Projects & Research

I evaluated whether ChatGPT could reliably assess job candidates by comparing AI-generated interview ratings to those from trained human evaluators. The study examined 183 interview responses to test whether AI ratings were accurate, fair, and consistent across different demographic groups.

What I Did: I analyzed how closely ChatGPT's scores matched human evaluators' ratings using statistical methods (intraclass correlation, Pearson correlation, t-tests, and Fisher's r-to-z transformation). I also tested whether rating patterns differed based on candidates' gender, race, or age.

What I Found: ChatGPT showed strong overall agreement with human raters (ICC = 0.94), but systematic problems emerged. The AI consistently gave higher scores than humans; an average of 6.5 points more on a 5-point scale; and showed demographic bias patterns that humans didn't exhibit. Specifically, ChatGPT underrated Black candidates and overrated middle-aged candidates.

Why This Matters: Organizations are rapidly adopting AI tools to screen job candidates, but this research shows those tools can introduce bias even when they appear to work well overall. Companies using AI in hiring need validation frameworks, fairness audits, and human oversight to catch these disparities before they affect real hiring decisions.

I examined the ethical implications of organizations adopting Generative AI tools (like ChatGPT) in Human Resources functions; particularly in hiring, onboarding, and performance management. Through literature review and qualitative analysis, I identified key ethical concerns including algorithmic bias, privacy violations, accountability gaps, and copyright issues.

Key Findings: AI tools trained on biased datasets can perpetuate discrimination against protected groups, even when organizations adopt them specifically to "reduce human bias." I found that many HR professionals lack awareness of how these systems work, creating transparency and accountability problems when AI-generated decisions affect people's livelihoods. The research also revealed privacy concerns around AI systems mining personal data from social media and other sources without candidates' knowledge or consent.

Recommendations: Organizations should require human oversight for all high-stakes HR decisions, conduct regular fairness audits of AI tools across demographic groups, and ensure candidates are informed when AI systems are used to evaluate them. The thesis argues for interdisciplinary collaboration between I/O psychologists, HR professionals, and AI developers to create ethical frameworks for responsible implementation.