The Policy Innovation Lab is committed to supporting transformative policymaking using responsible AI. As part of this mission, we have been building a platform for hosting a range of tools that utilise Large Language Models (LLMs). This technology is often underutilised in the public sector due to the risks concerning hallucinations, bias and data privacy in a high-stakes environment. This platform allows South African policymakers to access these tools securely and privately, while the Lab supports their responsible use with regular training, testing and feedback.
This week, the Lab conducted beta-testing of the platform and its tools. The tools currently developed use LLMs to analyse, summarise and compare documents, media and academic research for the purpose of making the policymaking process easy and informed. Members of the Policy and Research Services branch of The Presidency of South African enjoyed testing the different tools on the platform and gave useful feedback to help us improve the tools and ensure that they are fit for purpose.
Their feedback showed a clear understanding of the benefits of such tools, while the need to enhance the user’s ability to dig further into the data was highlighted. Furthermore, the issues of bias and representation when collecting English-only media were emphasised, highlighting the need for high-performance African language LLMs to advance the use of AI in South Africa.
Overall, the event helped the Policy Innovation Lab continue towards its goal of impactful research and user-centric app development.