Chapter 5
Recruitment and selection with artificial intelligence
Validity, accessibility and non-discrimination in access to employment.
5.1 Job analysis
Defensible selection begins by identifying the role's tasks, competencies and conditions. Criteria must relate to these requirements and be explainable. Copying characteristics of previous employees can perpetuate a historical workforce composition without demonstrating that those characteristics are necessary.
Tests must suit what they are intended to measure. Validity is not inherited by using a familiar technology. It must be examined whether the assessment provides relevant information and whether its consequences are consistent with its stated purpose.
5.2 Application screening
Artificial intelligence can help locate information in applications or compare explicit requirements. Missing information and differences in format must be treated cautiously. An unconventional résumé does not demonstrate lower ability, and an extraction error can exclude a suitable person.
Information retrieval should be separated from rejection decisions. Reviewable evidence must be retained and alternative routes offered where technical barriers exist. Criteria must remain stable and must not be changed covertly during the process.
5.3 Interviews and accessibility
Structured interviews use job-related questions and defined rating criteria. Digital tools can organize the process, but they do not justify unvalidated inferences about personality or emotions. Accent, disability or connection quality must not become hidden disadvantages.
Accessibility requires clear instructions, usable formats and reasonable accommodations in light of the circumstances and applicable framework. The alternative offered must allow a comparable assessment without penalizing someone who needs another participation channel.
5.4 Fairness and auditing
Fairness cannot be reduced to removing names or photographs. Other variables may act as proxies for protected characteristics. Auditing must examine the definition of success, historical data, selection errors and outcomes across the entire process.
Metrics must be interpreted with attention to sample sizes and uncertainty. A statistical difference requires investigation; similar averages do not prove an absence of discrimination either. Conclusions must be documented and linked to specific corrective actions.
5.5 Decisions and communication
Decisions must be linked to relevant criteria and sufficient evidence. Applicants need understandable information about the process and review channels. An explanation need not disclose other people's data or secrets unnecessary to justify the outcome.
The European framework identifies certain uses of artificial intelligence in employment as high risk. The classification, obligations and applicable timelines must be checked against current official sources; they do not depend solely on the tool's commercial label [R1].
Questionnaires
Chapter knowledge & Human Resources dilemmas
Dilemmas receive no moral or employability score and are not used to assess job candidates.