Humans Inherit Biases from Artificial Intelligence
Recommendations from artificial intelligence are sometimes wrong and biased. New research conducted by psychologists Lucía Vicente and Helena Matute from the University of Deusto in Bilbao, Spain, provides evidence that people can inherit artificial intelligence biases, which are systematic errors in AI results, in their decisions.
In their study, Vicente and Matute hypothesized that individuals performing a simulated medical diagnosis task with the assistance of a biased AI system would replicate the model’s bias in their own decisions, even when transitioning to a context without AI support. To test this hypothesis, three experiments were conducted.
Participants were asked to complete a medical-themed classification task, either with or without the aid of a biased AI system. The researchers found that the biased AI recommendations influenced the participants’ decisions. Interestingly, when these participants, who were initially assisted by the AI, later performed the task without any assistance, they made the same errors that the AI had made during the previous phase. This suggests that participants’ responses mimicked the AI’s bias, even when the AI was no longer providing suggestions. The results of these experiments provide compelling evidence of human inheritance of AI bias.
The Risks of AI Biases
The remarkable capabilities of artificial intelligence systems, such as engaging in human-like conversations, have led to the perception of high reliability in this technology. Many professional fields are increasingly adopting AI-based tools to support decision-making and minimize errors. However, AI is not immune to biases in its results.
It is crucial to consider that AI models are trained using data that reflects past human decisions. If this data contains patterns of systematic errors, the AI algorithm will learn and reproduce these errors. Numerous studies have already demonstrated that AI systems inherit and amplify human biases.
Humans Inheriting Biases from AI
The research conducted by Vicente and Matute reveals an intriguing finding: humans can also inherit biases from AI. This means that not only does AI inherit biases from human data, but people can also inherit those biases from AI, creating a potentially dangerous loop. The results of their study have been published in Scientific Reports.