AI and gender stereotypes: the importance of awareness

Large Language Models (LLMs), like ChatGPT, may contribute to the perpetuation of gender stereotypes. This issue has been extensively studied by Prof. Monica Landoni, an adjunct professor at the Faculty of Informatics at Università della Svizzera italiana (USI). She tackled the topic in an article published in LaRegione.

The problem can be highlighted with a simple example, explained Professor Monica Landoni. Just ask a machine translation tool to translate the phrase "a doctor and a nurse" from English into Italian. The result will be "un dottore (male) e un'infermiera (female)". "The system operates by creating a mapping in the requested language. In cases of ambiguity, it selects the most likely declension of the term, often assigning medical professions to men and nursing professions to women. This reinforces the stereotype that certain jobs are exclusively male while others are solely female, a notion that remains prevalent in society," explained Professor Landoni to clarify the reason for the translation.

Part of the problem could be because the data available to the AIs is biased: "Examining the issue from a gender perspective reveals a contrast. On one hand, much of the available data focuses on male-dominated workplaces and living environments, where men predominantly hold the most prestigious roles, while women are often assigned caregiving tasks. On the other hand, there is a significant lack of disaggregated data specifically concerning women," clarified the USI professor. There is a significant lack of data regarding women, as noted by Caroline Criado Perez in her 2019 essay. This issue is compounded by the fact that, as Professor Landoni pointed out, women are more challenging to study due to social and biological factors. These factors introduce a greater number of variables that can disrupt the regularity of their daily lives.

Another female peculiarity is that they have less linear routes, both in a literal sense (women tend to make more stops than men during their daily commute for practical needs such as shopping or childcare) and in a metaphorical sense: "We conducted an analysis of women's histories in informatics and found that their career paths are often less linear than those of men. For instance, many women initially study in other, more 'comfortable' fields before transitioning to informatics. This field is frequently not their first choice because they do not feel included in a world that is still perceived as predominantly male and exclusive, resembling a sort of boys' club." explained Prof. Landoni.

The number of female students enrolled in the Faculty of Informatics (at USI, they account for around 17% of total students) worries Landoni: "Restricting women's choice for this path greatly limits the diversity of profiles employed. As a result, algorithms will be developed in a one-dimensional manner, reflecting the needs and perspectives of only a limited segment of the population". There are many practical examples, one of them being the size of smartphones, which are too big for women's hands, making it difficult for women to type with one hand.

The "Tools for Assessing and Developing Affecting & Attractive Narratives for Girls in Informatics" (TADAA) project was created to break down the prejudice that the informatics environment is not for women. It aims to help young people, parents, and teachers understand that informatics is accessible to everyone: "The idea is to dismantle the preconception from an early age that informatics is not a subject for girls and to limit the mentioned tortuous paths that we often see in female colleagues. Additionally, it is essential to empower university staff to create welcoming environments for girls, ensuring they do not feel isolated or excluded".

The aim is to engage young participants in fun activities that encourage reflection on the fallibility of digital tools, helping them identify these tools' limits.

The complete interview with Monica Landoni, edited by Cristina Pinho for LaRegione, is available at the following link (Italian only).