Towards Gender Parity in AI Development and Deployment in MENA
2024-05-22
2024-05-22
By: Aliah Yacoub
In the rapidly evolving ‘Responsible AI’ landscape, the intersection of gender and artificial intelligence (AI) stands as a crucial focal point. As the world grapples to navigate the complexities of responsible AI development and deployment, it becomes evident that achieving gender parity in tech ensures that AI technologies are not only socially and ethically sound, but also technically proficient. Today, women remain underrepresented in every stage of tech production: from the theoretical to the technical. This is more pronounced in AI, where gender inequality has far-reaching implications and very real, possibly dangerous repercussions. Through the lens of Data Feminism, this blog post delves into the current state of gender representation in AI within the MENA region, exploring its technical ramifications, societal challenges, mitigation initiatives, and pathways forward.
The Statistics: Gender imparity in tech
Over the past two years, the global representation of women in STEM has increased from 20% to 29.2%.[1] Although this uptick underscores a worldwide commitment to tackling gender inequality in STEM, this positive trend has not been observed everywhere in the world, particularly in MENA. A McKinsey report revealed that women’s labor force participation in the Middle East is the lowest in the world at 24.6%[2]—half of the global average, despite the fact that female university graduates now outnumber men in some countries.
This is especially true in Egypt, where studies show that as of 2023, women made up 48.8% of the total number of students enrolled in both public & private universities, and accounted for 36.9% of those in STEM disciplines.[3] Yet, despite the more equitable representation of women in STEM education, these statistics do not carry over into the workforce. In 2020, women made up only 38% of those employed in STEM[4] and in 2023, they made up only 15.7% of the employable Egyptian labor force – one of the lowest recorded numbers over the past two decades.[5]
In light of these statistics, two questions arise. First, what are some of the factors contributing to the gender disparity in the Egyptian STEM workforce, despite the more equitable representation in STEM education? A survey of the literature shows that the main reasons as to why the transition of women into the tech workforce continues to lag behind include: binary logic and sociocultural stereotypes, structural pipeline challenges, financial exclusion, inaccessibility, and rampant workplace bias.
Although these issues necessitate a study of their own/in their own right, what this blogpost is more concerned with is the second question. Namely, why is the underrepresentation of women in every stage of tech production – from the theoretical to the technical – a problem for datasets, for organizations, and for society as a whole? And ultimately, what mitigation initiatives are underway in MENA?
The Ramifications: through the lens of data feminism
The underrepresentation of women in AI has multifaceted implications, from the conceptualization stage to the deployment stage.
Starting at the beginning, AI is the supposed replication or approximation of human intelligence. However, the very theory of ‘intelligence’ and the epistemology operationalized by dominant AI research has been focusing only on a specific form of knowing—a male form. Work in feminist epistemology[6] has shown that gender affects our acquisition and practice of ‘knowledge’ and that focusing on ‘male intelligence’ in AI effectively excludes female epistemologies. A now-famous example of the reification of gendered and racialized conceptualizations of ‘intelligence’ is virtual assistants (VA). Siri or Alexa, although feminized in name, are designed to ‘think like a man’ – it being the VA’s default epistemology.
The relationship between gendered epistemology and AI development intersects with work in Data Feminism, which emphasizes the importance of recognizing the situated knowledge and experiences of diverse individuals and how it impacts AI development. At the core of Data Feminism is the belief that contemporary injustices – such as gender inequity, racial profiling, and class segregation – are all deeply enmeshed “in historical and contemporary differentials of power”[7], and that a passive approach to technology and its development reproduces these inequalities into our modern technologies and, by extension, societies.
Through the lens of Data Feminism, we can move beyond the design and theoretical conceptualization stage to explore the persistent underrepresentation of women throughout the rest of the AI pipeline. Let’s briefly focus on two interlinked elements in the development stage in which the repercussions of women’s erasure are made clear: the organization setup and the dataset.
