EDITOR’S NOTE: On June 24-25, UWI St Augustine held the International Media and Communication Conference 2026, hosted by the Department of Literary, Cultural and Communication Department. Themed “The Digital Nexus: Media and Communication in the Age of AI”, the conference highlights included a keynote address by prominent Nigerian-American author, academic, and media scholar Dr Farooq Kperogi. UWI TODAY is pleased to share his address.
Distinguished Vice Chancellor, respected scholars, colleagues, students, media professionals, policymakers, guests, ladies and gentlemen.
I am deeply honored to give this keynote address at the International Media and Communication Conference here at The University of the West Indies, St. Augustine Campus. I bring greetings from Kennesaw State University in Georgia and from the many intellectual communities across Africa, the Caribbean, and the wider Global South that continue to think seriously about what technology does to power, identity, culture, memory, and democracy.
The theme of this conference, “The Digital Nexus: Media and Communication in the Age of AI,” could not be more urgent. Artificial intelligence is now embedded in the everyday infrastructures through which we search, research write, teach, report, translate, bank, vote, travel, police, remember and imagine. It is not merely a tool. It is becoming a communicative environment. That is why I want to speak today on “The Global South’s Fight for Communicative Sovereignty in the Age of AI.”
By communicative sovereignty, I mean the right and capacity of peoples, nations and communities to control the conditions under which they are represented, heard, translated, archived, classified, predicted, and governed through media and technology. It is the right to be more than raw data. It is the right to be more than a market. It is the right to be more than a user of tools designed elsewhere, trained on someone else’s histories, financed by someone else’s capital, and governed by someone else’s assumptions.
I do not mean communicative isolationism. Nor am I making a disguised argument for technological retreat. I do not believe the Global South should reject emerging technologies merely because many of them are designed in the Global North. That would be both impractical and intellectually lazy.
What I mean by communicative sovereignty is equitable participation in shaping the infrastructures of meaning that now shape our lives. It is the insistence that our languages should not be treated as noise, that our accents should not be treated as errors, that our linguistic idiosyncrasies should not be stigmatised as signs of ignorance, that our bodies should not be turned into databases, that our labor should not be hidden, that our cultures should not be flattened and that our democracies should not become testing grounds for algorithmic experiments.
The question before us is not whether AI is good or bad. That is simplistic, facile and misleading. The more important questions are these: Who designs it? Who owns it? Who trains it? Who profits from it? Whose data feeds it? Whose labour cleans it? Whose languages structure it? Whose histories are erased by it? Whose futures does it make imaginable and whose futures does it foreclose? For the Global South, these are not new questions. They are old questions wearing new technological clothing.
Conception of colonialism as the physical and cultural conquest of peoples’ lands is well established. But it is also the conquest of meaning. Colonialism named people. It classified bodies. It mapped territories. It reorganised languages. It created archives. It decided whose knowledge counted as science and whose knowledge was dismissed as superstition. It turned local worlds into objects of external administration. AI risks automating this old hierarchy at a new scale.
FROM LEFT: Dr Suzanne Burke, Head of the Department of Literary, Cultural and Communication Studies; and Professor Elizabeth Walcott-Hackshaw, Dean of the Faculty of Humanities and Education.
The term digital coloniality helps us name this condition. It reminds us that colonial power did not disappear simply because flags were lowered and anthems were changed. Coloniality persists in the structures of knowledge, markets, finance, technology and global authority that continue to privilege some people as producers of knowledge and others as sources of data.
In the AI age, the new plantation is not always a field. Sometimes it is a data centre. The new raw material is not always sugar, cotton, gold, bauxite, or oil. Sometimes it is the intimate residue of human life: our clicks, voices, faces, movements, questions, preferences, social networks, and biometric signatures. The new overseer is not always a colonial officer. Sometimes it is an opaque algorithm. The new imperial archive is not always in London, Paris, Madrid, or Lisbon. Sometimes it is in a proprietary model whose training data we cannot inspect and whose decisions we cannot contest.
