In the runup to our newspaper’s 22nd anniversary last week, media colleagues and readers told me they noticed that we are regularly crediting DeepSeek and Gemini for our explanatory footnotes.
Some said they believed that The Macau Post Daily is one of the world’s first newspapers – perhaps the first one – whose masthead lists its team members, printer, wire services (AFP and Xinhua), and company details alongside the AI tools that we are using.
Well, as a publisher, I delight in being a trailblazer.
We are also the world’s only English-language daily that publishes a full page of news and views about the Portuguese- and Spanish-speaking countries.
Earlier this year, we added both large language models (LLMs) to our masthead to inform our readers about the newswires and AI tools we use to continuously improve our newspaper’s quality.
Of course, AI is more than just a tool; it possesses agency. That presents a challenge not only for journalists, but for anyone using it in a professional context.
In artificial intelligence, agency refers to a system’s ability to act autonomously to achieve specific goals rather than merely responding to individual, step-by-step prompts.
Unlike a non-agentic model that relies on constant user prompts (e.g., ‘Translate this text’), an agentic AI independently evaluates options, plans multiple steps, uses external tools, and adapts based on incoming feedback.
Agency allows AI systems to make independent decisions to achieve predefined objectives. As a daily user of AI, I am well aware of its tendency to invent false facts, known as hallucinations. Therefore, professional journalistic standards require a human-in-the-loop (HITL) for any AI-assisted reporting, editing, and publishing.
Human-in-the-loop (HITL) is a design approach where humans actively intervene in an AI-driven system to guide, review, accept, or correct its actions.
In AI, including publishing, the term “loop” refers to the continuous cycle of automated data processing.
In other words, HITL describes any automated process that requires human oversight, judgment, or intervention at critical steps. Placing a human inside that loop ensures a person reviews, corrects, or approves the machine’s work before a final action is taken.
Instead of allowing an AI system to operate with complete autonomy, the HITL approach keeps human judgment at critical decision points to ensure safety, accuracy, accountability, and ethical alignment.
While AI can help summarise notes or suggest outlines, a journalist must always write and edit the final story.
After all, journalism is the process of gathering news and views, writing as accurately, even-handedly, and concisely as humanly possible, and editing the text to ensure it is reader-friendly, accurate, and fair – what I call the ‘ABC’ rule of journalism: accuracy, balance, and conciseness.”
Journalism is not a literary pursuit, but an act of communication.
Our newspaper mainly relies on AI assistance for three tasks: proofreading, background referencing for footnotes, and translating texts from Chinese and Portuguese into English.
Occasionally, we also use it for subediting, but that is a slippery slope: AI can get carried away and turn overly creative, risking the integrity of the original text. For professional journalism, that is unacceptable. Content is sacrosanct – only the reporter, subeditor, and editor bear responsibility for it.
“Consequently, as chief editor, my routine starts by prompting DeepSeek or Gemini: “Please proofread this Macau newspaper article for grammar and spelling, without changing any content or attributions”. More often than not, it works.
AI agency must stop where editorial judgement begins. Full stop.
Let’s be realistic: AI has become a strategic partner in our daily editorial work – not as a replacement, but as an amplifier of human journalism.
For us, high-speed AI is a useful partner, particularly when it comes to translation. While newspaper translations took hours or even days just two decades ago, today they take seconds. Like human translations, AI translations are not perfect. They must be edited by a human who is proficient in the language and, most importantly, well-versed in the content. That is often an issue with human translators – they frequently struggle to understand the subject matter. In my experience with both forms of translation (and as a translator myself decades ago), AI is often better at understanding the core context of an article. In the near future, I believe translators will primarily be tasked with reviewing, correcting, and polishing AI-generated translations.
What AI can certainly do better than a human is search through vast amounts of data and background information in a very short time. However, data compiled this way needs cross-checking by us journalists.
AI also comes in handy when we struggle to improve and shorten (“the shorter the better”—one of journalism’s oldest maxims) a headline or subheading. DeepSeek and Gemini are pretty good at it, and they often come up with two or three alternatives”.
