Picture the nightmare, as sketched by a growing chorus of scholars: artificial intelligence swallows the hiring process, and one day every resume in America slides across the desk of the same algorithm — a single silicon gatekeeper, an algorithmic monoculture, in which a rejection from one firm is automatically a rejection from every firm, and a whole class of unlucky souls is locked out of work forever. Systematic exclusion, the critics call it, and it has the ring of dystopia. Now two MIT researchers have taken that nightmare apart, bolt by bolt, and their verdict is: breathe easy, mostly.
In a paper in Philosophical Perspectives, Brian Hedden — a professor in MIT’s Department of Linguistics and Philosophy with a shared position in electrical engineering and computer science and a principal investigator in the Laboratory for Information and Decision Systems — and Manish Raghavan — the Drew Houston (2005) Career Development Professor at the MIT Sloan School of Management and in EECS, also a LIDS principal investigator — systematically evaluated the major objections to algorithmic monoculture. Their conclusion: systematic exclusion and many other arguments either fail outright or aren’t decisive against all forms of monoculture.
Start with the marquee fear. If every firm screens resumes with the same algorithm, and that algorithm bounces your application, you get bounced everywhere — surely a catastrophe for the bounced. But running a series of models capturing multiple situations, the researchers found the argument isn’t compelling: the overall number of people hired doesn’t change just because firms share an algorithm. All the jobs still get filled. And there’s a twist — the competition cuts the other way. “All the jobs get filled and the same number of people have jobs, but the firms are fighting over the same pool of candidates, which actually drives up wages,” Raghavan says. The monoculture, in this rendering, is not the job seeker’s executioner but his agent.
Monoculture, for the record, is no science-fiction scenario. Lending was once the province of individual bankers exercising idiosyncratic judgment; now every banker leans on standardized credit scores derived from the Fair Isaac Corporation’s FICO algorithm. A handful of resume-screening algorithms are already commonly used across many Fortune 500 companies. “The worry is that, as more people use AI and algorithms to get information and make decisions, there is more of a vehicle for this kind of correlation to occur,” Raghavan says.
What about fairness to the applicant as a person — the agency objection? If your resume is forwarded to every firm on the platform, you never get the chance to tinker with it and try again. “This seems like a good objection to bad forms of monoculture,” Hedden concedes. “But if you have a monoculture where you get to revise your resume and resubmit your materials, then this doesn’t hold up.” And the mirror-image worry — that one universal algorithm invites gaming, with candidates reformatting their resumes to please the machine — doesn’t obviously discriminate either. “In the latter scenario, you might just target a couple of firms’ algorithms and try to game them,” Hedden says of the polyculture world, “giving yourself a bit of advantage with a few employers.”
The objection that survives scrutiny is subtler and, in its way, more interesting: monoculture creates informational echo chambers that hinder exploration — and this the researchers prove mathematically. The logic runs through the “wisdom of crowds,” the social-psychology finding that a diverse group of independent decision-makers can outperform any single judge. Many firms using many different algorithms will, collectively, surface a higher-quality pool of hires; one algorithm everywhere tends to crown the same characteristics and credentials every time, at every firm, and the system stops discovering the oddball who might have been the better bet. “Monoculture might inhibit the amount of discovery that happens overall,” Raghavan says. “It is not clear if that is a bad thing, but it is definitely a worry when we think about designing AI for applications like science, art, or writing.”
Even that flaw has a patch. Build randomness into a monocultural platform, Raghavan suggests, and you induce more exploration. And performance depends on the algorithm itself: a single highly accurate algorithm can beat a market full of mediocre ones. Better yet, bundle the various firms’ hiring algorithms into one “ensemble” — a single score derived from the average — and the downsides shrink. In simulations of different hiring situations, the researchers confirmed that such an ensemble algorithm could sometimes outperform a polyculture of separate algorithms. How feasible ensembling would be in practice, Hedden notes, remains unexplored.
The pair are careful to bound their claims. Hiring was their laboratory, and the approach could extend to domains like lending, but generative-AI content creation or AI-guided scientific research may work differently — monoculture there may be genuinely more problematic. “A trend toward algorithmic monoculture is a realistic scenario, and a really important issue that is being brought about by the use of AI, but it is hard to say in the abstract whether monoculture would be a bad thing,” Hedden says. “It depends on the details, like the domain we are talking about and the accuracy of the algorithm itself.”
“A lot of the answers around the promises and pitfalls of algorithmic monoculture are going to be contextual,” Raghavan adds. “Even from a research perspective, there is still a lot of work to be done to figure out how we can approach these concerns from an empirical perspective.” The message, in other words, is not that the one-algorithm future is benign — only that the case against it has been argued with slogans where it needed mathematics, and the mathematics turns out to be fussier, stranger and more conditional than the slogans.