Parliament’s oversight committee is unsatisfied with the communication ministry’s presentation on the events that led to the withdrawal of the draft National Artificial Intelligence (AI) Policy and the actions taken by the department to remedy the situation.
Yesterday, the Department of Communications and Digital Technologies (DCDT), led by minister Solly Malatsi, briefed the portfolio committee about the way forward and consequence management processes, a month after pulling the plug on the AI policy.
The department expressed its commitment to finalising South Africa’s revised AI policy this financial year, with a target to open it for public comment in January 2027.
It revealed that an internal investigation, into how a policy featuring AI-hallucinated references was presented for public comment, was still underway. It said it has put together a panel of experts in AI, governance and academia, to fix the fundamental errors.
However, members of Parliament (MPs) questioned the department’s “brief” presentation, saying it fell short of what was required. The MPs felt the presentation did not focus on what really happened, but rather on reactive remedial processes.
They further expressed concerns about the composition of the advisory panel, the selection criteria, additional costs the DCDT would incur as a result, and questioned the selection of six members as opposed to a colloquium involving multiple stakeholders.
They were also concerned about the two suspended officials potentially being used as scapegoats for the department’s lack of oversight.
Answers sought
In his remarks, uMkhonto weSizwe MP Adil Nchabeleng pointed out that AI is pervasive and unavoidable, noting there is nothing, in principle, wrong with the use and reliance on AI.
However, Nchabeleng wanted to know the department’s guidelines in terms of its use.
Malatsi responded by acknowledging the department’s shortcomings, saying it has both a systems challenge and people challenge. “The two things that are critical…where there is use of AI and all its other features, it is incumbent that there is disclosure.
“Secondly, which is where the department fell short, is around internal processes for AI authentication and verification systems. If there aren’t any of those disclosures, because there will be instances where, whether intentionally or unintentionally, people don’t disclose their reliance on AI usage in the formulation of documents and policies, there must be internal guardrails.
“In the absence of these systems, we found ourselves vulnerable and were not able to pick up on these fictitious references.”
On the composition of the panel, Malatsi said their services will be offered pro-bono. The only cost will be that the department would bear the miscellaneous and logistical support costs, including transportation or arranging a venue.
“There won’t be any remuneration going directly to the advisory panel members,” he said. “We are grateful, given their expertise and their roles, that they’ve agreed to play a part in this process at no cost to the department when it comes to remuneration.”
He added: “They [advisory panel] are globally recognised individuals that are prominent in their work on AI governance and AI policy direction. It also does not mean the panel and internal teams in the department responsible for the drafting will be restricted. There’s still going to be engagement with external stakeholders.
“In establishing the panel, we looked at the variety of skills and a balanced approach around both academia, governance, technology and even components of cyber security. We have a manageable team that has a fair resemblance of the broad spectrum of anything related to AI governance and AI views in society that would provide the guidance in terms of the other engagements that will have to take place with the rest of society.”
Shaik Imraan Subrathie, ANC MP and whip of the ANC Study Group on Communications and Digital Technologies, wanted to find out whether AI was used at all in any of the processes of the policy development, and if it was used, if there was human oversight of the AI.
Malatsi answered: “At all the stages, the substance of the document…there had not been any red flags about any undeclared use of AI and on the referencing part at that time. In hindsight, which is where the oversight from my end comes in, is that we did not individually go through the verification of each reference.
“This is also part of the internal investigation that seeks to establish questions. At the various interventions that the ministry had with the policy document, there was no indication whether there was reliance on AI in the formulation of the policy, or non-disclosure around unverified references, which sometimes, during a draft document, there is that disclaimer.”
DCDT director-general Nonkqubela Jordan-Dyani added that when the department reflected on the root cause and the internal engagements with the officials, it has been indicated that ChatGPT was used for the editing of the policy document.
“The issue related to the fictious references, we’ve also been able to locate and pinpoint where it emanated from. This also talks to the ongoing investigation, and the outcomes will be communicated.”
On further details of the officials that have been placed on precautionary suspension, Jordan-Dyani said it does not give the presumption of guilt and asked that the ongoing investigation be concluded first, after which the department will be able to share the outcome.
Share
