Gender Inclusivity in AI Governance

Introduction

The rapid advancement of artificial intelligence has significant implications for how societies are shaped and governed, and it is essential that these transformative technologies are developed and deployed in an inclusive and equitable manner. In the African context, where historical and structural inequalities have long persisted, the role of artificial intelligence (AI) in addressing challenges and creating opportunities must be carefully examined through a gender-sensitive lens.¹

Gender inequality persists in various fields, including but not limited to fields within technology such as AI.² Having a more inclusive pool of individuals working on datasets utilised for training AI models who are aware of the discrepancies surrounding AI on a sociological level has been described as key to addressing the inherent biases that are translated from humans to AI.³ The predictive nature of algorithms in converting datasets into patterns, alongside the underrepresentation of women in the developmental phase of AI systems, contributes significantly to the biases that manifest within them.⁴ AI systems globally have been testamentary to the gender biases embedded within various sectors such as employment, financial services, gender-based violence (GBV) and femicide detection, administrative services, education, healthcare, media, search engines and cyberspaces.⁵ Perpetuation of bias has been widely identified as a consequence of such AI systems, however beyond this, the methods through which such systems are eventually tested, also known as ‘benchmarks’, have equally been identified as contributory to perpetuation.⁶

The intersection of gender bias and algorithmic bias

Broadly, three main factors from social, institutional, design and implementation angles are identified as contributory to the gender biases in AI systems. These include gender stereotyping, lack of AI regulations, biased training datasets, lack of diversity in developers, AI’s nature amplifying the bias, and other contextual factors.⁷ A racial factor to such biases is also indicative of how specific groups of women, such as Black women, are affected based on stereotypical factors as outlined in a study.⁸ The study involved identifying biases in LLMs through five distinct scenarios-how a purchase is conducted, predicting the winner of a chess match, predicting the winner of an election, identifying the best athlete and employment-related queries.⁹ By applying different contexts and names that can be closely linked to specific groups of people racially or through gender, the responses LLMs gave in terms of advice, general perceptions and outcomes were evaluated.¹⁰ Consequently, it was found that biases that were most disadvantageous affected Black women.¹¹

The need to go beyond a technical angle to such biases and tap into a socio-centric view has been heavily emphasised with the intent to promote ethical considerations, debunk stereotypes and create more inclusive AI systems.¹² Kong’s study on intersectionality reveals that discrimination faced by coloured women does not necessarily stem from their gender or race but rather their diffused identity of being both.¹³ The root cause of such discrimination yet again falls on the attributes of datasets themselves being male and white, the biases within crowd workers who source datasets and overall structural injustices affecting women of colour as evidenced by various systems such as face-recognition algorithms.¹⁴

The case for developing countries

In developing countries, gender inequality persists against women with respect to AI, especially due to legal systems that are inherently unfavourable to women with respect to their right to privacy/data protection, hiring systems, as well as inaccessibility to education and technology.¹⁵ Other factors that contribute to gender bias include AI systems’ ability to engage in GBV and harassment as well its potential to replace women in the job market.¹⁶

‘Colonial models of AI’-Formulating AI systems with little to no contextual considerations in Africa have been described as a component that hinders the inclusion of diverse, marginalised, and underrepresented groups in processes where participation is key.¹⁷A closer look at the African continent reflects similar biases in gender due to factors such as accessibility and inclusion in the developmental and participation processes of AI systems.¹⁸ The lack of diversity and representation in the field of AI development¹⁹ has been well-documented.²⁰ Women, particularly those from the global south, face significant barriers in accessing necessary education, training, and opportunities to participate in shaping these powerful technologies. This disparity has limited the ability of AI applications to be effective within African contexts, as the tools developed often fail to account for the needs and experiences of marginalised communities.²¹ Additionally, the impact such biases have on African women trickle down to their privacy and security due to contextual differences between AI developers and marginalised groups.²² Global power imbalances and politicised effects of technological advancement are also identified as influential in the systemic injustice that marginalised groups such as women face on a policy scale.²³ This has been equally expressed in the space of AI education.²⁴

Women in Africa have been identified as subjected to inequality economically, through education, the digital divide, from an entrepreneur angle, participation in science, technology, engineering and mathematics (STEM) as a field, as well as leadership.²⁵

Datasphere-described as almost a collation of datasets and its relationship with society- has been highlighted as a concept that can mitigate gender bias within AI.²⁶ In ensuring that data is sourced, processed, handled and protected, recognizing gender bias and establishing measures to this effect, there is a potential for gender bias to be mitigated.²⁷ African countries such as Kenya’s efforts to re-train its language models, Nigeria in establishing rules to promote responsible AI governance and South Africa’s audits to ensure that healthcare systems are not disproportionately impacted by gender are evidentiary of African countries’ efforts towards addressing gender bias in AI.²⁸

