Class, gender and race stereotypes reproduced by generative AI: recommendations for users

Abstract

Objective: To establish recommendations for prosumers to make ethical use of these tools, thus avoiding results that fall into class, race and/or gender stereotypes.

Method: Analysis of documentary material developed by different entities: international institutions (UNESCO and Council of Europe) and companies developing AI tools (OpenAI, Google, Microsoft and Anthropic).

Finding: AI can perpetuate stereotypes as it uses data that reproduces a context where minorities and women are underrepresented. Major corporations claim to have developed codes of conduct and self-regulation. This governance is in line with international recommendations and governs their behaviour, protects them from liability and underpins their CSR. They include recommendations for users in the processes of using the tools.

Conclusions: It is not yet possible to completely avoid biases but it is advisable for users to detect stereotypes and minimise them. Self-regulation contributes to less discriminatory content, as well as incorporating women and minorities as developers. They propose codes of conduct to detect inappropriate content. The findings are useful to encourage a critical perspective and focus on the role of users, to whom recommendations are offered, in a process in which the use of tools is fundamental.