09: The challenge of Artificial Intelligence
Artificial Intelligence (AI) caught attention worldwide in late 2022 when OpenAI introduced its ChatGPT service to the general public. ‘Chat’ indeed – that was the gist of the new tool and what made people sit up and notice. And experiment. An app you can talk to as in a real life conversation, ask questions, prompt to deliver content on topics of your choice, at any time. Too good to be true? An invaluable daily assistant? Or a nightmare?
The basics
GPT stands for Generative Pre-Trained Transformer, in the words of Chat GPT itself: “an AI model that understands and generates human-like text. It’s trained on large amounts of data to predict the next word in a sentence, allowing it to answer questions, write stories, and hold conversations.”
Artificial Intelligence, a form of intelligence demonstrated by machines as opposed to human intelligence, has been around for decades, What used to be known as “thinking machines” are computer-based systems taught to imitate or reproduce certain human tasks or perform cognitive functions such as sensing, collating and processing data, “learning”, “reasoning” and solving problems. [We add the quotation marks because evidently computers do not learn or reason like humans, although some people use these terms as if there is no difference]. AI tools are created by human scientists and programmers who feed in all the requisite data and train them to perform specific tasks for application in science, medicine, industrial production or logistics – in which further data is usually added and processed..
Building foundational models under ChatGPT, in this case the Large Language GPT model depends on access to powerful computing chips and to huge data centres – facilities beyond African possibilities. This leaves us largely as users of applications built upon models created elsewhere – both of which are shaped by needs and values outside the continent.
All of us already use AI , often without realising it. It is running on our phones behind the scenes to assist in navigating our journey, organise our day, or to help us take better photos, translate into dominant languages or stream music and video. The technology is embedded in search engines such as Google, and it drives recommendation systems used by YouTube, Amazon or Netflix. And it is integrated in strategic games such as Minecraft, chess or others we play online. Like ChatGPT, all the while collecting more and more data to make money for the companies concerned.
The African Union’s Declaration of Principles on Freedom of Expression and Access to Information in Africa expressed a caution in 2019:
“States shall ensure that the development, use and application of artificial intelligence, algorithms and other similar technologies by internet intermediaries are compatible with international human rights law and standards, and do not infringe on the rights to freedom of expression, access to information and other human rights.”
The AU’s Continental AI Strategy 2024 further warns us: “… there is a need to ensure the integrity of information and sustain trust in the use of AI systems and platforms. The responsible and ethical use of AI should also cover media and information literacy that stimulate critical thinking, lifelong learning, global citizenship, freedom and autonomy and cultural competencies.”
Other platforms
Following pioneer ChatGPT , other AI apps have come on the market – such as Gemini, developed by Google; Perplexity (which uses Microsoft cloud); and Claude by Anthropic. Mistral is a French-owned company that has gained as an alternative to closed source AI. Meta’s AI (available also on WhatsApp) claims to be open source, but does not share key parts of its model nor the data it is trained upon (which include Facebook posts and user behaviours). In February 2025, Elon Musk’s xAI launched Grok 3. Musk described the new platform as “a maximally truth-seeking AI, even if that truth is sometimes at odds with what is politically correct.” Going on previous experience, this probably means that there will not be safeguards against prompts that lead to output stereotypes and harmful deepfakes, nor curbs on the use of the tool for attacks on human rights.
The Chinese platform DeepSeek joined the competition in January 2025. This has raised concerns and questions, especially about the origin of training data and data security. The system was trained with a Chinese worldview – a potential problem due to the country’s authoritarian government and its use of information to serve its political ends. And there is high potential for large-scale data leaks to the Chinese government. This is because user data is stored in China and companies are obliged to disclose such to the authorities when requested to do so for law enforcement purposes. On the other side of the world, , US-based AI services are also subject to that country’s ‘Data Request Policy’ which empowers government authorities to demand the disclosure of personal data relating to a particular internet user, with minimal protections for nationals and nothing for foreigners..
NEED FOR CRITICAL ASSESSMENT
Chatbots are ready to answer (almost) any question anyone may want to ask, either orally or by text, usually within seconds. They produce texts – and also images and audio – on demand in various styles and lengths, as prompted. For example, write a news report in the style of a given newspaper, or create a replica of a real news reader but speaking with fake words. Less risky, they translate or summarize, and give answers that partly replace the role of traditional search engines and search functions on platforms. And they are able to give the appearance of debating complex issues in an ongoing dialogue.
But they are certainly not a human being nor a trustworthy friend, even if they refer to themselves as “I”.
Chatbots are fed with terabytes of data compiled from all kinds of sources, basically everything that is available on the Internet – meaning a mass of English-language and Western content. They are not genuinely creative, they merely regurgitate through making links between data points. They do not distinguish between right and wrong, true or false. They will tell you nothing that is not based on what they have been fed before. And they choose content they pass on to you not on the basis of quality but of quantity: what you get is what they have been fed most often.
