different question.
Your screening machine has never once found you a great candidate. It has only ever removed the people the great ones had already stopped competing with.
The opportunity almost nobody is funding is the simplest one of all. Stop trying to catch the worst and start earning the trust of the best. They are not hard to find; they are just waiting to be trusted.
“Stop treating the candidate's judgment as the thing to be replaced and start treating it as the thing to be recruited.”
FAQs
What is algorithmic monoculture in AI hiring?
Algorithmic monoculture happens when many employers rely on similar or identical screening systems, causing the same candidates to be rejected repeatedly across different hiring processes.
What did the Stanford AI hiring study reveal?
The study highlighted the risk of bias in AI hiring tools, especially when large numbers of employers use similar screening algorithms to assess candidates at scale.
Why are AI hiring filters limited?
AI hiring filters are designed to sort candidates against a threshold. That can help identify who may be unqualified, but it does not necessarily identify exceptional candidates, especially those with unconventional experience or non-linear career paths.
What is the alternative to screening-first hiring?
The alternative is to give candidates clearer, more truthful information about the role, the work, the expectations, and the conditions so they can make better decisions about fit before they apply.
How does Happydance approach candidate experience differently?
Happydance helps employers create careers website experiences that inform, guide, and empower candidates before they apply, rather than treating candidate judgment as something to replace with another screening tool.