The era of the augmented newsroom and generative AI began in 2018 when Reuters launched Lynx Insight, an AI assistant designed to aid journalists by scanning vast datasets, spotting anomalies, suggesting story leads, and assisting with background research.
By 2022, LLMs shifted the dynamic to interactive newsroom tools – supporting subediting, translation, background checking, headline generation, and search optimisation under human editorial supervision.
In late 2022, OpenAI unveiled ChatGPT (Generative Pre-trained Transformer), introducing human-like dialogue to the public. The model was pre-trained on vast amounts of internet text and fine-tuned for conversational tasks using Google’s Transformer neural network architecture, which allows the system to process sequences of words efficiently, understand context, and evaluate word relationships within a sentence.
When we launched our newspaper on August 27, 2004, all of this was still futuristic. Initially, we focused on our print edition, but about a decade later, we gradually developed our online presence.
Nowadays, the unabridged e-edition of our 16-page newspaper is released daily on our website. However, I am convinced that a newspaper should live up to its name – it must always have a printed edition, as that strengthens readers’ confidence in our product.
Unlike digital-only news outlets, whose articles can be silently edited or removed in seconds, printed newsprint builds trust through its permanence.
Over the past 22 years – and increasingly assisted by AI in recent years – our newspaper has carved out a unique space in Macau’s media landscape through strict adherence to the principles of professional journalism.
We continue to strictly separate news from opinion and adhere to the time-tested 5Ws of journalism (Who, What, When, Where, Why) – requiring full source attribution and ensuring that all news photos carry clear descriptive captions and source credits. However, we often use collective captions (also called omnibus captions) for photo features and exhibition coverage. These avoid repetitive text under every individual picture, keep the layout clean, and provide structured context.
In my view, every news photo must carry a caption built on the 5Ws. Anything less shortchanges our readers and compromises the standards of professional journalism.
AI may be humanity’s greatest invention, just as electricity was its greatest discovery. Electricity is arguably mankind’s greatest technological achievement, because without it, modern life as we know it would be impossible.
AI is arguably humanity’s greatest invention because it expands human intellect. The danger lies in potential cognitive laziness. Ultimately, because AI relies on human-written code, human-labelled data, and statistical probability, it is a reflection of humanity itself.
As the probably most important tool ever invented by us humans, AI should be a common asset of all humanity. Incidentally, within AI’s global context, I think we should all gratefully remember Alan Turing (1912–1954), the British mathematician whose work during World War II laid the foundations for modern computing. In 1950, Turing published a paper titled “Computing Machinery and Intelligence,” where he posed a question that we all now know was profoundly pertinent: “Can machines think?” Depending on how one defines thinking, the answer today is increasingly looking like yes.
We celebrated our newspaper’s 22nd anniversary about a month after President Xi Jinping’s call for joint efforts to build a just and equitable system for AI governance. Xi made the appeal on July 17 when the World Artificial Intelligence Cooperation Organisation (WAICO)* was created in Shanghai.
“All countries should take a people-centred approach and develop AI for the public good.
“We should ensure that AI is an important driver for shared prosperity and common security”, Xi said.
China is a global leader in AI development. Reportedly, its open-source large AI models have been downloaded more than 10 billion times.
“AI development should not be a solo performance by a single country, but a symphony of international cooperation”, Xi said. It remains to be seen how the US and EU, for instance, will respond to his sensible appeal.
As a dyed-in-the-wool optimistic realist, I expect AI to play a positive role in humanity’s future development, particularly in fields such as medical diagnosis, news reporting, climate risk analysis, weather forecasting, and natural disaster warning systems. However, we must remain on guard so that such a potent tool – arguably the most powerful ever created – does not slip beyond our control. Unlike traditional human tools, such as the simple hammer invented by our ancestors three million years ago, AI possesses true agency.
AI is an extraordinary tool for virtually everyone – provided there is always a human in the loop.
– Harald Brüning
* “I recommend the following article by University of Macau Associate Professor Francisco José Leandro: ‘WAICO: Towards a More Inclusive and Representative AI Governance Architecture”.
-
https://www.macaupostdaily.com/news/28973
Circuit brain icon for AI hardware. – Source: Magnific via Gemini
