Fake citations have caused disruption once more. South Africa’s aspirations to lead in AI innovation across the continent have encountered a significant setback: the draft national AI policy was retracted due to discovering fictitious, seemingly AI-generated citations. According to Reuters, the policy was nearing completion in parliament when fabricated references appeared in the citations list. Communications Minister Solly Malatsi revealed the withdrawal on April 26, describing the oversight as a critical compromise of the policy’s credibility.
“This mishap is not just a technical glitch but undermines the policy’s integrity,” Malatsi explained. “It seems AI-generated citations were involved without proper checks. This highlights the necessity for rigorous human oversight in AI applications.”
AI-generated citations, particularly in legal documents, continue to be an ongoing issue, with several U.S. lawyers facing reprimands for submitting legal briefs peppered with inaccuracies. Damien Charlotin, a lawyer and data scientist, maintains an online database of legal AI hallucinations, recording over 900 instances in the U.S. alone, plus four in South Africa, excluding this latest incident.
The retracted policy proposed establishing a national AI commission, an ethics board, and a regulatory authority, as well as offering tax incentives, grants, and subsidies to drive private-sector investment. Reuters notes South Africa aimed to establish itself as Africa’s foremost AI innovation hub. Malatsi has not indicated when a revised draft will be available.
AI hallucinations persist as a tricky challenge with language models, a fact regularly reported by Mashable.