A medical student applied to jobs, got no interviews, and decided to find out why. He spent six months writing Python code to investigate whether an AI screening algorithm had filtered him out before a human ever saw his application. Wired reported on his investigation - and while the ending isn't tidy, the story is a clear look at a problem millions of job applicants face without knowing it.
The difficulty wasn't access to data or technical skill. It was that the system he was trying to investigate is designed, by default, to be opaque to candidates. Employers don't have to disclose which screening tools they use. Vendors protect their scoring models as proprietary. Rejection notices rarely say whether anyone looked at the application at all. Even six months of dedicated investigation couldn't produce definitive proof.
How ATS Filtering Works
Most large companies now route applications through ATS (applicant tracking system) software before a recruiter gets involved. The simplest versions do keyword matching: your resume needs to contain the specific language from the job description or it may not clear the first filter. A highly qualified candidate who writes "overseeing projects" instead of "project management" can fail a simple text check and never know it happened.
More sophisticated systems use scoring models trained on historical hiring data, looking at patterns across thousands of past successful hires. The problem: "successful hires" reflects whoever the company chose in the past. If that history included patterns of bias, the model encodes them without anyone explicitly programming it to. A hiring algorithm isn't neutral just because it's automated.
The Regulatory Picture
New York City's Local Law 144, passed in 2021, requires employers to audit AI hiring tools for bias before deployment. Those audits have been criticized as superficial, often run by firms hired by the same vendors being evaluated. The EU's AI Act classifies hiring algorithms as high-risk AI systems requiring documentation, transparency to affected individuals, and meaningful human oversight - a more serious framework, but one that applies to EU employers, not US ones.
The practical options for US job applicants right now: match resume language to job description keywords precisely, apply through referrals when possible, and document your applications if you suspect a pattern. None of that gives you the explanation you're owed.
The medical student's story matters not because he found a smoking gun - he didn't - but because he tried with real tools and still couldn't get there. That gap between the systems making decisions about people's careers and those people's ability to understand those decisions is the actual story here.