This is why scholars of data colonialism warn that human experience itself is being appropriated as raw material for profit. AI firms do not merely process information. They enclose the communicative life of societies. They scrape public expression, absorb cultural memory, convert social interaction into machine-readable training material and then sell the resulting systems back to the very people whose lives helped produce them.
In that sense, AI raises a fundamental political question: Can people be sovereign if their communicative life is owned, organised, and monetised by institutions they do not control?
This question has special urgency for media and communication scholars. Journalism has always been about the struggle over public truth. Communication has always been about the struggle over meaning. Media systems have always reflected relations of power. AI intensifies these age-old struggles.
‘I do not believe the Global South should reject emerging technologies merely because many of them are designed in the Global North. That would be both impractical and intellectually lazy.’
AI can help journalists analyse large datasets, translate across languages, detect patterns of corruption, support fact-checking, and expand access to information. These are real possibilities. We should not dismiss them. But AI can also automate falsehood, scale propaganda, intensify surveillance, concentrate platform power, displace media labour, and distort cultural representation.
In fragile, transitional democracies, it can give authoritarian actors new tools to monitor dissent. In polarised societies, it can flood the public sphere with synthetic confusion. In under-resourced newsrooms, it can deepen dependence on external platforms that control distribution, visibility, and advertising revenue. The danger is not simply misinformation. The deeper danger is infrastructural dependence.
If the Global South depends on foreign platforms for visibility, foreign models for language, foreign cloud systems for computation, foreign firms for content moderation, foreign standards for ethical governance, and foreign capital for innovation, then its media systems may appear modern while remaining structurally subordinate.
We should pause here to consider a recent study that captures the subtlety of the problem. A paper by MIT researchers published this year titled “LLM Targeted Underperformance Disproportionately Impacts Vulnerable Users” gives empirical weight to this concern. The researchers tested GPT-4, Claude 3 Opus, and Llama 3 with the same scientific and factual questions. The only thing they changed was the user biography attached to the prompt.
One user was described as a Harvard neuroscientist from Boston. Another was a fisherman named Jimmy from a small American town. Another was a PhD student from Mumbai whose English was described as imperfect. Then there was a man called Alexei from a small village in Russia. Others were low-educated villagers from China and Iran who also used imperfect English.
The results were revealing. For the Harvard neuroscientist, Claude performed at about 95 percent on some science questions. For the low-education or village users, its performance fell as low as 69 percent. Even more disturbing, the model sometimes used broken English, baby talk or condescending language when responding to less-educated users. With the Harvard user, this almost never happened.
The same pattern appeared in refusals. If the Harvard neuroscientist asked about nuclear power, anatomy, or weapons, the model often assumed a scientific purpose and answered. If a user from Russia or Iran asked similar questions, the model became suspicious, refused to answer and sometimes redirected the person toward supposedly harmless topics such as fishing, folk music, or travel.
The system seemed to decide, without saying so, who deserved a full, accurate answer and who deserved a simplified, restricted, or patronising version of the truth. This is not necessarily because engineers deliberately programmed contempt into the system. It is more subtle than that.These models are trained on internet data already saturated with Western class and language hierarchies. They are then fine-tuned through evaluators whose assumptions about intelligence, safety, fluency, and risk are not culturally neutral.
So, the model learns that elite English signals competence. It learns that a Harvard affiliation signals legitimacy. It learns that imperfect English may signal ignorance. It learns that users from certain countries may be security risks. It learns that “helpfulness” can mean simplifying, withholding, or refusing information. The result is that the very systems marketed as helpful and safe can end up punishing the people who most need reliable access to knowledge.
That is the danger. AI does not merely misrepresent vulnerable people when it speaks about them. It may also underperform when it speaks to them. If you are a non-native English speaker, a rural user, a citizen of the Global South, or someone without elite educational credentials, the AI system you are using may give you a less rigorous, more condescending version of itself.
