Can AI become a bias interruptor?

A novel approach to debiasing decision-making

‍

Cognitive bias has been a persistent problem in behavioral decision making, particularly in high stakes social contexts where fairness matters. At this year's Academy of Management Annual Meeting, I presented my research examining a question that has become increasingly important: Can AI reduce human bias in high-stakes decisions?

Addressing bias in employment decisions

Employment is one prominent example of high stakes social decision-making in which perceived bias contributes to thousands of discrimination claims each year.

‍

Many of the programs which organizations have relied upon, including unconscious bias training, have produced limited practical impact. And, recent challenges to policies which serve to mitigate bias have left a gap in bias interventions.

‍

This highlights an important opportunity: If human bias persists despite our best intentions, we must identify a method of interrupting it more effectively.

The AI paradox

Today, recruiters increasingly use AI to improve search productivity. Real world examples ranging from Amazon to Unilever have demonstrated an AI paradox in the recruiting context: AI, the same technology that accelerates bias, can also interrupt it.

‍

In one case, historical data trained the algorithm to embed historical bias — amplifying human bias. In another, competency anchored algorithms evaluated candidates consistently against the same rubric and resulted in greater diversity than human decision-making alone.

‍

This paradox has understandably fueled concern about AI in employment decisions. The resulting key question is: Under what conditions can AI overcome human bias?

The question organizations should be asking

Much of today's employment fairness discussion in US practice focuses on whether AI results in disparate impact in hiring. My research suggests that may be the wrong question.

‍

Instead, organizations should ask: How does the level of bias in AI decision-making compare to human decision-making?

‍

If algorithms embed human bias, they will no doubt accelerate it at scale. On the other hand, I found that AI built with fairness principles can overcome human bias.

‍

This distinction has profound implications for the future of fair and responsible AI. New data from Warden AI reinforces this: recent research reveals AI systems scoring an average of 0.94 on fairness metrics, compared to 0.67 for human-led hiring — a difference with direct implications for the discrimination claims organizations are already managing.

Looking ahead

The future of AI in high-stakes decision making will not be determined solely by advances in model capability.

‍

It will be determined by whether we intentionally design systems that help people make better, fairer decisions rather than faster versions of the same biased ones.

‍

As organizations continue integrating AI into consequential decisions, the opportunity is not simply to automate judgment, but to improve it. That starts with building AI on responsible principles — not just deploying it for speed.