Generally, if datasets used to train AI models are not diverse and inclusive, the resulting algorithms may be inaccurate, and may not represent or serve the needs of women and other underrepresented or marginalized groups. After all, biases and assumptions can permeate algorithmic decision-making processes, often reflecting the perspectives of those who hold power in the industry—predominantly men. Models that aren’t inclusive – for example, trained on faces of white people and therefore incapable of recognizing POC or veiled women – reflect, and often even amplify, structural inequalities present in society. Ultimately, this means that your ‘product’ will not work as well as it should.
Representation in the workplace presents a way to mitigate this issue, as bringing in diverse voices and stakeholders always ensures a more inclusive output. The lack of diversity in AI development teams can lead to biased algorithms and technologies that fail to account for the diverse needs and experiences of all users, potentially exacerbating existing inequalities in society. Underrepresentation also creates what researchers refer to as the “discriminatory feedback loop,” where gender inequity in the workplace results in inequitable access to resources, the development of inequitable tools, and ultimately the creation of discriminatory products that benefit those they were built by and for.
To put it quite simply, women’s underrepresentation in tech is not just a ‘social justice’ issue, it’s actually a problem for your dataset, your business, and the whole of society. By embracing gender diversity in AI development, we can work towards creating AI systems that are not only technically robust but also ethically sound and socially just.
The MENA-based Solutions
Despite the challenges, the MENA region has witnessed a remarkable upsurge in endeavors aimed at addressing gender disparity in the tech sector. These initiatives can be delineated into four key domains: venture capital and investment, Femtech, educational enhancement and upskilling, and data-centric initiatives targeted at curtailing technical biases in AI outcomes.
The scale and support behind these initiatives vary widely based on access to capital, the backing of government and corporates, presence of adequate regulatory frameworks, and ecosystem support. Nevertheless, a unifying thread among all these initiatives is a clear commitment to fostering a future characterized by gender-inclusive tech innovation. Now, let’s explore a few exemplary initiatives.
A Wamda report showed that in 2022, out of $622 million raised by startups in the MENA region, less than $6 million was invested in startups founded solely by women – amounting to less than one percent of the total. According to WEF, women-founded startups received 2% of the funding the following year. The uptick is promising, and is in part due to initiatives such as the following.
Leading the path to equitable investment is Organon, a global healthcare company dedicated to improving women’s health, which recently partnered with Flat6Labs, MENA’s leading seed and early-stage venture capital firm, and announced the launch of the second cycle of the Women’s Health Accelerator Program. The initiative aims to empower digital health startups with solutions to enhance women’s health across the Middle East, North Africa, and Turkey (MENAT) region. Similarly, campaigns such as the Womenpreneur Tour, spanning countries like Jordan, Morocco, and Tunisia, serve as catalysts for promoting quality investments in women in tech. By advocating for enhanced financial and social capital, equitable access to resources, mentorship, and policy formulation, these initiatives foster an environment conducive to women’s flourishing in the tech sphere.
Use case-specific Femtech initiatives have also gained considerable momentum in recent years. Studies show that as of last year, the UAE hosted a one-third of femtech companies in MENA, such as MyLily, Lizzom, Nabta Health and many others. In Egypt, initiatives like Motherbeing and Mumerz exemplify the innovative digital platforms tailored to address women’s unique health and wellness concerns by leveraging AI and data analytics to deliver personalized services to women. In Riyadh, KSA, another great example is Cura, a provider of telemedicine and medical consultation services. The Lebanese Siira, a start-up that developed a mental health platform designed to accompany women in their daily struggles related to parenting, relationships and work, is also a great example of the work done at the intersection between gender, tech, and mental wellbeing.
Beyond the concerted efforts to support and invest in female-founded tech startups, as well as fostering product development and market viability in the booming femtech sector, other initiatives focus on the tech itself. The underrepresentation of women in the data sets used to train AI is arguably one of the biggest challenges, resulting in AI systems that are biased against women and that perpetuate gender inequality. The production of gender-inclusive tech is not an easy feat, but a beacon of hope and innovation in tackling gender biases and fostering inclusivity in AI is found at Dubai-based Mullenlowe MENA. After recognizing the consistent pattern of gender bias that exists in AI representation, Mullenlowe focuses on AI-led corrective measures, to rectify gender bias ingrained in AI systems’ datasets, striving for more gender-neutral representations. Similarly, in Saudi Arabia, King Abdullah University of Science and Technology launched “Dear AI” which tackles mis/under-representation of women in AI software by retraining the algorithms on appropriately selected, unbiased data sets.