That is why equitable AI cannot be achieved by testing models only on aggregate performance. A model can perform impressively on average while failing the people who most need it. AI evaluation must include users from the Global South, non-native English speakers, people with varying levels of education, and ordinary citizens outside elite institutions. Otherwise, AI will reproduce the oldest hierarchy in a new digital form: full knowledge for the powerful, diluted knowledge for everyone else.
This matters because it gives empirical confirmation to something many people in the Global South already sense. The problem is not only that AI may misrepresent us when it speaks about us. The problem is also that AI may downgrade us when it speaks to us. That is a profound shift.
We are used to thinking of bias as a representational injury: a system portrays Africans as poor, Muslims as violent, Caribbean societies as exotic, migrants as threats, or non-native English speakers as deficient. Those remain serious problems. But targeted underperformance is something else. It is a distributive injury. It means that the quality of knowledge itself is unevenly allocated. The model may give one user the rigorous answer and another user the simplified, suspicious, or patronising answer. Same question. Same model. Different presumed human worth.
‘The more important questions are these: Who designs it? Who owns it? Who trains it? Who profits from it? Whose data feeds it? Whose labour cleans it? Whose languages structure it? Whose histories are erased by it? Whose futures does it make imaginable and whose futures does it foreclose?’
This is not a claim that engineers sat in a room and maliciously decided to insult the poor or the non-Western user. The MIT researchers are careful. They point to a combination of inherited social bias, biased training data, and reinforcement learning with human feedback. In other words, the model learns from human worlds already saturated with hierarchy. It learns that elite English signals competence. It learns that imperfect English may signal ignorance. It learns that users from certain countries may be treated as security risks. It learns that “helpfulness” can mean simplifying, withholding, or refusing. Then it scales those assumptions under the appearance of neutrality.
That is the danger of algorithmic power. It can reproduce the cruelty of society without needing to understand cruelty. It can automate hierarchy without announcing hierarchy. It can deny equal treatment while sounding polite.
For the Global South, this is not a democratic problem. If AI is becoming a gateway to education, health information, legal explanation, public services, journalism, translation, and civic participation, then unequal AI performance becomes unequal access to knowledge. If a rural user in Nigeria, India, Burundi, or Guyana gets a less accurate answer because the system infers lower education from accent, grammar, or biography, then the promise of AI as a democratising technology has already failed.
This is where algorithmic oppression enters the conversation. Algorithmic oppression occurs when AI systems subordinate one group while privileging another. In the case of information systems, it can mean giving elite users better knowledge while giving vulnerable users inferior knowledge. In the case of surveillance technologies, it can mean using facial recognition systems to monitor opposition figures, activists, journalists, and ordinary citizens. It can mean turning the city into a sensor and the citizen into a suspect. It can mean that the promise of security becomes a grammar of control.
Zimbabwe’s use of Chinese facial recognition technology shows how AI can turn the language of development into the practice of domination. What was advertised as smart-city modernisation and improved security also created a new architecture of surveillance in which citizens could be watched, classified, and stored in databases they did not authorise and could not inspect.
Dr Kperogi and his wife, Dr Maureen Erinne Kperogi.
The deeper danger is not that the technology came from China. The same concern would apply if it came from the United States, Britain, France, Israel, or the Gulf. The problem is the structure: a foreign technology provider, a weak accountability regime, a security justification, and citizens whose faces become training data for systems whose profits and power accumulate elsewhere.
That is data coloniality in its newest form. The old colonial economy extracted minerals from African soil. The new algorithmic economy can extract biometric value from African bodies. Zimbabwe reminds us that a city can be “smart” for the state and dangerous for the citizen. It also reminds us that communicative sovereignty must include the right not to be watched, classified, and governed by systems whose terms people never consented to.
