The many promises and numerous perils of AI in South Africa's newsrooms

27 Nov 2023

The most promising development for South African newsrooms would be the democratisation of AI model development. What it would mean in practice are laws and regulations that allow access to information for the purposes of learning in combination with data-sharing agreements that favour the public interest.

The impact of foundational AI models on many aspects of life has been a trending topic this year. Associated Press, for example, has revised its style guide to follow AI best practices.

South African journalists are also adapting to this innovation cycle, but with the knowledge their field has acquired from three decades of managing free expression in the digital age. These newsrooms have sought to gradually incorporate AI systems into news production.

In 2020 News24 deployed AI systems for comment moderation and in 2021 for analytics. Recently the company reported that they are looking to use these systems to augment their news production, assist with workplace efficiencies and other aspects of their business. Their efforts highlight how AI will likely affect newsrooms in three ways: coverage, distribution, and financing.

While journalistic AI ethics are starting to create norms and guidelines to help journalists make decisions, we offer some thoughts about how coverage, distribution and financing can be explored.

Coverage

Upfront, chatbot-generated content should never be used as a source. In addition to hallucinations and producing fictionalised information, African expressions and experiences are not part of most AI model’s training sets. Using these tools for seemingly benign tasks like summation could ultimately damage public trust and the reputation of the news media as cultural differences accumulate.

This is why transparency and full disclosure of such usage is an important norm to establish now.

To improve usage, newsrooms can ask critical questions to model providers to develop their own collective understanding of how these models are created. This in-depth knowledge about these AI systems may likely end up becoming a prerequisite for audience trust. And it can then be shared with audiences.

At the same time, AI is more than foundational generative models like OpenAI’s ChatGPT or Google’s PaLM 2, so conversations about this range of technologies can stand to be more imaginative. One promising area where AI can assist is with modern news-gathering techniques, like the use of satellite imagery and map data for geolocation, or forensic video and image analysis.

This type of open data journalism means that newsrooms can be less reliant on using spokespersons as sources, especially when these people are motivated to engage in spin. The same applies to online sources, which can be products of disinformation.

To act on this vision, newsrooms need in-house skill development. Journalists and their managers should learn how to use AI ethically and effectively. This requires skills such as data sourcing, analysis, programming, visualisation, storytelling, verification and even in-house experts in data and AI governance.

Distribution

Newsrooms know about how platforms’ AI recommendation systems play a role in circulating news online. Some newsrooms depend on online audiences and are well aware of how even slight changes to a platform’s algorithm can have cascading effects that disrupt media work.

As they work in different sectors, newsrooms and platforms do not have the same interests or considerations. And so there is going to be a degree of friction as they navigate each other’s interests, assumptions and decision-making.

This does not mean that the near future will be smooth. Newsrooms may need to establish new assumptions like foregoing specific by-lines to have work attributed to named multidisciplinary teams.

Another hurdle is that some AI projects may be unaffordable for newsrooms operating in small markets or in low-resource settings. However, there are open-source products that can assist with some of these items.

Open-source products might not be able to address reasonable, perennial worries about the future of news workers and their employment prospects. The issue of AI in Africa and the fear of losing jobs due to simplified stories and curation is a fair concern. It is vital for democracy that newsrooms do not turn into automated content creation warehouses; rather there is scope to emphasise the importance of creative, original storytelling in journalism.

Read more in Daily Maverick: Catch up or cut a new trail — AI’s expanding role in African journalism

Financing

While existing business models have frayed, there is value in imagining how to rebuild newsrooms so that everyone has access to high-quality, fact-based news.

While it is too early to tell how the entire local news industry adapts to AI, it is reasonable to expect that successful newsrooms would prioritise strategy over technological adoption for its own sake. Based on past behaviour, technology companies will be quick to promise that their products can help newsrooms monetise other aspects of their business, like using their archive as a revenue source.

In light of this, we might want to think about whether AI accountability encompasses a company’s transparency about how product development and investment decisions are made.

We also expect AI to complement existing data journalism initiatives. How to make these kinds of initiatives financially viable remains an open question. Perhaps open data journalism can find new ways of engaging an audience, while AI systems can be used to repackage and spread content on platforms and generate another revenue stream, one with relatively low overheads.

There are open legal and ethical questions about data scraping and copyright infringement that raise issues of data collection, use and abuse, and ownership. Nevertheless, experimenting with new genres and business models is likely to increase in the years ahead. This means that notices that transparently explain how AI was used in the production become normalised. These initiatives can reassure consumers that they are not subject to audience scoring, user surveillance, or data-sharing agreements within the branches of a conglomerate a newsroom may be part of.

Future work

Perhaps the most promising development for South African newsrooms would be the democratisation of AI model development. What it would mean in practice are laws and regulations that allow access to information for the purposes of learning in combination with data-sharing agreements that favour the public interest. When money determines access, inequalities hinder public interest journalism — the kind that is absolutely needed to curb the abuse of power.

What is needed is a comprehensive understanding of the scope, scale, and speed of changes in journalism, and how AI can help or hinder them. Newsrooms also need to explore new ways of engaging with their audiences and stakeholders and creating value for them.

In sum, the ethical use of AI in journalism is vital for preserving credibility, a task made more pressing in the face of reactionary attacks on the free press occurring across the world. DM

Scott Timcke is a Senior Research Associate with Research ICT Africa. He is affiliated with the University of Johannesburg’s Centre for Social Change and the Center for Information, Technology, and Public Life, University of North Carolina at Chapel Hill.

Herman Wasserman is Professor of Journalism and Chair of the Department of Journalism at Stellenbosch University. His books include Tabloid Journalism in South Africa; Media, Geopolitics, and Power; Media, Conflict and Democracy in Africa; and Disinformation in the Global South (with Dani Madrid-Morales). He is a Fellow of the International Communication Association and an elected member of the Academy of Science of South Africa.

Categories:

Tags:

DM is much better if you are signed in...

Maverick Citizen

Coverage

Distribution

Financing

Future work