Structured Work Samples and Interviews as Alternatives to Degree Screens Under AI-Contaminated Assessment

Abstract

Generative AI complicates the inference that assessed academic work provides evidence of individual competence, but removing degree requirements may shift selection toward discretionary interviews, referrals, or network advantage. We develop a stylized simulation of screening for non-licensed occupations in which employers choose among degree-only, interview-only, degree-plus-interview, degree-plus-work-sample, work-sample-only, direct-hybrid, all-signal augmented-hybrid, and network-fallback regimes. Across 600 design scenarios with 5,000 synthetic applicants each, signal validity, AI contamination, degree costs, direct-assessment bias, direct-signal dependence, referral inequality, and hire rates vary over literature-informed ranges. The direct-hybrid regime combining work samples and structured interviews maximized a performance-equity objective in 40.5 percent of scenarios, followed by the all-signal augmented hybrid in 26.7 percent and work-sample-only screening in 21.2 percent. The direct hybrid exceeded degree-plus-interview screening on the objective in 92.7 percent of scenarios and degree-plus-work-sample screening in 83.3 percent, but exceeded the all-signal augmented hybrid in only 60.2 percent. Sensitivity checks show that the direct-hybrid advantage largely disappears when direct and degree signals have matched validity, high dependence, and no modeled group penalties. The model supports a conditional claim: degree-free reform depends on structured, job-relevant, accessible evidence rather than informal networks or costly, biased, duplicative assessments.