Ortho-Heterodox Biases and the Economist Algorithms of ChatGPT

Abstract

Recommendations for economic policies can be based on different theoretical perspectives and may present hidden biases. Identifying these biases is challenging when they are embedded in recommendations from sources with high technological and social disruptive potential, where a good level of impartiality is expected, such as contemporary large language models. Thus, a questionnaire was administered to economists affiliated with the Brazilian academic community to assess their perception of orthodox/heterodox biases in economic policy recommendations derived from interactions with ChatGPT. The results showed that: i) there is still no consensus on the concepts of orthodoxy and heterodoxy in Brazil; ii) there are indications of a positive relationship between how self-proclaimed heterodox (orthodox) an economist is and how heterodox (orthodox) the perceived bias in an economic policy is; iii) it was not possible to identify a consistently orthodox or heterodox bias in ChatGPT's recommendations, which exhibited a good degree of impartiality.