Enhancing journalism through AI: What are the actual possibilities?
Earlier this year, I attended the KAS Media Africa Local Journalism Conference in Dar es Salaam and had the honour of contributing to two panel sessions: one on disinformation and the second on ‘artificial intelligence in investigative journalism’. What I am sharing in this article is essentially the same things I said during that programme (and more).
First, it’s important to note that artificial intelligence tools have been around for quite some time. They are not new. Once you understand the basic mechanics behind AI, you will see how it is already incorporated into many things we’ve grown accustomed to.
Google Translate, for example, has never seen that email written in French before, but it is able to tell you what it means because of AI. It has been trained on a large dataset so that it can make such predictions with startling accuracy. Google Assistant is able to tell you what music is stuck in your head just by hearing you hum it for a few seconds, again thanks to machine learning. Oh, and why do you think Google Lens is able to identify objects and characters and brands from pictures? Why do you think cars are now able to drive themselves? Even text-to-speech and dictation apps have gotten so much better thanks to advances in artificial intelligence. Let’s not forget the autocomplete feature in your Gmail and Google Docs. We have that because the platform has processed tons of similar articles and emails and so is able to predict what you are trying to say or what you ought to be saying based on recognised patterns.
Whenever you play chess and opt to play against a computer opponent with a specific difficulty level, that is also artificial intelligence. As far back as 1997, a model trained by IBM through brute force called Deep Blue defeated Garry Kasparov, the reigning world champion.
The reason for the latest frenzy is what is called generative AI. What this means is that the computer is not only able to predict; it is also able to create. Given a prompt, it can come up with a relevant picture, video, audio, or text seemingly from thin air — something that has previously not existed. It can also clone and fill in gaps in existing pictures, videos, texts, and audio files. In short, it can mimic humans. Nowadays, that mimicry is so good that we can barely tell the difference between it and the real thing.
Advancements in computing power and the application of artificial neural network approaches (which mimic how the human brain functions) to machine learning have aided the development of Large Language Models (LLMs) such as ChatGPT. In secondary school, we were taught that computers could only understand programming languages such as Java and Python. But now, LLMs are able to give instructions in a computer-intelligible format using the languages most humans speak using something called natural language processing. They are like an interpreter between the layman and the machine. That is why you can tell ChatGPT to give you a recipe using a set of ingredients, and it would understand without you having to rely on coding skills. Pretty cool.
- Watch this TED-Ed video to understand how different AI models are trained.
- Check out this YouTube channel to see examples of deep reinforcement learning in the field of motion and strategy.
- Check out Teachable Machine to train your own machine learning model (free of charge) essentially for object identification and analysis.
Now, to the reason we are here. In what ways will this artificial intelligence revolution help journalists? I have a couple of ideas (eleven in total). Some of them are painfully obvious. Others, not so much.
Ideation
You are looking for ideas of stories you can work on based on your interests or previous publications? You have written an article but have no idea what headline to use? Or you have written one, but it still feels like something is missing and you can’t place your finger on it? You want to create art illustrations for a report and are not sure what direction to take? AI tools can help with recommendations.
Imagine if there was a tool that scoured the news for articles that might interest you, and every week, just before the editorial meeting, it sent you a list of reports and what angles you could explore for a story.
Research
Beyond just giving you story ideas, they can also do the research for you. Previously, people mainly relied on search engines like Google, which combed the internet for content related to the query. The conventional method is like having a librarian who is familiar with all the books and knows where to find them. But with new AI tools, what you have is a librarian who has mastered all the books and simply answers any questions you might have instead of giving the books to you. Because of the problem of hallucination (that is, generative AI making things up and presenting them as fact) and to ensure transparency, many tools are trying to merge both approaches. So, the librarian answers your questions and then hands you the book so you can confirm that those answers are correct and do more research yourself.
