What do Large Language Models ‘think’ determines occupational prestige?

In this paper we ask a Large Language Model (GPT-4o) to rate the social status associated with a list of common occupations. We then ask it to justify these judgements. Based on the rationale that the associations present in Large Language Models represent a reflection of human opinions and biases, we analyse the model’s responses in relation to existing theory and evidence on the determinants of occupational prestige. We find that GPT-4o invokes a number of well-established determinants of occupational prestige – including economic rewards, education and training, social value, and authority over others. However, we also find that the model raises a number of determinants that have been largely overlooked in the existing occupational prestige literature – including, most notably, the ‘visibility’/fame associated with a particular profession.