As employers turn to algorithms to sift through job applications, the poor are more likely to lose out
Even when wrong, their verdicts seem beyond dispute - and they tend to punish the poor — few years ago, a young man named Kyle Behm took a leave from his studies at Vanderbilt University in Nashville, Tennessee. Tweets: @ilparone , @mkelley , @ilparone , @digiphile and @digiphile Tweets: Jarno M. Koponen / @ilparone : “We're not told what the tests are looking for. The process is entirely opaque.” http://www.theguardian.com/... by @mathbabedotorg #ml #privacy Mike K / @mkelley : “the Kronos test can be considered a medical exam, the use of which in hiring is illegal under the ADA of 1990.” http://twitter.com/... Jarno M. Koponen / @ilparone : “We're not told what the tests are looking for. The process is entirely opaque.” https://www.theguardian.com/ ... -@mathbabedotorg v @Blakei #ml #ux Alex Howard / @digiphile : “these algorithms have the potential to create an underclass...inexplicably shut out from normal life"-@mathbabedotorg http://www.theguardian.com/... Alex Howard / @digiphile : “the algorithms that power the data economy are based on choices made by fallible human beings"-@mathbabedotorg http://www.theguardian.com/...
Context & Ripple Effects
This Guardian investigation is an early entry point in a story that keeps widening: Kyle Behm's rejection by a Kronos personality test — criteria undisclosed, verdicts effectively undisputable — is the hiring-side instance of a scoring regime that later reaches far beyond employment. By 2020, predictive algorithms were setting police patrols, prison sentences, and probation rules on both sides of the Atlantic.
What connects the coverage is who bears the error rate: low-income applicants and defendants with the least recourse. The same year the Guardian reported on hiring screens, Predictim began AI-scoring babysitters from years of social media activity for parents — character scores sold as consumer products — while [[a:960741|civil lawyers started building litigation strategies against automated systems that deny the poor basic services]]. The through-line is opacity plus asymmetry: the scored party cannot see the rubric, so the only check is legal.
First-order effects
- Low-income job applicants like Behm are screened out by tests whose criteria are never disclosed, with no meaningful way to appeal a wrong verdict.
- Employers deploying the Kronos test carry direct legal exposure if observers are right that it functions as a medical exam, which would make its use in hiring illegal under the ADA.
Second-order effects
- Vendors of screening algorithms face pressure to defend their instruments against disability-law challenges, since one successful ADA claim would unsettle every employer customer relying on the same test.
- The scoring model spreads to adjacent markets — gig work, childcare vetting — where platforms sell automated character judgments to buyers rather than employers, multiplying the population scored without consent.
Third-order effects
- If the pattern holds, algorithmic gatekeeping consolidates across hiring, policing, sentencing, and probation into a single structural problem: the poor face machine-made denials at every institutional door, and adversarial litigation becomes the primary accountability mechanism because the systems themselves offer no appeal path.
- Regulatory attention follows the litigation: once courts treat opaque scoring as a disability or due-process issue, disclosure requirements for automated decision tools become the likely policy response.
The trend: Opaque algorithmic scoring is expanding from hiring into policing, sentencing, and everyday vetting, with courts and civil-rights law emerging as the main counterweight.