I

Digital platforms - spaces defined by being beyond the physical and the tangible - still leave room for discrimination against women. It is hard to accept that fields such as technology and science, which fly the banner of progress and human wellbeing, can at the same time create and reproduce gender injustice . It has to be said that technology and science are only thought incompatible with such injustice if we take the impartiality, neutrality and objectivity of scientific disciplines as given. The reality is that the sciences are deeply embedded in society, and society interacts with them. From the design, the goals pursued and the production through to the use made of Artificial Intelligence, it concerns us all because it affects us all.

Artificial Intelligence has become hugely important in a short time, appearing across many areas of activity - work, care, defence, culture, society, the law - and prompting a broad social response. Its influence is such that we depend on the decisions AI arrives at, which makes the presence of gender bias within it a serious problem (Belloso, 2023). Yet the speed at which this sector has developed has not been matched by careful questioning and assessment of how Artificial Intelligence behaves and what effects it has on our lives.

One of the main reasons this has not happened is the prestige and unquestioned respect granted to STEM fields in general. The idea is that scientific disciplines rest solely on facts and that therefore the design, the methodology and the product contain no errors. Resting solely on facts, bias would be impossible, since bias is a product of human intelligence, where value judgements are built (consciously or unconsciously) into the work. In this way Artificial Intelligence draws a radical separation between facts and values. That separation means the social sciences (value judgements) do not intrude on the pure sciences (judgements of truth).

This justification carries with it the idea that ethical, legal or gender-aware assessment of AI, and the application of legal rules to AI, will limit and constrain scientific progress. It also gets in the way of training software engineers in the moral impact of designing and programming AI (Belloso, 2023). The debate about whether technology is bad in itself or is merely misused thus becomes beside the point: the focus has been placed solely on technology (the design, the goals it pursues, the data) being misused, absolving and immunising every earlier stage. That is why, in the case of AI, people speak of the myth of the fair algorithm (Belloso, 2023). It is time to look at these biases. It is time to apply guiding principles, values beyond progress and wellbeing. It is time to acknowledge that the separation between facts and values is only wishful thinking. Artificial Intelligence and the technological innovations derived from it are not neutral: they discriminate, and they do so precisely because of the gender bias in the data they handle and the bias in their design.

II

Why exactly is there gender bias in AI? To answer that, we need to look at the digital gender gap, both in access to technology and in digital skills. European Union figures from 2014 show an 8-point gender gap, and according to Emerj data from 2017 only 13% of positions of power in technology companies working on Artificial Intelligence are held by women (Rodríguez, 2021). Computing in schools is, along with mathematics, literature and sport, one of the subjects with the widest gender inequality between male and female pupils, with boys outperforming girls. STEM subjects (science, technology, engineering and mathematics) are rarely chosen by teenage girls as university degrees, leaving women under-represented in those sectors.

This male dominance of STEM fields feeds a lack of concern for women's issues, a lack of research from a gender perspective, stereotyped representations of women, and the sexist language reproduced in the design and in the data these systems use.

It is also worth noting that the vulnerable groups in our society are at greater risk from algorithmic bias, so that the data ends up reinforcing their own marginalisation (Balmaceda, 2021).

III

Exposing AI algorithms to human interaction has the effect of reproducing what it takes to be representative of society - namely, discrimination on grounds of gender. Take the Microsoft Twitter bot Tay, which learned from its conversations with users and had to be switched off by the company within 24 hours because of the sexist and racist behaviour it had picked up from its human interlocutors (Rodríguez, 2021). Although several countries are trying to use Artificial Intelligence precisely to root out gender bias, technology companies carry out no systematic assessment to actively look for such bias in AI, root it out and correct the design and the data input these systems use.

It is worth noting that the prejudices that shape the results algorithms produce are a problem at the prescriptive rather than the descriptive level. That is, taking the well-known case of Amazon's recruitment of women and how they did not fit the profile of past candidates (men) (Rodríguez, 2021), algorithms reproduce what happens rather than what ought to happen.

Some of the consequences of gender bias in Artificial Intelligence are these: difficulty finding jobs traditionally coded as male, and chatbots and voice assistants that embody roles assigned to women, such as service and obedience, among others.

IV

The solutions necessarily involve bringing the gender perspective into the digital world, using technology to overcome patriarchy rather than sustain it. To that end, cyberfeminism - a term coined in 1994 by the philosopher Sadie Plant in order to “theorise, criticise and exploit the internet” (Vita Activa, 2021) - must focus on every area where sexism persists in technological processes (Mujeres en Red, 1997).

In the particular case of Artificial Intelligence, people have for some years spoken of algorithmic justice o fairness (Belloso, 2023), whereby technological innovations must meet ethical and legal criteria such as beneficence, non-maleficence, autonomy, justice, intelligibility and accountability. There must also be incentives for girls to choose STEM degrees and to rise into leadership and positions of power in technology companies, along with a gender perspective in the methodology, design and monitoring of AI, sex-disaggregated data and far more data representing women.

In short, there must be interdisciplinarity between the social and the pure sciences, so that scientific output in general is properly assessed and cared for . We cannot let our guard down on the reproduction of gender inequality on digital platforms.

Written by: Ana E. Marcos

Edited by: Nuria de Pablo Sánchez

REFERENCES
  • Balmaceda, T. et al. (2021): “Bajo observación: Inteligencia artificial, reconocimiento facial y sesgos”, Artefactos, Revista de estudios de la ciencia y de la tecnología, Vol. 10 (2): 21-43. Available at https://revistas.usal.es/cinco/index.php/artefactos/article/view/25057/26997
  • Galloway, A. (1997): “Un informe sobre ciberfeminismo. Sadie Plant y VNS Matrix: análisis comparativo”, Mujeres en Red: El Periódico Feminista. Available at https://www.mujeresenred.net/spip.php?article1531
  • Vita Activa (2021): “Ciberfeminismo y tecnofeminismo”. Available at https://vita-activa.org/ciberfeminismo-y-tecnofeminismo/