Accessibility
Artificial intelligence tools can help make our reports more accessible to all kinds of people. Already, browsers like Google Chrome come equipped with page translation features. On top of that, you can now present readers with a text-to-speech option for your articles that translates them from the original language to multiple other languages. That way, you can reach people from all parts of the world. Earlier this month, OpenAI announced GPT-4o, which, among other incredible abilities, can translate speeches from one language into another with incredible speed. This means not only can journalists reach new audiences with their stories, they can also go into various corners of the world to report events without worrying too much about language barriers.
Another beautiful thing is that publications can now easily be produced in formats that are accessible to people with disabilities. The text-to-speech technology, which has now been imbued with human-like voices or can even be made to sound like a particular person, helps people with visual impairment. AI-powered closed captioning (automatic subtitles) can help people with hearing disability. Given the awesome improvement in object-identification AI, you can also add auto-generated alt descriptions to pictures included in a report so that those who cannot see them can have an idea of what they are showing. Also, you can present summaries of articles (of various lengths) generated by AI to readers who do not have the time to read long-form journalism. And, of course, you can easily convert stories from one format to another with AI, e.g. from text to audio, video to text, etc.
Newsrooms with a large collection of printed publications or aired broadcasts can also digitise them using optical character recognition (OCR) and transcription tools and then make them available to their users.
Automation
AI tools allow you to automate processes you would ordinarily need humans for. You will see this playing out in the other examples in this piece, but another thing I have in mind is writing stories from press statements or writing summaries for newsletters. You can do that now entirely or mostly using artificial intelligence while you focus human resources on developing more mentally tasking and creative reports.
Presentation
You can make your work look more professional and presentable using artificial intelligence tools. If you cannot afford to pay a professional illustrator, you can have the computer do it for you. You can generate videos for various social media platforms from one report. You can convert a set of data into a visually appealing infographic with little to no design knowledge.
We might also start more instances of newsrooms using deepfake in a controlled way to maximise resources. They can have an AI clone of someone’s voice or looks to present a podcast or read the news. The aim here could be to reduce workload or ensure consistency. Similarly, you should be able to correct errors in videos and audio files that were noticed too late without affecting the smooth delivery — saving you the stress of re-recording/re-shooting. It’ll now be as easy as updating a word document.
Consider this possibility as well: using one person to do the voice acting for different sources, regardless of their age and gender, rather than scrambling to find multiple actors. Wouldn’t that be nice?
Interactive storytelling
One word: digital replicas. Sorry, that was two words.
AI allows us to create more fluid digital characters that can engage in human-like conversations. There are at least two ways this can serve the purposes of journalism: chatbots and NPCs (non-player characters).
Imagine being able to chat with an internally displaced person from a particular region or a homeless person. You can ask them questions on various topics like how they earn money, what their shelter is like, how they raise their children, whether they look forward to returning home, and so on. The responses they provide, of course, would be based on tons of available data, interviews, and reports. We can take this a notch higher and make the digital replicas humanoid or even three-dimensional, so it is like you are sitting across from them.
We can use the same technology for informative gamification.
Analysis
We can now analyse large volumes of data in a shorter amount of time thanks to artificial intelligence tools. Google’s Pinpoint, for example, allows you to upload tons of documents, including audio files, and easily search through those documents (through optical character and speech recognition technology). It will also identify different elements in the documents (such as people, places, and organisations) and allow you to filter out portions pertaining only to those elements.
We can use this tech for investigations, too. You can tell a story by gathering a large amount of pictures and looking for patterns or outliers (or rather instructing a computer to look for them). Imagine being able to tell if security agents are recycling the same pictures to give an impression of continued military successes or recycling weapons that are planted on innocent people who are paraded as criminals. It would take forever for the human eye alone to spot such patterns.
The Centre for Journalism Innovation and Development (CJID) has developed a fascinating tool intended to monitor radio broadcasts (including those in pidgin English) and document claims made (separating them from other kinds of statements) so that fact-checkers can easily tackle misinformation spread through that medium.
