Machine Intelligence in Africa: a survey

Allahsera Auguste Tapo,Ali Traore,Sidy Danioko,Hamidou Tembine

2024-02-04

Abstract:In the last 5 years, the availability of large audio datasets in African countries has opened unlimited opportunities to build machine intelligence (MI) technologies that are closer to the people and speak, learn, understand, and do businesses in local languages, including for those who cannot read and write. Unfortunately, these audio datasets are not fully exploited by current MI tools, leaving several Africans out of MI business opportunities. Additionally, many state-of-the-art MI models are not culture-aware, and the ethics of their adoption indexes are questionable. The lack thereof is a major drawback in many applications in Africa. This paper summarizes recent developments in machine intelligence in Africa from a multi-layer multiscale and culture-aware ethics perspective, showcasing MI use cases in 54 African countries through 400 articles on MI research, industry, government actions, as well as uses in art, music, the informal economy, and small businesses in Africa. The survey also opens discussions on the reliability of MI rankings and indexes in the African continent as well as algorithmic definitions of unclear terms used in MI.

Computers and Society,Artificial Intelligence

What problem does this paper attempt to address?

The main problems that this paper attempts to solve are the challenges and opportunities faced by the African continent in the field of machine intelligence (MI). Specifically, the author focuses on the following key issues:

1. **Under - utilization of audio datasets**:
- Although African countries possess rich and large - scale audio datasets, current machine intelligence tools fail to fully tap the potential of these datasets, resulting in many Africans being unable to benefit from the business opportunities brought by machine intelligence.

2. **Lack of cultural awareness**:
- Many of the existing advanced machine intelligence models lack cultural awareness, and their ethical adaptability is also questionable. This deficiency is particularly evident in African application scenarios, as many models fail to take into account the local cultural background and social needs.

3. **Reliability of machine intelligence rankings and indices**:
- The author points out that it is unethical to rank the machine intelligence development of African countries using a single weighted average index, because this method ignores data transparency and specific actions taken by the government, and these factors are crucial for evaluating the actual impact of machine intelligence.

4. **Insufficiency in education and public awareness**:
- The paper emphasizes the importance of enhancing the public's understanding and awareness of machine intelligence, especially among those who are unable to read and write. Through education and training, more technical professionals can be cultivated to promote the development of machine intelligence in Africa.

5. **Multi - objective optimization problems**:
- The author cites the concepts of basic game theory and Pareto optimality, and points out that when multiple interdependent objectives are involved, simple scalarization techniques (i.e., mapping multiple objectives into a single value) are not a good method. For example, the vector \((1,0)\) is not better than \((0,1)\), and vice versa. These vector elements usually represent the well - being of the local population, including monetary, non - monetary, technical and non - technical aspects, which cannot be measured by a single number.

In summary, this paper aims to summarize the development of machine intelligence in Africa in recent years from a multi - level, multi - scale, and culturally - sensitive ethical perspective, and present its application cases in research, industry, government actions, as well as art, music, the informal economy, and small enterprises in 54 African countries. At the same time, the paper also explores the reliability of machine intelligence rankings and indices, and discusses the ambiguous terms in algorithm definitions.