Summary — "How companies use AI to choose who gets hired and fired" (NPR, TED Radio Hour)

Overview

This episode explores how companies increasingly use artificial intelligence across the entire employee lifecycle — from sourcing and screening applicants to monitoring performance and deciding who gets fired. Investigative journalist Hilke (Hilka) Shellman, author of The Algorithm, describes where these tools help, where they fail, and how they can amplify bias and invade worker privacy.

Key points & main takeaways

• Hiring has moved from classifieds → job boards → massive online volume, which created pressure to automate screening.
• Applicant volume is enormous (examples: Google ~3M applications/year; IBM ~5M), prompting companies to rely on software and AI to triage candidates.
• Early machine-learning resume tools (notably Amazon’s) learned from historical hires and reproduced prior biases — Amazon’s system penalized resumes that included “women” or references to women's groups, so it was shelved.
• Newer tools include:
  • One‑way video interviews evaluated by algorithms using words, intonation, facial expressions.
  • Game‑style assessments and AI-powered background checks.
  • On‑the‑job monitoring: keyword logging, Zoom meeting analysis (who spoke, sentiment), emotion detection from video/text.
• Many of the claims (e.g., facial expressions or a brief behavior in an interview predict job success) lack robust scientific validation; experts warn much of this is correlation, not causation.
• AI can multiply harms: a biased algorithm applied at scale can systematically disadvantage whole groups rather than isolated human decisions.
• Secrecy is common: companies rarely disclose what algorithms do or what data they use, partly out of liability concerns — creating a "cloak of silence."
• The problem is complex: humans are biased too, so the solution isn’t simply reverting to human-only hiring. Instead, careful testing, transparency, and governance are needed.

Notable quotes / insights

• “It got so easy to apply for a job that everyone started.” — on how application volume exploded after job boards.
• “The algorithm assumed that men would be best suited for the job.” — summary of Amazon’s resume tool failure.
• “An algorithm that is used across all of the resumes… could, like, just multiply the harms.” — on scale effects of automated bias.
• “This cloak of silence now extends to how AI is being used to hire people, track them on the job, and even decide who gets fired.” — on lack of transparency.
• Experts’ critique: facial-expression analysis for hiring “is just correlation… pure rubbish” and may drive discrimination rather than reduce it.

Topics discussed

• Evolution of hiring (classifieds → online job platforms → AI)
• Applicant-tracking systems and resume-screening algorithms
• Machine learning failures and bias (Amazon example)
• One-way video interviews and facial/emotion analysis
• AI-driven assessments, games, and background checks
• Workplace surveillance: keystroke/keyword logging, meeting analytics, sentiment/emotion monitoring
• Scientific validity of predictive claims and the risks of opaque systems
• Legal/liability concerns and the social harms of automated decision-making

Action items & recommendations

For employers and vendors

• Run independent audits of AI tools for accuracy and disparate impact before deployment.
• Require and publish validation studies demonstrating predictive value tied to demonstrated job performance.
• Keep humans in meaningful oversight roles; don’t let opaque models make sole determinations about hiring/firing.
• Avoid using unvalidated behavioral signals (facial expressions, micro‑expressions) as decisive criteria.
• Be transparent with applicants and employees about what tools are used, what data is collected, and how decisions are made.

For policymakers & regulators

• Mandate transparency and explainability standards for employment algorithms.
• Require bias testing and documentation (e.g., adverse impact analyses).
• Create enforceable notice, consent and data‑protection rules for workplace monitoring.

For job seekers & employees

• Ask employers what tools they use, request explanations of automated decisions, and document communications.
• Where possible, request human review or alternative assessment methods if you suspect automated screening harmed you.
• Advocate for workplace policies limiting invasive monitoring and requiring informed consent.

For researchers & journalists

• Continue independent investigations into how algorithms are trained, validated, and applied across companies.
• Publicize cases where algorithms produce demonstrable disparate impacts to spur reform.

Final note / further reading

• The episode’s reporting is based largely on Hilke Shellman’s book, The Algorithm, which dives deeper into concrete examples, investigations, and the broader implications of workplace AI.