For news-watching technologists with a vested interest in building tools that support the production of quality journalism, the release of Generating Change: A global survey of what news organisations are doing with AI by Polis at the London School of Economics is important.

The Polis team, including Charlie Beckett, Mira Yaseen and South Africa’s own Tshepo Tshabalala, have long been at the forefront of understanding the impact of machine learning and artificial intelligence on news media; so their study — published less than a year after the launch of ChatGPT — is as revealing and insightful as expected. And, while you should definitely read it, here’s what we believe needs to be understood about the key takeaways for the African context.

It opens with a discussion of what it is not about: and that’s the broader and well-hyped existential dangers and opportunities of generative AI. Instead, it is grounded in real-world experience of what tools are currently being used, and where editors and publishers are expecting the benefits to come from in the future. Many acknowledge the risk that computer-generated articles could trigger a rush to the bottom for editorial quality, in much the same way that clickbait stories did two decades ago. But most have some optimism in the practical application of AI today and going forward.

Yet perhaps even more importantly, the first finding helps to foreshadow the political economy that should be foregrounded before we begin any discussion on the introduction of technology:

“Artificial Intelligence (AI) continues to be unevenly distributed among small and large newsrooms and regionally among Global South and Global North countries.”

Bearing this in mind, we at OpenUp went ahead with our reading of the report strongly conscious of our own agenda: how can we extract the most benefit from this environment for the media partners we work with?

How AI is helping journalists today

The report’s broader findings reflect our own interactions with journalists from across southern Africa. We train journalists in the discovery, extraction, analysis and visualisation of data, and the single most common issue raised is how to get information out of printed documents. Data remains a foundational regional challenge for innovation of all sorts, which is why we have also been exploring Large Language Models with Optical Character Recognition and entity extraction in our own tool, Dexi.

For journalists, the challenge is that as much as the world has gone digital, investigative reporters still receive much leaked information in printed form, and public bodies maintain the obstreperous habit of publishing PDFs rather than Excel sheets for numbers. Polis identified ingesting documents and using optical character recognition to extract information as one of the primary use cases for AI. This doesn’t surprise us.

Indeed, it is revealing that many other use cases may be ones for which journalists may not even know they are using AI at all. Social media analytics and SEO keyword research are common day-to-day uses of AI in the newsroom, but does a writer leaning on Ubersuggest for Google-friendly headlines even know they are using machine learning for assistance?

The future is here, just unevenly reported

Let’s get back to the most important issues that the report raises, however: the inequalities inherent in AI implementation. As might be expected, big publishers in rich countries have already made substantial investments into AI trained on useful datasets. They are mastering the tools and racing ahead. Small publications, and those in Global South countries (especially where English is not the lingua franca), find themselves at a perceived disadvantage when it comes to models that are trained on local data, and supposedly lack access to the skills and financial resources needed to test out new ideas.

Yet our clarion call is not to race alongside, consequences be damned (after all, we’re champions of risk-conscious development). We’d like to help shift us from a “Move Fast, and Break Things” to a “Move Fast, and Fix Things” mindset. To do this requires two types of learning: an introductory understanding of both the relevant technology and the relevant politics. To start on the politics, the rapid adoption of AI-as-a-service by Global South companies only feeds into an already growing context of the data extractivism (and resource extractivism) inherent to digital colonialism.  We should not divorce our use and adoption of technology from its political consequence: so the question then becomes, how can our very own development and/or implementation of technologies contribute directly to the kind of African renaissance much of our work is principally seeking to contribute to?

To move to the more technical knowledge itself, we need to build our skills first around a consciousness of the problem we want to solve, rather than simply a concern to stay “ahead” with media houses in the Global North that have a very different context, and agenda, to our own. This focus helps prevent both direct cost and time costs in trying to embed inappropriate technologies that distract our newsrooms from their most immediate priorities (like investing in boring old basic digital infrastructure, for instance). But moving further, to tie our political and technical priorities, it is worth reflecting on another key finding from the survey from respondents, which was:

“Tech companies are driving innovation in AI and other technologies, but survey participants voiced concerns about their profit-driven nature, the concentration of power they enjoy, and their lack of transparency.”

How the technology we use is made, and by whom it is made, matters. The nature by which technology is made ultimately influences its impact. And there are plenty of great local technologists doing great, considered development (a shout-out for instance to the fascinating work being done at Lelapa AI).

That being said, the first step is to play around. Play around with some AI offers like ChatGPT, but also realise that most of you are already probably engaging machine learning and other AI technologies already. The AI hype machine is designed to keep you in awe, and commercially dependent. When a newsroom starts with the question “What is my problem?”, the answer is often much more low-tech, and cheaper, than the AI world would have us believe. And we’d guess that nine out of 10 times the problem’s going to be “we need the data”.

Ultimately, though, the report is a refreshingly honest and hype-free assessment of where we are today, and the level of seriousness with which it suggests publishers are looking at AI is reassuring. As with any new technology, if the tools make journalists more efficient, productive and improve the quality and accuracy of serious reporting, they can be embraced. AI, and generative AI in particular, may seem to carry more risks than most emerging tech, but that doesn’t mean we should shy away from their use. DM

Gabriella Razzano is the Executive Director of OpenUp, an Atlantic Senior Fellow in Social and Economic Equity and an Expert Advisor to the African AI Observatory.

Adam Oxford specialises in innovative news product design and strategy, and works with OpenUp on projects that support quality journalism throughout Southern Africa.