I believe a similar tool can be created to aid accountability journalism. What it’ll do is crawl the web for radio and TV interviews and create a searchable database of appearances by different public office holders and the statements they make. So every time an interview of a minister is uploaded to YouTube, it would transcribe and publish it. Every time they mention a promise, a figure, or a claim, it would highlight it.
Efficiency
If you have ever used Grammarly, then you know just how handy an AI assistant can be for copyediting. Sometimes, I disable the app just to see how clean I can make an article. It’s nearly impossible to catch every single typo, especially in long documents. And applications like Grammarly do not only check your spelling and grammar but also recommend better ways to construct your sentences. So, small newsrooms that cannot afford to hire five experienced editors, for example, can instead have two who lean on AI tools to deliver error-free copies. Another exciting prospect is training the model with your newsroom’s editorial style guide.
Commercialisation
The ability to easily convert content from one format to another comes with an opportunity to make a profit. You can have a report that is free to watch on YouTube or free to listen to as a podcast but behind a paywall as a written article on the website. You can charge people to access an in-depth study while making the summary available for free. That way, you are able to balance your need for financial sustainability with the right of the public to access information. If you develop an AI tool like CJID is doing or you train a model that is useful for others, you can also charge people to use them.
Verification
A great example of how AI can be used for fact-checking is PimEyes, an advanced face recognition search engine. You have a picture of someone you cannot identify? No problem. Simply upload it to the platform and it will look for where pictures of that person have surfaced on the internet, even if those pictures were taken when they were much younger.
FactCheck Africa has built a brilliant tool, MyAIFactChecker, which is able to verify claims with stunning accuracy by aggregating and summarising articles from reliable websites — using the large language model motherboard.
Though, as far as I know, no publicly available app does it yet, it is theoretically possible to have an AI tool that geolocates pictures using various clues from it — just like human fact-checkers, but much faster.
Distribution
At the KAS Media conference I mentioned at the beginning of this article, someone introduced participants to a digital periodical published in an African language. However, there was a huge problem with the distribution system. To get a copy of this magazine every week, you needed to text a number on WhatsApp monitored by someone who then added you to a list. Several people were managing different lists. They would send the magazine to every subscriber one after the other and it always took them several hours to complete the process. I’m sure you’re exhausted just reading this. Yes, there are bots that can simplify and automate the process. And now, with AI features, those bots can perform advanced tasks like supplying previous editions of the publication without requiring specific instructions etc.
Some final thoughts:
- From the 1950s to the present day, a lot has changed. Still, nothing has changed. We are still in the era of computer-assisted reporting, even with the emergence of generative artificial intelligence. The tools are here to assist us and to enhance our work.
- Newsrooms are not places for writers, reporters, and editors alone. We must create spaces for computer scientists, programmers, and even so-called prompt engineers.
- Every newsroom must have an AI conversation and develop a policy to guide its use of AI tools. This is new territory, and we must stretch our understanding of ethics to cover the developments. How do we ensure accuracy and originality do not suffer because we are trying to embrace technology? What should be our position on disclosing when a report or a part of it is generated by AI?
- The possibilities are endless. Constant training is important so journalists can stay abreast of the latest trends and see how they can benefit from them.
I have seen a news website dedicated entirely to content generated by artificial intelligence (both text and images). I have listened to podcasts entirely scripted and voiced by artificial intelligence. We might argue these things often come across as soulless and robotic, but I wonder if we only say that because of our awareness of AI’s involvement and our bias. We have already reached a stage where you can tell the machine to sound less like a machine and more like a human. It would pause and chuckle and crack jokes just like people have a tendency to do, and unless you were told, you would not even be able to tell the wires apart from the veins.
One day, many years into the future, when we have finally crossed the Artificial General Intelligence (AGI) milestone, the things written and created by humans will have the same appeal as handcrafted products or homemade food — as opposed to those that are factory-made or mass-produced. And that just might be their only selling point.
Lo! The gears of capitalism may grind against the time-consuming and resource-sapping labour of man, but there will always be an appetite for it. Well, as long as mankind itself survives.
