AI for the World, or Just the West? How Researchers Are Addressing Global Gaps in AI Development

Since

The Global AI Power Imbalance

The advent of AI technologies has intensified debates around power dynamics within the tech industry, with many arguing that Big Tech continues to shape the global AI landscape to prioritize the interests of the West. In response to this trend, the Distributed AI Research Institute (DAIR) stands at the forefront of advocating for an AI that is not dictated by major corporations. DAIR is an international group of technologists and researchers dedicated to fostering “independent and community-rooted AI research,” focusing on the needs of marginalized communities rather than those of wealthy multinational firms.

Researchers within DAIR, such as Nyalleng Moorosi and Timnit Gebru, are challenging the status quo by developing Africa-centric AI solutions that aim to address local societal issues. They believe in empowering the very communities that have been historically excluded from AI development, instead of relying on Western-centric models that often miss cultural and regional nuances.

Building AI for African Communities

Nyalleng Moorosi, based in Lesotho, is a key member of DAIR. Her background in machine learning, combined with her experience teaching in South African public schools, has influenced her approach to AI. Moorosi’s time at the University of Forte, a historically significant institution in South Africa, exposed her to the struggles of students living in poverty while pursuing their studies. This experience shaped her belief that AI development should be more inclusive and equitable.

Previously employed at Google’s Africa AI research lab in Ghana, Moorosi began questioning the lack of African representation within the company, particularly after a colleague, Timnit Gebru, reached out to her regarding the topic. Gebru, a former leader on Google’s ethical AI team and DAIR’s founder, shared her concerns about how Big Tech seemed to suppress voices aiming to highlight the negative impacts of technology. This shared vision led them to form DAIR, which centers power within local communities and employs AI to benefit them directly.

The South African Township Study

In 2018, Moorosi, Gebru, and fellow researcher Raesetje Sefala initiated a project to analyze the spatial transformation of South African townships. Townships, historically underdeveloped urban areas with predominantly Black populations, are emblematic of South Africa’s post-apartheid struggles. By collecting satellite imagery and applying machine learning, DAIR sought to track changes in these areas over time, focusing on the social and economic impact of historical segregation.

Their findings faced resistance from predominantly Western academic institutions, which rejected the study as “too niche” despite the use of standard AI methods and computer vision technologies. Moorosi and Gebru argue that this reflects a broader issue: the global AI industry often dismisses research that does not align with Western perspectives, even when it addresses critical issues in African and post-colonial contexts.

Serving Underserved Communities with AI

In addition to DAIR’s work, Asmelash Teka Hadgu, co-founder of Lesan AI and a DAIR fellow, is developing language models for African languages that are underrepresented in current AI systems. Lesan AI focuses on languages like Amharic and Tigrinya, which are often overlooked by Western tech giants. Hadgu explains that, while major companies like OpenAI’s ChatGPT fail to support these languages adequately, his approach is rooted in local community needs, emphasizing the importance of capturing the cultural and linguistic specifics of African populations.

This issue of underrepresentation extends beyond language support. According to a study, over 90% of the data used to train AI models comes from North America and Europe, with Africa contributing a mere 4%. The lack of adequate data for African languages is compounded by the fact that many African languages are not readily available online in digitized formats, unlike English and other Western languages. As a result, AI systems trained on such data are ill-equipped to serve African communities effectively.

The Struggles of Language Representation in AI

One of the critical issues in AI development today is the lack of meaningful representation for low-resource languages. While AI giants focus their efforts on popular languages such as English, the translation and transcription needs of millions of African speakers often go ignored. Despite some efforts from companies like Meta, which is attempting to integrate more African languages, these languages often face challenges in preserving their cultural uniqueness and tonal nuances in AI systems.

Hadgu’s work at Lesan AI aims to provide accurate translations and open up digital resources to these communities. However, he warns that Western corporations’ approach to AI often dismisses these languages as “long-tail,” relegating them to the margins despite being spoken by millions of people. The AI systems built for these languages often fail to capture their cultural specificity, which is essential for providing meaningful solutions.

What Undercode Says:

The shift toward community-based AI development is a crucial one, particularly when examining the historical and ongoing inequalities between the Global South and the West in the tech industry. Big Tech’s focus on profit-driven innovation has led to the exclusion of regions that are not seen as lucrative markets for AI development. However, as we see with DAIR, Lesan AI, and other initiatives, there is a growing movement to create AI systems that cater to local needs rather than global monopolistic interests.

The insistence on “community sovereignty” in data and AI development is a vital step toward bridging this gap. By involving local experts and communities in the development of AI systems, researchers like Hadgu and Moorosi are not only making AI more inclusive but also addressing the underlying power imbalances in the tech world. These initiatives highlight the importance of creating AI models that are rooted in local culture, history, and needs, offering a more diverse and equitable approach to technological advancement.

The role of governments in shaping AI policy will also be critical to addressing these disparities. Several African countries, including South Africa, Ghana, and Tunisia, have already begun drafting national AI strategies, recognizing the importance of equitable access to technology. These frameworks aim to combat the influence of multinational corporations and ensure that AI development benefits local populations, especially those in underserved or rural communities.

In conclusion, the future of AI lies not just in the hands of Silicon Valley giants but also in the innovative efforts of technologists across the globe, particularly those from the Global South. As these researchers continue to challenge the status quo, their work will pave the way for a more inclusive and equitable AI future.

Fact Checker Results:

There is a strong movement within Africa to ensure AI represents local languages, cultures, and needs.
Despite some efforts by Western companies, African languages are severely underrepresented in current AI systems.
African governments are slowly starting to draft AI regulations that aim to limit Big Tech’s influence on the continent.

References:

Reported By: https://www.zdnet.com/article/ai-for-the-world-or-just-the-west-how-researchers-are-tackling-big-techs-global-gaps/
Extra Source Hub:
https://www.github.com
Wikipedia
Undercode AI