AI Report on Southern Africa http://misa.org At the Johannesburg meeting, the Google AI assistant was tested to evaluate its effectiveness in understanding and responding to African communities’ unique needs and nuances. We asked if the Google AI assistant could speak Shona (one of Zimbabwe’s main languages). The Google AI assistant response is shown in the screenshot below. Figure 2 Extract of bard AI interaction Despite claiming to be able to speak Shona, the Shona speakers in the room were quick to point out that the translation provided by the Google AI assistant needed to be more accurate and accurately capture the nuances and intricacies of the Shona language. For example, the Shona phrase ‘taurai zvakanaka’ ‘does not mean goodbye; ‘house kuita’ is not Shona for you are welcome. We did more tests using other popular Southern African languages, and the results were similar - the translations needed to be more consistently accurate and capture the true meanings of the phrases. Since AI can only function with data, more high-quality data is needed. The efficacy of machine learning techniques is contingent upon the quality of the data they are provided with. AI algorithms may incorporate biases present in the data or the biases of the individual who developed the algorithm, leading to the propagation of societal inequalities. In Africa, it is particularly crucial to be cautious about adopting machine learning algorithms that have been developed and taught outside of the continent. These algorithms may not accurately represent or have biases against significant portions of the African population (Kathryn Hume 2017). To facilitate the adoption of AI solutions by researchers, developers, and consumers, it is necessary to have a more extensive, comprehensive, and easily available collection of data. In emerging economies, specifically in regions characterised by instability or conflict, the availability and accessibility of accurate data are often limited (Ajadi 2020). If the training data for an AI system does not accurately represent the demographic factors of the intended population, the AI is likely to fail in several instances. To deliver accurate responses to users, a Chatbot system relies on extensive knowledge of its operations. If the user requests information that is not stored in the system’s database, it will be unable to provide a response. 11