Africans’ Contributions to AI Can Reduce Bias

Kenechukwu Agbo 

Artificial Intelligence (AI) has revolutionized numerous industries, promising efficiency and accuracy. However, there is a dark side that exists beneath AI’s gleaming surface: bias against Africans. This bias shows in various AI applications, showing discrimination and reinforcing existing social inequalities. Lots of studies and research have exposed AI’s limitations in recognizing non-white individuals’ images and speech patterns, leading Black AI researchers at tech giants to sound the alarm about potential harm to their community, and highlighting the need for more diverse and inclusive AI development teams.

One primary source of bias is the data used to train AI systems. Since AI learns from data, it inherits the prejudices and imbalances present in the data. Historical datasets often contain limited or distorted representations of African cultures, languages, and experiences. As a result, AI systems develop a skewed understanding of African identities, leading to misidentification, misclassification, and marginalization. Statistics show that Nigeria is the only African country in the top 25 countries of internet users, thereby, leading to a very poor representation of Africans on the global stage. Because artificial intelligence (AI) is trained on publicly available data, there are not enough Africans generating enough data to reduce the bias against the continent.

Facial recognition technology, for instance, has been shown to be less accurate for individuals with darker skin tones, leading to potential misidentification and wrongful arrest. Language processing AI often struggles with African languages, favouring Western languages and limiting access to important services. Large language models (LLMs), on the other hand, have been shown to potentially deliver harmful medical information to Black people. A study published in npg Digital Medicine demonstrates this concern. The study, led by Stanford researchers, was conducted on four commercially available LLMs — Google’s Bard, Anthropic’s Claude, Open AI’s ChatGPT and GPT-4 — and found that they all could cause harm by producing debunked, and racist information. When asked questions about calculating patients’ kidney function and lung capacity, all of the LLMs tried to justify race-based medicine. These two areas are where race-based medicine practices used to be common but have since been scientifically debunked.

Another issue is the lack of diversity among AI developers. AI teams composed mainly of white and Western individuals often overlook African viewpoints, perpetuating a cycle of bias and exclusion, disregarding the unique challenges and experiences of African communities. It is a positive thing that leading AI tech companies are very particular about diversity in their staffing. This will help to bridge the gap in the long run.

The consequences of AI’s bias against Africans have a devastating impact on individuals, communities, and society as a whole. Biased AI systems perpetuate and amplify discrimination in crucial areas like employment, healthcare, and finance, further marginalizing already vulnerable communities and reinforcing existing social and economic inequalities. Moreover, AI-driven disinformation and propaganda can fuel hate speech, xenophobia, and misinformation, threatening social stability, undermining trust in institutions, and potentially leading to violence and conflict. The perpetuation of these biases also reduces the level of trust in AI systems themselves, hindering their potential to drive progress and innovation. It is, therefore, paramount that we take immediate and sustained action to address and overcome these biases, working tirelessly until we bring the bias in Artificial Intelligence to a complete halt, ensuring that AI is a force for good, rather than a tool of oppression.

To effectively combat AI’s bias against Africans, it is essential to adopt a multifaceted approach that addresses the root causes of this bias. First, diverse and representative data sets that reflect the complexity and richness of African experiences must be included in the training set, ensuring that AI systems are exposed to a wide range of perspectives and contexts. This requires a deep understanding of the continent’s history, its present challenges, and its future aspirations. Second, African developers and researchers must be included in AI development teams, bringing their unique insights and expertise to the table. Initiatives like Deep Learning Indaba, which brings together African AI researchers and provides a platform for knowledge sharing and collaboration, are crucial in this regard. By supporting and amplifying such programs, we can ensure that AI is developed in a way that serves the needs of African communities and promotes a more equitable and inclusive future for all.

In conclusion, the bias in AI against Africans is a pressing issue that requires immediate attention. However, it is not a lost cause. Ultimately, AI has the potential to empower Africans, but only if we address and overcome the biases ingrained in these systems. By prioritizing inclusivity and diversity, we can harness AI’s potential to drive African progress and innovation, rather than perpetuating harmful biases. Let us work together to create a future where AI lifts up, rather than holds back, the African continent. The time to act is now.

*Agbo, a Machine Learning engineer, writes from Lagos