The African Observatory reaffirmed such biases in the country on Responsible AI’s representatives who expressed the need to incentivize women’s participation on all fronts given that the issues AI seeks to address primarily affect women.²⁹ Carnegie Mellon University Africa’s representatives equally spoke to the positive impact that such integration would have socioeconomically through AI.³⁰ Aside from participation and policy-making efforts, the need to deconstruct biases on a societal level was equally highlighted in the OECD-AU AI Governance Dialogue.³¹

In terms of policymaking, the need to preserve gender neutrality has often been highlighted as key to mitigating gender bias in Africa.³² However, the ability to foresee biases and outcomes resulting from such biases in data processing has been expressed to be a limitation.³³ Emphasis has equally been made to having multi-sectoral inclusivity of women in developing AI to ensure systemic change, be it the development of the system itself or external factors surrounding regulation/governance.³⁴Aligned with the ambition of promoting inclusivity on all facets stands the AU’s strategy on Gender Equality and Women’s Empowerment (GEWE), which has key pillars aimed at promoting education, tackling GBV/domestic violence, enhancing enforcement through regulation and overall participation of women in the decision-making process.³⁵The necessity for AU member states to ratify the Maputo Protocol on Women’s Rights has been highlighted as critical to protecting women’s interests in the technology sector whilst promoting gender-specific programs to provide women education on AI and its impact across sectors.³⁶

Recommendations

Addressing these challenges requires a comprehensive and multifaceted approach that prioritises gender equity and inclusivity. Governments, private sector actors, and civil society organisations must work collaboratively to invest in STEM education and training for women and girls, ensuring that the pipeline of talent for AI development is diverse and representative.³⁷ Additionally, policies and regulations must be put in place to mitigate the risk of algorithmic bias and to ensure that AI-driven solutions are designed with the needs of marginalised communities in mind.³⁸

International collaborations and knowledge-sharing can also play a crucial role in fostering a distinct African AI ecosystem that is responsive to local contexts and priorities. By embracing a gender-inclusive approach to AI governance, African nations can harness the transformative potential of these technologies to drive sustainable development, improve service delivery, and empower historically underrepresented groups.³⁹ Through strategic partnerships and exchange of best practices, African stakeholders can learn from global experiences, adapt these learnings to their local realities, and develop innovative AI-driven solutions that address the unique needs and challenges of the region.⁴⁰ This collaborative approach can help build local capacity, foster innovation, and ensure that the benefits of AI are equitably distributed across diverse communities. By prioritising gender inclusivity in the development and deployment of AI in Africa, these technologies can become powerful enablers of social and economic progress, empowering women and other marginalised groups to participate fully in the digital transformation of the continent.⁴¹

Simultaneously, policymakers must develop comprehensive regulatory frameworks that mandate the rigorous consideration of gender and intersectional impacts throughout the entire lifecycle of AI systems – from design and development to deployment and governance.⁴² Such frameworks should ensure that marginalised communities, particularly women and gender minorities, are actively involved in shaping AI initiatives’ priorities, processes, and outcomes, thereby promoting more inclusive and equitable AI-driven solutions.

Additionally, the development of AI-powered solutions must be grounded in a deep understanding of the local context, drawing on insights and lived experiences of marginalised communities. This will not only enhance the effectiveness of these technologies but also contribute to a more equitable and inclusive AI ecosystem in Africa. Engaging with marginalised communities, including women, ethnic minorities, and people with disabilities, to understand their unique needs, challenges, and perspectives is crucial for designing AI systems that are responsive to the diverse realities and priorities of the African continent. By actively involving these stakeholders in the design, development, and deployment of AI-driven solutions, policymakers and technology developers can ensure that the benefits of AI are equitably distributed and that no one is left behind in the digital transformation of Africa.⁴³

Conclusion

Embracing a gender-inclusive approach to AI governance in Africa is crucial for harnessing the transformative potential of these technologies to drive sustainable development, improve service delivery, and empower historically underrepresented groups. By prioritising diversity, equity, and inclusion throughout the entire lifecycle of AI systems, African nations can build a distinct and responsive AI ecosystem that addresses the unique needs and challenges of the region. Through strategic partnerships, knowledge-sharing, and rigorous policy frameworks, stakeholders can work collaboratively to ensure that the benefits of AI are equitably distributed and that no one is left behind in the continent’s digital transformation.