Answers are grounded in this paper and cite their sources.
The article examines the effectiveness, risks, and ethical limitations of artificial intelligence (AI) in recruitment. A distinction is made between deterministic automation, algorithmic systems, machine learning, and generative AI, which makes it possible to avoid equating any digital HR tool with artificial intelligence. Based on a systematized review of recent academic literature, comparative and content analysis, as well as analysis of regulatory documents, the evidence regarding the impact of AI on recruitment performance is synthesized. It is shown that the strongest evidence concerns operational effects: the ability to process large volumes of applications, automate repetitive tasks, support sourcing, screening, matching, and communication, and reduce recruiters’ workload. At the same time, the evidence is substantially less conclusive with respect to quality of hire, candidate experience, and the elimination of bias. The article analyzes algorithmic bias, indirect discrimination, opacity, false exclusion of qualified candidates, privacy and information-security risks, automation bias, and the potential dehumanization of recruitment. Particular attention is paid to meaningful human oversight and to the current regulatory framework, including the EU AI Act, the GDPR, and Ukrainian legislation. An author-developed matrix for assessing the appropriateness of AI use in recruitment is proposed. It combines the expected effect, strength of evidence, discrimination and privacy risks, severity of potential error, and the required level of human oversight. Three modes of AI use are substantiated: automation-preferred, AI-assisted decision-making, and human-led decision-making. The principle of proportionality between automation and risk is formulated, according to which the permissible autonomy of an AI system should decrease as the potential harm of an algorithmic outcome increases.
No takes yet. Share an insight, caveat, or question.
Драган et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: