LABEL RESPONSIBILITY, NOT TOOLS

World Food Day, 16 October 2026

On World Food Day, 16 October 2026, an agricultural organisation published a poster celebrating one of humanity’s oldest and most important journeys: the journey from soil to plate. Behind that poster was a human process. Someone conceived the message and understood the farmers, crops and value chains being celebrated. Someone decided what the public needed to know, selected the images, shaped the words, reviewed the composition, checked the facts and approved the final publication. Yet the platform reduced all of that work to two words: AI content.

Those two words may describe something about the technology involved, but they say almost nothing about the human effort, editorial judgement, responsibility or intention behind the publication. The machine became visible, and the human became invisible. That experience raises a question far larger than one agricultural poster in Tanzania: who owns human creativity in the age of artificial intelligence, and who decides when a person’s work is no longer human enough to deserve recognition?

I write as a publisher, a communicator and a farmer in East Africa. I believe in technological progress, in editorial accountability and in the public’s right to know when information has been manipulated. I do not oppose transparency. I oppose incomplete transparency: a system that identifies the instrument of production while obscuring the person responsible for the work, and a regulatory direction that risks confusing technological assistance with the absence of human authorship. This is not an argument against responsible AI regulation. It is an argument for better regulation.

1. We do not label a farm by its tractor

Consider the modern farmer. She may prepare land with a tractor, plant with precision equipment, monitor crops through satellite imagery, irrigate with automated systems and harvest with a combine. When her maize reaches the market, nobody labels it “tractor-grown maize”, as though the machine rather than the farmer were the producer. We recognise the farmer, we judge the quality of the harvest and we demand compliance with food safety standards. When something goes wrong, we hold the responsible producer or trader to account. The tractor does not become the farmer merely because it performs part of the physical work.

The same principle applies to the architect using computer-aided design, the accountant using spreadsheets, the photographer using editing software and the journalist using a word processor. These tools have transformed production without extinguishing human authorship. Artificial intelligence does introduce important new capabilities, including the ability to generate substantial creative material, and that difference deserves serious consideration. But it does not justify treating every use of AI as though the human contribution has vanished. The question should not simply be whether AI was used. It should be what AI did, what the human contributed, what the audience is being asked to believe, and who accepts responsibility. A machine can help cultivate a field without becoming the farmer, and it can help compose an image without necessarily becoming the author.

2. Meta’s own experience demonstrates the problem

The history of AI labelling on major social media platforms shows why crude classifications are inadequate. In 2024, Meta introduced labels identifying content as “Made with AI”. Photographers soon complained that genuine photographs were receiving the label after relatively minor edits with AI-enabled software. In one widely reported case, removing a tiny visual imperfection with an AI-assisted tool was enough to trigger the label. The photograph still showed a real scene, and the photographer remained the photographer, yet the label suggested something fundamentally different about the image’s origin.

Meta subsequently changed the wording to “AI info” and later made labels less prominent on material that had been edited rather than generated. These revisions were welcome acknowledgements that context matters, but they also exposed the deeper problem. A technical signal showing that AI software touched a file does not establish how much creative work the AI performed, whether the depicted event was real, or whether the publisher is misleading anyone. Metadata provides useful evidence about a production process; it cannot, by itself, settle questions of authorship, authenticity or truth.

A genuine photograph lightly retouched with AI and a wholly fabricated image of a political leader accepting a bribe may both involve artificial intelligence, yet their meaning, purpose and potential for harm are radically different. A transparency system that cannot explain this distinction risks informing the public badly while claiming to inform it well.

3. The software industry exposes the inconsistency

Consider how modern software is built. Programmers increasingly use AI assistants to generate code, find bugs, explain unfamiliar systems, write tests and accelerate development. A developer may describe a complex business requirement and receive hundreds of lines of suggested code in seconds. That is where the real work begins. The developer must judge whether the architecture is appropriate, whether the code is secure, whether it meets the customer’s needs, whether it protects personal information and whether it will run reliably. A small script may take minutes; a serious platform may take months or years of daily human direction, involving thousands of decisions about databases, security, payments, permissions, infrastructure and risk. Throughout, the developer pays for the tools, the servers and the hours.

AI assistance may be extensive, but human responsibility remains indispensable. When software fails, the customer does not sue a chatbot; the developer or business answers. When sensitive data is exposed, the fact that an AI suggested the code does not absolve the operator. And when a business succeeds, its founders and engineers are not stripped of recognition because they used AI. This does not mean AI-generated code raises no questions of intellectual property, security or accountability. It means those questions must be answered with precision. Why, then, should a publisher’s poster, a journalist’s article, an educational illustration or a farmer organisation’s video be treated as though AI use automatically diminishes the human contribution? If human direction, verification and responsibility matter in software engineering, they should matter in creative communication too.

4. We must distinguish assistance from generation, and generation from deception

One of the greatest weaknesses in the current debate is the tendency to treat all AI involvement as a single category. Using AI to correct grammar, remove background noise, generate a new illustration or fabricate a realistic video of an event that never happened are profoundly different acts. So too is the difference between a publisher openly using synthetic imagery to illustrate an idea and someone deliberately presenting that imagery as photographic evidence. These distinctions should form the foundation of any credible labelling framework, which should recognise at least four situations:

1. AI-assisted human work, where a person substantially creates the work and uses AI for editing, enhancement, translation, research or other supporting tasks.
2. Human-directed, substantially AI-generated work, where a person provides the concept, direction, selection and editorial control, but AI generates significant parts of the final text, imagery, audio or video.
3. Synthetic representations of reality, where AI creates or alters realistic depictions of people, voices, places or events that audiences might reasonably mistake for authentic recordings.
4. Deceptive or harmful synthetic content, used to impersonate, defraud, manipulate, defame or mislead, particularly when presented as authentic evidence.

These categories are not interchangeable, and they do not deserve identical labels or identical regulatory responses. A system that draws these distinctions is not weaker on transparency. It is more transparent.

5. Authorship is a human interest, not merely a technical classification

Article 27(2) of the Universal Declaration of Human Rights recognises everyone’s right to the protection of the moral and material interests arising from their scientific, literary or artistic production. That principle is directly relevant to the AI debate. It does not mean every AI-assisted work automatically qualifies for copyright; intellectual-property law differs across jurisdictions, and the degree of human creativity required remains a live legal question. But the broader human interest deserves recognition: people have legitimate interests in being identified with their contributions, protecting their reputations and receiving fair acknowledgement for work they direct and produce.

A publisher may spend days researching an agricultural value chain, interviewing farmers, verifying information and refining a message. If AI assists with the final illustration, that contribution does not cease to exist. The same holds for a small business owner preparing marketing materials, a teacher developing learning resources or a community organisation producing public-health information. A label that mentions only the machine leaves audiences with a distorted picture of how the work was made. Recognising AI involvement and recognising human authorship are not competing objectives. A well-designed system should do both.

6. Misinformation existed long before artificial intelligence

Falsehood is not a technological invention. Propaganda, forged documents, fabricated quotations, staged photographs, impersonation and fraudulent advertising all existed long before generative AI. The printing press could distribute lies, radio could broadcast propaganda, television could manipulate public opinion, and the internet made misinformation faster and cheaper to spread. Societies did not respond by labelling the printing press, the camera or the typewriter. They responded with responsibility: publishers who stand behind what they print, editors who check facts, laws against fraud and defamation, and a public taught to question sources.

Artificial intelligence is a further escalation, because convincing synthetic material can now be produced and distributed at extraordinary speed and scale. That is a serious danger. But it does not follow that everyone who uses AI should be treated as a potential deceiver. We should strengthen the mechanisms that address real risks: source verification, provenance records, editorial standards, media literacy, independent fact-checking, protection against impersonation and enforceable remedies for fraud. AI-specific disclosure is especially valuable where synthetic material could reasonably be mistaken for authentic evidence, but disclosure is no substitute for accountability. A false statement written entirely by a human is still false, and a truthful explanation developed with AI assistance is not thereby deceptive. Truthfulness cannot be determined by the tool used to produce a sentence.

7. The European Union has already recognised the importance of editorial responsibility

The European Union’s AI Act provides an important starting point. Article 50, whose transparency obligations took effect on 2 August 2026, covers certain AI-generated and manipulated content, including deepfakes and AI-generated text published to inform the public on matters of public interest. Significantly, the provision on public-interest text contains an exception where the content has undergone human review or editorial control and a natural or legal person holds editorial responsibility for its publication. That is a meaningful recognition that AI involvement does not erase the relevance of human editorial oversight.

The exception is not, however, a universal exemption for AI-assisted content, and it does not extend to synthetic imagery, audio or video. The law also separates the obligations placed on AI-system providers from the disclosure duties placed on those who deploy them. Those distinctions matter, and they lead to a straightforward question for European policymakers: if human review and identifiable editorial responsibility justify different treatment for public-interest text, why should comparable safeguards not inform a proportionate approach to other creative media? An agricultural poster, a documentary, a public-awareness video and an educational illustration can all involve substantial human direction and editorial control. They should not be treated as equivalent to a fabricated recording of a real person. Europe has an opportunity to set a global model that protects the public without diminishing legitimate creative work, and that opportunity should not be wasted.

8. Africa must not become collateral damage in global AI regulation

In much of the world, AI is discussed chiefly as a threat to employment, authenticity or established creative industries. In Africa the picture is more complex, because AI also offers a way past longstanding constraints in access to design, translation, technical expertise and digital production. A farmer cooperative in Iringa may have no budget for a graphic designer, translator, photographer or video editor, yet it must still communicate with buyers, explain production standards and compete in digital markets. A young entrepreneur in Nakuru can use AI to build a website, write product descriptions and test software. An extension officer in rural Tanzania can use AI-assisted translation and illustration to make technical knowledge accessible to farmers in their own language.

These tools guarantee neither quality nor commercial success, but they widen participation for people with limited resources. If AI-assisted work is routinely presented in ways that imply inferior authenticity, smaller creators may carry a reputational burden that large institutions can easily absorb. That risk deserves investigation, not dismissal. African creators should not have to choose between affordable technology and recognition for their work. We must not create a digital hierarchy in which wealthy institutions can afford conventional production while resource-constrained creators are marked as less legitimate for using accessible tools. The rules of AI should empower emerging economies, not quietly reinforce existing inequalities.

9. The Global South deserves a meaningful voice in rule-making

Regulatory decisions taken in Brussels and platform policies written in California shape what people see across Africa. A labelling policy designed for Western media markets is applied, unchanged, to a farmer cooperative in Tanzania, a teacher in Uganda, a trader in Kenya or a small publisher in Zambia. These users have little influence over the systems governing their work, yet they live with the consequences. That raises genuine questions of procedural fairness, representation and digital sovereignty.

Who designs AI labels, and whose understanding of authenticity informs the terminology? Are labels tested with Kiswahili speakers and users of other African languages? Are small publishers consulted, and do appeals work effectively across countries? Have the consequences for African SMEs and civil-society organisations been assessed, and do platforms publish data showing whether automated labels are applied consistently across regions, languages and content types? These questions belong in mainstream AI governance. African participation must mean real consultation, representation in standards bodies, independent evaluation and accessible channels for challenging decisions, not an occasional invitation to a conference. Africa should not merely receive the rules of the AI economy. It should help write them.

10. Transparency without explanation can become a form of misrepresentation

A label should clarify something meaningful. When a platform marks a publication as “AI content”, what exactly is the reader meant to understand? That the whole image was generated? That AI adjusted the lighting or removed a background? That the text was translated, or drafted with an assistant? That a realistic scene was fabricated? Or merely that metadata from an AI-enabled application was detected? These are different facts, and folding them into one vague category may be convenient for platforms, but convenience is not transparency.

The public deserves meaningful information about the nature and significance of AI involvement. The creator deserves a fair representation of the work performed, and the platform should be able to explain the basis of its classification. Where a label is applied automatically, the publisher should receive an understandable reason and an effective route of appeal. Nobody should have to prove the authenticity of their entire creative identity because software detected a technical signal. Transparency should explain reality, not replace one ambiguity with another.

11. AI companies and platforms must be held to the same standard of accountability

There is a further contradiction worth confronting. Major technology companies rely heavily on AI for software development, advertising, content moderation, translation, recommendation and customer service. These systems sit at the heart of how platforms earn revenue and exercise influence. Yet individual creators find their own use of AI highlighted in ways that overshadow their contribution. The comparison is not exact, since platform operations and public-facing synthetic media raise different transparency questions, but the principle of consistency stands.

If AI involvement is relevant to accountability, platforms should explain how their own systems affect users. How are automated labels generated, and how often are they wrong? What evidence supports a classification, and what happens when a publisher disputes it? Can the system tell a minor edit from substantial generation? Do users receive comparable information about significant AI-driven platform decisions that affect their reach and reputation? Transparency must be reciprocal. Platforms cannot demand clarity from creators while withholding meaningful explanation of their own automated decisions.

12. The right to appeal is as important as the label itself

Any automated classification system operating at scale will make mistakes, and its governance must reflect that reality. A mistaken label may seem trivial to a platform processing billions of posts, but for a small publisher, reputation is an asset built over years. If readers take a label as evidence that a photograph is fabricated or that an article lacks human authorship, the resulting confusion can damage trust. Such harm should not be presumed in every case, but it is plausible and deserves study.

Publishers therefore need more than a generic help page. They need clear reasons for a label, the opportunity to submit supporting evidence, timely review and prompt correction. Where a platform discovers a systematic error, it should correct affected labels at scale rather than forcing each creator to fight alone. A platform that can apply a label automatically must also invest in making its mistakes reversible. Due process should not disappear merely because a decision is made by software.

13. A better system is possible: recognise the author and disclose the assistance

The choice is not between hiding AI involvement and stamping everything with a crude warning. A more informative system can acknowledge both the person and the technology, using categories such as:

• Human-created, AI-assisted: the creator used AI for limited supporting tasks.
• Human-directed, AI-generated elements: significant elements were generated by AI under human direction and editorial control.
• Substantially AI-generated: AI produced most of the expressive content, with the named publisher responsible for its release.
• Synthetic depiction: the material realistically represents something that was not authentically recorded.
• Verified deceptive manipulation: the content has been found, through a fair, evidence-based process, to misrepresent material facts or impersonate someone.

These categories would need careful definition, testing and safeguards against abuse, and not every one requires a prominent public warning. Minor assistance may warrant no visible label at all, while materially synthetic depictions of real-world events should carry clear disclosure. The guiding principle is proportionality: a small edit should not trigger the same message as a realistic fabricated recording, and a fictional illustration should not be treated like a forged document. Equally, human editorial responsibility should be acknowledged without being used to conceal significant synthetic generation.

A useful publication notice might read: “Published by [Organisation], which takes editorial responsibility for this content. Illustrations created with AI tools under human direction.” Such a notice tells the reader everything that matters: who stands behind the work, what role the technology played and whom to hold accountable. It respects the audience, the creator and the truth at the same time.

14. What we ask for

To the European Commission and its AI Office, we ask that the implementation of Article 50, and any future review of it, be guided by proportionality and by the principle already present in the Act: that identifiable human editorial responsibility matters. Guidance and codes of practice should distinguish clearly between AI assistance, AI generation and deceptive synthetic depiction, and should extend the logic of the editorial-responsibility exception, with appropriate safeguards, to images, audio and video produced under human direction.

To Meta and every platform that labels content, we ask for labels that describe what actually happened rather than what metadata implies; for the accountable author or publisher to be displayed alongside any AI notice; for detection systems to be tested for accuracy before their verdicts are shown to the public; for clear reasons and a fast, fair appeal process; and for regular public reporting on error rates, broken down by region and language.

To African governments, regional bodies such as the East African Community and the African Union, and our own media institutions, we ask for active engagement in global AI standard-setting, for national guidance that focuses on deception and harm rather than on tools, and for investment in media literacy so that citizens can judge content on its truthfulness and its source.

And to creators across our continent, we ask for the discipline that gives this argument its force: use AI openly, verify everything, put your name to your work and accept responsibility for it. Our claim to recognition rests on our willingness to be accountable.

15. Conclusion: label the liar, not the ladder

Artificial intelligence will soon be as ordinary as the tractor, the camera and the computer. Like every tool before it, it can be used to build or to deceive. The question that matters was never which tool a person used. It has always been whether the person tells the truth, and whether they stand behind their work.

A good law should reward that responsibility, not erase the human behind it. It should protect the public from deception without teaching the public that every AI-assisted work is suspect, and it should guard against fabrication without diminishing the farmer cooperative, the teacher, the developer or the publisher who uses new tools to do honest work better.

Label the tool where it genuinely informs the reader. Expose the deceiver wherever deception occurs. But never let the machine take the place of the human who conceived, directed, checked and published the work.

Label the liar, not the ladder.

Anthony Muchoki
Publisher, Kilimokwanza.org