Artificial intelligence (AI) is increasingly being used in the hiring process. However, a study suggests that AI may actually form biases more often and more strongly than humans do. Researchers found that when AI models were given the task of hiring for different jobs based on candidates from four fictional ethnic groups, they rapidly started segregating roles based on early observations.

The findings are concerning because these biases could persist even in sectors where human bias is traditionally less pronounced. Moreover, as AI systems gain better memory and personalization features, their potential to form and act on biased stereotypes increases. This raises questions about how fair the hiring process might become in the age of advanced AI.

The study also revealed that simply telling these models to be fair didn’t significantly change their behavior. However, providing them with more relevant information about individuals could reduce their biases. This suggests a potential avenue for mitigating AI bias but highlights the need for careful design and oversight.

As LLMs learn from experience in making decisions that affect people’s lives, we must consider the ethical implications of these new biases. The models might start showing preferences or prejudices that they have never been explicitly taught by humans. This could lead to a world where AI perpetuates social divisions and inequalities without us even realizing it.