New research by psychologists Lucia Vicente and Helena Matute from the University of Dústo in Bilbao, Spain, provides evidence that people can have artificial intelligence biases (systematic errors in AI decisions) in their decisions.
Humans gain bias from artificial intelligence
|
Summary
Artificial intelligence recommendations are sometimes wrong and biased. In our research, we hypothesize that people performing a (simulated) clinical diagnostic task with the help of a biased AI system will reproduce the model’s bias in their own decisions, even when moving to an environment without AI support. In three experiments, participants completed a medical-themed classification task with or without the help of a discriminating AI system. Biased AI recommendations influenced participants’ decisions. Furthermore, when those participants went to perform the task unaided with the help of the AI, they made the same errors that the AI had made in the previous phase. Thus, even when the AI does not make recommendations, participants’ responses reflect the AI’s bias. These results provide evidence of human inheritance of AI bias.
|
comments
Surprising results obtained by artificial intelligence systems, for example, being able to maintain a conversation like a human, have given this technology an image of high reliability. More and more professional fields are implementing AI-based tools to support decision-making to reduce errors in experts’ decisions. However, this technology is not without risk dependence In AI results. We must consider that the data used to train AI models reflects past human decisions. If you hide these forms of data Systematic errors, the AI algorithm will learn and reproduce these errors. Indeed, there is ample evidence to suggest that AI systems inherit and amplify human biases.
A more relevant finding of Vicente and Matute’s research is that the opposite effect can also occur: what Humans gain bias from AI. This means that not only can AI derive its biases from human data, but people can also inherit those biases from AI, which risks getting stuck in a dangerous loop. Scientific reports Publishes Vicente and Matute’s research results.