In Egypt, many local companies attempt to fix misrepresentation in datasets while concurrently developing their products. One such example is the Azka Vision team at Synapse Analytics. The team encountered a problem where veiled women were misclassified and misgendered as men due to biases in publicly available datasets, which primarily featured non-veiled individuals. This issue mirrored broader biases in datasets, such as FairFace, where class representation was imbalanced, particularly regarding gender and race. To address this, a simple rebalance of class representation was introduced in FairFace and mitigated the issue, thereby improving accuracy in gender classification.
The cross-sector educational and upskilling efforts are also witnessing a considerable increase. In the United Arab Emirates (UAE), one such effort is the UAE National Program for Coders, a project aimed at educating 100 Emirati women in AI technology, encompassing cybersecurity. Similarly, the Dubai Business Women Council (DBWC) and Oracle have joined forces and launched ‘sAIdaty’: an initiative that aims to bolster AI-related skills among female professionals and entrepreneurs in the UAE and beyond. Sirius Labs in the UAE is also proactively addressing the underrepresentation of women in tech by offering AI-driven training apps, equipping women with the requisite skills for promising career trajectories. Another great example of educational initiatives is set by the SheCodes Foundation, which is teaching thousands of women in several MENA countries — most notably in Morocco and Tunisia – to code for free.
The Pathways Forward in MENA
Despite the strides made, the journey towards gender equality in tech and AI remains an ongoing endeavor. It demands sustained collaboration between public and private sectors, academia, civil society, and grassroots movements to dismantle systemic barriers and cultivate an inclusive ecosystem where women thrive and contribute meaningfully to the technological landscape.
As we chart a course towards gender parity in AI development and deployment in MENA, it is imperative to amplify the voices of women in tech and dismantle the barriers that impede their progress. Through targeted interventions, such as mentorship programs, policy formulation, and investments in female-led startups, we can create an inclusive tech ecosystem that harnesses the full potential of diverse perspectives.
Finally, embracing the principles of Data Feminism is essential in ensuring that AI technologies are not only socially and ethically sound but also technically proficient. By challenging existing power structures and advocating for data equality, we can pave the way for a more equitable future in tech.
[1] “Global Gender Gap Report 2023.” World Economic Forum, June 20, 2023. https://www.weforum.org/publications/global-gender-gap-report-2023/in-full/gender-gaps-in-the-workforce/
[2] ‘Women at Work: Job opportunities in the Middle East set to double with the Fourth Industrial Revolution.’ Chiara Marcati and Rima Assi, Mckinsey & Company, 2020.
[3] CAPMAS. (2023) Annual Bulletin of Students Enrolled, Teaching Staff, Higher Education. Retrieved from https://www.capmas.gov.eg/Pages/Publications.aspx?page_id=5104&YearID=23350
[4] CAPMAS. (2020b). Quarterly Bulletin: Labour Force Survey. Retrieved from https://www.capmas.gov.eg/Pages/Publications.aspx?page_id=5106&Year=16603
[5] CAPMAS. (2023b) Quarterly bulletin: Labour force survey (Analytical report)
[6] According to the Standford Encyclopedia of Philosophy, feminist epistemology ‘identifies how dominant conceptions and practices of knowledge attribution, acquisition, and justification disadvantage women and other subordinated groups, and strives to reform them to serve the interests of these groups.’ Most petinently here, feminist epistemologists argue that ‘dominant knowledge practices disadvantage women by (1) excluding them from inquiry, (2) denying them epistemic authority, (3) denigrating “feminine” cognitive styles”. https://plato.stanford.edu/entries/feminism-epistemology/
[7] Catherine D’Ignazio and Lauren Klein (2020). Introduction: Why Data Science Needs Feminism. In Data Feminism. Retrieved from https://data-feminism.mitpress.mit.edu/pub/frfa9szd