Do AI-Generated Biases Shape Human Cognition? Evidence from a Reaction-Time Experiment

Since recent advances in Artificial Intelligence (AI), new questions have emerged regarding how these tools influence human behavior and cognition. In the case of Large Language Models (LLMs) —algorithms trained on vast corpora for language generation— their proximity to human linguistic abilities has been increasingly debated. As human communication is goal-oriented, a central concern is whether the information conveyed by AI exhibits bias and whether such bias affects users information processing positively or negatively. Evidence indicates that AI systems can reproduce biases embedded in their training data (Bender et al. 2021), raising the risk that certain statements —potentially non-objective or disrespectful— may become fossilized in human cognition. Our study collects statements about men and women from Google's autocomplete feature and presents them to participants while measuring reaction times to assess if and how AI-related biases influence cognitive processing. This approach examines whether exposure to such information modifies human opinions or behaviors from a gender-based perspective.