When hiring algorithms think alike: the risk of AI monoculture
Most conversations about AI bias in recruitment focus on a single employer's tool making unfair decisions. But a new study - Algorithmic Monocultures in Hiring - asks a bigger question: what happens when employers use the same tool?
The study found that, when multiple organisations rely on the same AI screening systems (or systems trained on similar data), their decisions end up highly correlated. The same candidates get through everywhere, and the same candidates get rejected everywhere. The researchers refer to this as “algorithmic monoculture” (a concept borrowed from ecology, where a lack of diversity in a system makes it more fragile).
Typically, a candidate rejected by a biased hiring manager at one company can still find success elsewhere. However, a candidate filtered out by a widely used algorithm may find that every door is closed to them, not because of anything they have done, but because every employer's "objective" tool has reached the same conclusion. This is a fundamentally different kind of problem that existing accountability frameworks, which focus on individual employers, are not well set up to address.
Our take
For businesses, this might prompt some uncomfortable questions. For example, do you know how many of your competitors use the same recruitment AI as you do? What profiles does it systematically favour or penalise? Ultimately, if the same tool produces the same blind spots across the market – particularly if these correlate with protected characteristics – there is a risk of discrimination that cannot be ignored.
AI hiring tools are not plug-and-play. Employers should ask vendors hard questions about training data, optimisation targets, and market penetration. They should also think carefully about whether a single algorithmic screening process, without meaningful human input at the filtering stage, is actually reducing risk or quietly concentrating it.
