South Africa's New Government Chatbot Plagued by Flaws and Misinformation

South Africa's Government Communication and Information System (GCIS) recently launched "RSA Facts," a chatbot intended to combat misinformation and provide reliable government information. However, within days of its October 28 debut on X, the chatbot was widely criticized by developers for its shoddy technical implementation, including being hosted in the US on a third-party platform with a "Testing" title and public connection details, raising concerns about data sovereignty and procurement.

Further testing by the non-profit news site GroundUp revealed significant functional failures. The chatbot generated a fake press release about a social grant increase, provided unsubstantiated one-word verdicts on political parties just before elections, and incorrectly attributed its creation to municipal officials. While it did correctly debunk some false claims, its overall performance highlighted a critical inability to handle complex or probing questions, often producing misleading or fabricated responses.

This incident marks the second high-profile AI-related embarrassment for the South African government this year, following the withdrawal of its Draft National AI Policy due to fabricated sources. The repeated failures underscore a broader challenge in effectively integrating AI into public services, despite other departments, like Public Service and Administration, moving forward with their own AI initiatives such as the Batho Pele AI Chatbot for civil servants.

The swift exposure of the chatbot's flaws by critics and media outlets has sparked public debate on the responsible development and deployment of AI in government. It emphasizes the need for rigorous testing, robust procurement processes, and clear guidelines to prevent AI tools from inadvertently spreading misinformation, especially in politically sensitive contexts.