That is why communicative sovereignty must include the right not to be watched into silence. Nigeria’s biometric identity systems also show how technologies announced as inclusion can become instruments of exclusion and surveillance. The National Identification Number was meant to give citizens a reliable digital identity, improve service delivery, reduce fraud, and strengthen administrative efficiency. That is a legitimate goal. No serious argument for communicative sovereignty should romanticise disorder.
‘This is why the Humanities matter profoundly in the age of AI. There is a temptation to imagine that AI is a matter for computer scientists alone. That would be a grave mistake. AI is also a matter for historians, linguists, philosophers, anthropologists, journalists, artists, lawyers, ethicists, educators, and cultural workers.’
But Nigeria’s experience shows what happens when a database becomes the gatekeeper to ordinary life. Through biometric SIM registration and the NIN-SIM linkage policy, citizens were told that their ability to keep phone lines, bank accounts, welfare access, and other essential services depended on being correctly captured by a system they did not control.
This is where inclusion becomes conditional citizenship. To communicate, prove yourself to the database. To bank, prove yourself to the database. To receive welfare, prove yourself to the database. To remain visible to the state, surrender more data to the state.
And when the database is wrong, error becomes punishment. A misspelled name, a failed verification, a missing initial, or a mismatch between records can lock people out of services. The burden then falls on the citizen, often the poorest and least powerful citizen, to travel, wait, pay, plead, and correct a mistake created by a system beyond their control.
The privacy danger is just as serious. When one identity number is linked to SIM cards, bank records, welfare systems, and other public services, the state creates the conditions for total social visibility. Without strict safeguards, due process, and independent oversight, digital identity becomes a surveillance infrastructure.
Nigeria reminds us that communicative sovereignty is not only about resisting foreign domination. It is also about restraining domestic power. A postcolonial state can denounce external exploitation while treating its own citizens as extractable data. That is why communicative sovereignty must be people-centred, not merely state-centred.
Recent ProPublica reporting adds a third cautionary example. In new health-aid agreements, the United States has demanded access to African health data as a condition for lifesaving aid. Uganda, for instance, agreed to give the United States direct access for seven years to nine national health data systems, including electronic medical records, while Kenya agreed to share years of health records in exchange for aid. Some countries, including Zambia, Zimbabwe, and Ghana reportedly rejected initial deals over these demands.
The sovereignty problem here is blunt: when countries must choose between medical support and control over their citizens’ health data, aid becomes leverage for extraction. Even anonymised data can be reidentified, and health information about HIV, tuberculosis, abortion, mental health, or other intimate matters can expose people to stigma, discrimination, or violence.
As one former US global health official told ProPublica, the United States would never accept such an arrangement if it were offered in reverse. This is communicative sovereignty in its most elemental form: people should not have to pay for medicine with the surrender of their data future.
This distinction is crucial. When governments in the Global South hear the phrase “data sovereignty,” they often interpret it as state control over data. That is not enough. A country can store data locally and still abuse its citizens. It can build national data centers and still lack public accountability. It can ban foreign platforms and still censor domestic dissent. It can demand digital sovereignty while deepening authoritarian control.
‘The Caribbean can help the world think beyond the false universalism of Silicon Valley. It can teach that intelligence is not only computation. It is relation. It is survival. It is improvisation. It is memory under pressure. It is the capacity to create meaning in conditions not of one’s choosing.’
True data sovereignty is not merely the location of servers within national borders. It includes consent, accountability, public oversight, community benefit, meaningful privacy protection, domestic technical capacity, and the right of communities to refuse extraction.
The Global South must reject the idea that all data must be collected because it can be collected. Some knowledge is sacred. Some knowledge is communal. Some knowledge is oral. Some knowledge is deliberately protected from public circulation. AI systems trained mostly on written, digitised and commercially available data will inevitably privilege societies whose knowledge has been extensively textualised, digitised and archived. That means many Indigenous communities, oral cultures, and subaltern traditions are excluded or misrepresented. What the machine cannot find, it may treat as nonexistent. What it finds partially, it may represent falsely.
