AI Displays Bias when Judging People, and that Matters for some of Its Most Common Uses

Why It Matters

AI‑driven decisions are moving from advisory to decisive roles; systematic bias and model variance threaten fairness and legal compliance across hiring, credit and litigation. Understanding these patterns is essential for firms to mitigate risk and choose appropriate models.

Key Takeaways

• AI scores competence, integrity, kindness as separate columns
• Biases persist across age, religion, and gender
• Model‑to‑model differences can flip outcomes for the same trait
• Choosing a model directly shapes loan, hiring, and legal decisions

Pulse Analysis

The latest study from Hebrew University’s Business School reveals that today’s most advanced language models, including those powering ChatGPT and Google’s Gemini, are not merely processing data—they are forming judgments that resemble human trust. By breaking down trust into discrete components—competence, integrity and benevolence—these systems apply a spreadsheet‑like logic that is far more consistent than human intuition, yet it lacks the holistic nuance people use when evaluating others. This structural similarity explains why AI can predictably assess risk, but it also creates a new vector for bias.

When the researchers ran over 43,000 simulated decisions, a clear pattern emerged: AI consistently favored certain demographic groups. Older individuals often received more favorable financial offers, while religion and gender also swayed outcomes, sometimes in opposite directions across different models. The bias was not random; it was systematic, predictable, and in some cases stronger than human prejudice. Moreover, the study showed that two ostensibly similar models could reach opposite conclusions about the same person, meaning that the specific AI chosen can dramatically alter real‑world results.

For businesses, the implications are immediate. In recruitment pipelines, AI‑screened candidates may be unfairly advantaged or penalized based on age or gender, exposing firms to discrimination lawsuits. Financial institutions using AI for credit decisions risk regulatory scrutiny if demographic bias is embedded in their algorithms. Legal practitioners relying on AI for risk assessment must also account for model‑specific quirks. Companies should therefore audit model outputs, diversify model portfolios, and maintain human oversight to ensure that AI augments rather than distorts equitable decision‑making.