Large language models show human-like content biases in transmission chain experiments

Abstract

As the use of Large Language Models (LLMs) grows, it is important to examine if they exhibit biases in their output. Research in Cultural Evolution, using transmission chain experiments, demonstrates that humans have biases to attend to, remember, and transmit some types of content over others. In five pre-registered experiments with the same methodology, we find that the LLM ChatGPT-3 replicates human results, showing biases for content that is gender-stereotype consistent (Exp 1), negative (Exp 2), social (Exp 3), threat-related (Exp 4), and biologically counterintuitive (Exp 5), over other content. The presence of these biases in LLM output suggests that such content is widespread in its training data, and could have consequential downstream effects, by magnifying pre-existing human tendencies for cognitively appealing, and not necessarily informative, or valuable, content.