Exploring gender biases in ML and AI academic research through systematic literature review

Abstract

Automated systems that implement Machine learning (ML) and Artificial Intelligence (AI) algorithms present promising solutions to a variety of technological and non-technological issues. Although, industry leaders are rapidly adopting these systems for anything from marketing to national defense operations, these systems are not without flaws. Recently, many of these systems are found to inherit and propagate gender and racial biases that disadvantages the minority population. In this paper, we analyze academic publications in the area of gender biases in ML and AI algorithms thus outlining different themes, mitigation and detection methods explored through research in this topic. Through a detailed analysis of N = 120 papers, we map the current research landscape on gender specific biases present in ML and AI assisted automated systems. We further point out the aspects of ML/AI gender biases research that are less explored and require more attention. Mainly we focus on the lack of user studies and inclusivity in this field of study. We also shed some light into the gender bias issue as experienced by the algorithm designers. In conclusion, in this paper we provide a holistic view of the breadth of studies conducted in the field of exploring, detecting and mitigating gender biases in ML and AI systems and, a future direction for the studies to take in order to provide a fair and accessible ML and AI systems to all users.

Gender diversity management in NATO for sustainable security and peace

“…A widespread, systemic, and implicit gender bias exists in most of the fields in which AI-ML is widely used, including search and ranking algorithms; systems of online recommendations; robotics; Natural Language Processing (NLP); and automated decision-support systems that are used in social programmes, national defence, justice, medicine and healthcare, and policing. They are programmed with a binary concept of gender which does not reflect the real world, and completely ignores the complexity of identities, most common among the Y-and Z-generation members (Shrestha and Das, 2022). Algorithmic unfairness has been starkly evidenced in the groundbreaking paper "Gender shades: intersectional accuracy disparities in commercial gender classification" (Buolamwini and Gebru, 2018), which demonstrated that facial recognition systems were more than 30% inaccurate at classifying the faces of women of colour but had the greatest accuracy for white men.…”

“…However, in reality, cyberspace is forged by the intersectionality of age, economic status, culture, gender, and severely biased algorithms. While many gender-bias detection and mitigation methods have been proposed in the literature, they are not widely applied nor are the ethical and legal aspects discussed (Shrestha and Das, 2022).…”

AAPM task group report 273: Recommendations on best practices for AI and machine learning for computer‐aided diagnosis in medical imaging

“…A potentially significant, yet subtle, consequence of improper data collection might be an algorithm that performs poorly for certain subgroups or subpopulations with the targeted disease or condition as a result of under‐representation of those subgroups in the training set 30,31 . In radiology applications, it is important to be vigilant so that training/validation dataset selection incorporates safeguards to minimize underlying distortions for under‐represented and/or vulnerable populations and so that already‐existing health‐care inequities are not perpetuated or exacerbated 27,32–34 …”

Artificial Intelligence and User Experience in reciprocity: Contributions and state of the art

“…Concerning fairness of AI, Shrestha and Das point out that although, the relevant discussion within the realm of ML and AI is a recent development, fairness problems and discrimination have roots within human society where the unfair treatment toward the minority is documented in the data that has been created over time and therefore, the inadvertent learning and perpetuation of implicit biases is a longstanding issue in ML/AI systems due to historical biases in the data we have accumulated, which can lead to discriminatory outcomes such as an advertisement algorithm showing more high-paying technical jobs to men than women [190].…”