The Presidency says it is treading carefully as it probes whether artificial intelligence (AI) ChatGPT was used to write one of President Cyril Ramaphosa’s speeches earlier this month. 

Presidency spokesperson Vincent Magwenya said the office was investigating the matter after a social media post alleged that parts of a speech delivered at a department of basic education (DBE) event were developed using AI. 

“The first problem we have is that it is well documented that these tools are not reliable. However, that being the case, we then requested the department of basic education, which submitted the initial draft, to assist us in determining whether those parts of the speech were indeed developed via AI — [and] if so, why, and who was involved — so that the necessary actions can then be taken. 

“The Presidency will not and does not sanction the use of AI in developing content such as the president's speeches. We are slightly concerned that these tools have not been proven to be reliable.”

Magwenya referred to a news report which tested the AI tools against a 2021 state of the nation address, and it produced a result that said 40% of Ramaphosa’s speech may have been developed via AI.

“However, at that time I think ChatGPT was not in existence, so it’s kind of curious how it could make that determination when the entire technology was not in operation at that time. We need to be careful and circumspect in terms of how we approach that issue. 

“We do not want to unfairly taint colleagues at DBE for misrepresenting the content that they provided to us, so we need to balance that there is still a lot of work that is being done to improve the applicability of these tools.

“On the other hand, we do not want to dismiss the entire issue on that basis, so we are checking. We are going to institute an additional layer of quality checks whenever we receive drafts from departments to ensure that we don’t run foul of such issues.” 

TimesLIVE