Bias-Free Algorithms for Better Innovation

In recent years there has been a growing interest in integrating the variables of gender and sex in research from different disciplines, including business. However, the inclusion of these variables are still ignored today, which may lead to high costs at the monetary level and, on various occasions, may put human lives at risk.

What is the difference between the terms sex and gender?

Sex refers to the biological attributes of people, including physical characteristics, chromosomes, genes, and anatomy. On the other hand, gender is a multidimensional psychological concept, which includes social-constructed roles and behaviors, attitudes, and lifestyles that identify girls, women, boys, men, and other gender identification persons.

In 2016 in the United States, 1.4 million people (0.6% of the population) identified as transgender, according to a study conducted by The Williams Institute. In 2020, only 18 countries (including Argentina, Germany, Austria, Australia, and India) allowed a third category of sex in legal documents such as birth certificates, passports, etc.

Why should companies include sex and gender in their R&D departments?

Innovations, especially in Artificial Intelligence (AI), influence many aspects of our society. From algorithms that determine who to hire or fire to the inclusion of robots in our daily lives (such as Siri and Alexa). These may be biased if fundamental variables are not considered in their algorithms, resulting in the discrimination of certain groups.

Research has demonstrated that businesses have failed when developing their products, brands, or innovations, especially with women and people of color. This is illustrated with the following examples:

• •The word embedding tool Word2Vec, which captures word associations, was found to be programmed with gender bias in its analogie “Man is to Computer Programmer as Women is to Homemaker". What is the reason? Some artificial intelligence algorithms have biased information in their databases.
• According to research, males are five times more likely than women to be shown Google Ads offering executive-level positions when conducting a Google search. What is the reason? The bias is in the database. Google's algorithm is configured to serve ads to the right person. Therefore, the algorithm assumes that men have a better salary and are best suited for executive positions.
• Some heart rate monitors, including some versions of the Apple Watch, do not work for people with dark skin color, which may put their health at risk. What is the reason? The technology that measures blood flow is absorbed by the pigment in the skin of dark-colored people, which reduces the accuracy of the sensors that measure heart rate.
• In 2009, Nikon developed anti-flicker software. However, the system was not able to identify whether an Asian person was blinking or not. Instead, it just prompted up with the message: "Is someone blinking?". What is the reason? The software had biases with the images of Asians, as it reads them as if they are continuously blinking.

What do these examples have in common?

At the time of their development, these innovations focused on only one or two dimensions of intersectionality: gender, ethnicity, sexual preference, etc.

Intersectionality describes how race, gender, disability, age, and other social categories are mutually shaped and interrelated. This is best illustrated with the Gender Shades facial recognition experiment, developed at MIT. They evaluate bias in automated facial analysis algorithms and datasets concerning phenotypes results. The research shows that the system performed better on men's faces than women's (gender) and that darker-skinned females are the most misclassified group (intersectionality).

What is the solution?

More exact science is developed when integrating variables such as sex, gender, and intersectionality in R&D, innovations, and knowledge creation. As a consequence, better research methodologies are developed, and social inclusion and diversity are promoted. The benefit of this approach also encourages economic development.

To accomplish so, interdisciplinary teams must be formed, including software engineers, lawyers, and specialists in diversity, gender, and social policy, among others. They need to determine how to measure and what dimensions of sex, gender, and intersectionality are applied to the innovations.  Furthermore, the team needs to conduct an analysis and a report of each variable to make more inclusive decisions.

Las autoras son alumna del Doctorado en Ciencias Administrativas de EGADE Business School (Claudia Gómez) y alumna del Doctorado en Ciencias Administrativas de la Universidad de Curvinus de Budapest (Anna Törok).

REFERENCES

Bolukbasi, T., Chang, K. W., Zou, J., Saligrama, V., & Kalai, A. (2016). Man is to computer programmer as woman is to homemaker? debiasing word embeddings. arXiv preprint arXiv:1607.06520.

Buolamwini, J., & Gebru, T. (2018 ). Gender shades: Intersectional accuracy disparities in commercial gender classification. In Conference on fairness, accountability and transparency  77-91.

Datta, A., Tschantz, M. C., & Datta, A. (2015). Automated experiments on ad privacy settings: A tale of opacity, choice, and discrimination. Proceedings on Privacy Enhancing Technologies, 92-112.

Flores, A. R., Herman, J.L., Gates, G.G. & Brown T.N (2016). How many adults identify as transgender in the United States? Williams Institute. https://williamsinstitute.law.ucla.edu/publications/trans-adults-united-states/. 

Schiebinger, L. , Tannenbaum C., Falk-Krzesinsk, H.F &Miles, J. (2021) How to integrate sex, gender, and intersectional analysis into research Elsevier. https://researcheracademy.elsevier.com/communicating-research/sustainable-development-goals-researchers/integrate-sex-gender

Rice, C., Harrison, E., & Friedman, M. (2019). Doing Justice to Intersectionality in Research. Cultural Studies, Critical Methodologies, 19(6), 409–420.