What they get fed, by the way, is not very much in terms of African languages and content, for example. For African users, AI apps can communicate in Arabic and the colonial languages English, French and Portuguese, but not in most indigenous African languages. ChatGPT, for example, says that in Zulu it can do “basic things like simple greetings or phrases, basic translations and short common sentences”.
Responses can include factual errors and problematic guidance. The models inherit and often amplify biases and stereotypes in regard to gender, race or political opinions. They can also be misused to generate harmful content (e.g., spam, deepfakes, fake news, phishing emails).
The same goes for pictures, audios and videos produced by AI based tools. After – or even during – major news events, the Internet is flooded with forged images – faked just for fun or with intent to spread misleading information. Such visuals are quickly reshared by others, often without any context, and become part of ‘reality’ for uncritical or unthinking users. When these are shared and re-posted, they re-enter the AI data scraping process – further reinforcing what gets output – and increasing what is often termed “AI slop” online. .
Image generator tools such as Midjourney, DALL-E, ChatGPT-4 and others can generate realistic images in seconds – complete with detailed backgrounds, with little more than a simple text prompt. Or they can take existing pictures as material for their own ‘creations’, again responding to prompts by the user who chooses from a set of options.
In short: The need for critical assessment applies to all information generated by AI. Those who are familiar with an issue may find the information offered by an AI service to be useful as pointers or corrective suggestions. Those who have little clue on a matter may take everything at face value – and end up more uninformed or disinformed than before.
OWNERSHIP OF TRAINING MATERIAL
The data mined by ChatGPT and similar AI tools from a huge array of sources are owned by someone. This ‘someone’ will probably have the copyright on his or her work, even if they put it in the public domain. Should the companies be allowed to use material without the agreement of the original creator, and without paying any licence fee? Who, for example, owns a work of art based on an original – the user, the developer of the software or the artist whose work thrown into the mix for the training of the AI tool may be untraceable in the outcome? The same applies to journalism, and of course African journalism in particular.
A number of legal charges has been brought before the US-American Federal Trade Commission against the tech giants for their indiscriminate crawling of the net and stealing copyright-protected content. In an Open Letter to OpenAI and other companies like Alphabet/Google or Meta, the Authors Guild in the USA says:
“Generative AI technologies built on large language models owe their existence to our writings. These technologies mimic and regurgitate our language, stories, style, and ideas. Millions of copyrighted books, articles, essays, and poetry provide the “food” for AI systems, endless meals for which there has been no bill. You’re spending billions of dollars to develop AI technology. It is only fair that you compensate us for using our writings, without which AI would be banal and extremely limited.“
In the wake of ChatGPT’s success, Meta scrambled to deliver its own Meta AI and had to identify useable training material speedily. The company found huge libraries of pirated copies of publications. Their origin is Samizdat, the movement that flourished in the previous century when Russian academics evaded censorship by establishing shadow libraries (later digitalised) without obtaining copyrights. In fact, these were stolen books. And now Meta is stealing their prey. (YouTube, which takes ownership of what users post on that platform, has now found itself being exploited by new AI apex predators.)
In 2025, Meta found yet another way of feeding more material into its AI system. It will now mine content from its social media services Facebook, Instagram and WhatsApp. Users were given one month to agree or disagree, with no reply taken to be agreement, of course.
And what about news media content? Already, media houses in Africa are not paid when Google, for example, uses snippets from articles, audio and videos as part of its links. Nor do they derive any benefit when people share their news stories on social media sites, other than perhaps getting traffic to their own website, as they sometimes do when search engines or social media carry links or pointers to their content. But now, with AI-based generative technologies, news media are again adding value to intermediary companies, without even getting a link back to themselves. In other words, from the point of view of news media, the exploitation of their news content without permission or recompense is going from bad to worse.
In all cases, the AI systems are not just built on past data (much gleaned without permission), but such up new data – whether via your prompts, or your automated agents on Zoom chats, or your documents put into the system for getting out a summary. In short, AI tools are also AI spies.
IN SUMMARY
Journalists need to up their game in reporting to the public the issues around AI and what’s needed to govern the balance of opportunity and risk around the continent. They need to be watchdogs against both the hype and the abuse of AI services to damage human rights and sustainable development. That includes also paying attention to the environmental issues of big data centres, the ownership of African data, biases and errors in AI output, and the growing international AI divide.
This INFO BITE is selected from the online course on Media
and Digital Policy in Africa, offered by Stellenbosch University
in association with Namibia Media Trust.
There are free and paid options available for the full course.
Explore more BITES on a number of related topics
