AI Inequalities in Low-Income Countries

Please find below a description of my MPhil dissertation, written in 2020, with the title:

Towards Inclusive and Sustainable Artificial Intelligence: Case Studies on User-Centric Business Models for Development in Low-Income Countries

While the use of artificial intelligence (AI) for sustainable development is growing rapidly, recent studies indicate that the design and deployment of AI systems pose risks to achieving truly sustainable AI. The current dynamics of the industry, driven by the mantra ‘the more data, the better’, often involve the extraction of data from human lives. This results in, or exacerbates, power asymmetries between those who own and control AI systems and those from whom data are extracted (‘data subjects’). Such mechanisms present significant threats, particularly in the context of low-income countries (LICs). The literature highlights the need for research into alternative paradigms for the data business, as well as approaches that place data subjects at the heart of business operations.

This research explores businesses that position data sources as central stakeholders in their operations - so-called ‘user-centric businesses’. Case studies were conducted to assess the potential of these approaches to respond to the problematic dynamics of the data industry in LICs. Informed by business model theories, a framework was developed to systematically capture the contexts and strategies of user-centric businesses and to guide interviews with founders and employees. Within- and cross-case analyses revealed three key themes across these business model approaches: engagement (direct social value return), participation (indirect social value return), and self-reflection.

Although direct social value returns - such as using data primarily to understand people’s contexts and perspectives, enabling possession or control over data, and ensuring non-identifiability - have the potential to challenge the extractive logic of data markets, findings indicate a persistent tension between delivering services to customers and returning social value to data subjects. Moreover, it remains unclear whether individuals perceive social value returns as sufficient compensation for the extraction of their data. The study also reveals that certain forms of data ownership and participatory practices can offer data subjects opportunities for agency and empowerment. However, significant challenges remain in developing robust and trustworthy systems that provide individuals with data possession and privacy.

Recommendations are offered to researchers, practitioners, and policymakers to support progress towards more inclusive and sustainable AI.