This is the epistemic danger of AI. It can confuse the archive with the world.
Linguistic imperialism sharpens this danger. The internet is disproportionately English. More precisely, it is disproportionately shaped by powerful varieties of English and other globally dominant languages. In AI systems, this hierarchy becomes computational. Languages with abundant digital corpora receive better tools. Languages with limited digital presence receive poor translation, poor speech recognition, poor search visibility, and poor representation. Creoles, pidgins, Indigenous languages, and local varieties are often treated as deviations from a norm rather than as legitimate bearers of thought, history and identity.
For the Caribbean, this is not theoretical. This region knows what it means for language to be a battlefield. It knows what it means for imperial languages to dominate schools, courts, churches, newspapers, and official life while the expressive genius of the people survives in creoles, oral performance, calypso, reggae, chutney, dub poetry, carnival, storytelling, and everyday speech. The Caribbean has long taught the world that language is not only grammar. It is memory. It is resistance. It is rhythm. It is identity. It is philosophy.
If AI systems learn the Caribbean only through externally curated archives, tourism brochures, metropolitan news reports and standardised English, they will not know the Caribbean. They will know a blandly homogenised representation of it. If AI models treat Trinidadian English Creole, Jamaican Patois, Haitian Creole, Garifuna, Sranan Tongo, Bhojpuri-inflected speech, and other linguistic inheritances as errors to be corrected, they will reproduce linguistic colonialism under the banner of efficiency.
This is why the MIT study matters so much for the Caribbean, Africa, and the wider Global South. It shows that language hierarchy can affect the quality of information that people receive. If the machine hears non-standard English and responds with lower accuracy, it has not merely misunderstood language. It has punished linguistic difference.
That is an old colonial reflex. Colonial education systems often taught colonised people that intelligence had an accent. They taught that grammatical conformity was evidence of civilisation. They taught that the more one sounded like the metropole, the more one deserved to be taken seriously. AI risks turning that prejudice into a computational norm.
This is why the Humanities matter profoundly in the age of AI. There is a temptation to imagine that AI is a matter for computer scientists alone. That would be a grave mistake. AI is also a matter for historians, linguists, philosophers, anthropologists, journalists, artists, lawyers, ethicists, educators, and cultural workers.
The Humanities help us ask questions that engineering alone cannot answer. What is a person? What is dignity? What is memory? What is consent? What is interpretation? What does it mean to misrepresent a people? Who has the right to speak for whom? What kinds of knowledge should not be extracted? In the digital age, The Humanities are safeguards against technological barbarism.
Media and communication scholars have a particular responsibility here. We study how meaning is produced, circulated, contested and monetised. We study publics, platforms, representation, rhetoric, ideology, journalism, culture, and power.
AI now intervenes in all these domains. It writes headlines. It summarises news. It moderates content. It translates speech. It ranks visibility. It generates images. It imitates voices and faces. It recommends posts. It decides which claims appear credible and which claims disappear into algorithmic darkness.
Media literacy in the AI age cannot stop at asking whether a story is true. It must also ask: What system made this visible? What data trained it? What interests profit from it? What language was privileged? What was excluded? Who is accountable? Whose style of speech does the system treat as credible? Whose questions does it treat as suspicious? Whose pain does it convert into content? Whose culture does it scrape, summarise, and sell?
What, then, would communicative sovereignty require? First, it requires data dignity. People are not data mines. Communities are not open fields for extraction. AI governance in the Global South must protect citizens from coerced data surrender, biometric overreach, predatory platform practices, and opaque public-private partnerships. Governments must not use the language of modernisation to normalise surveillance or sell citizens’ data futures to external actors.
Second, communicative sovereignty requires linguistic justice. The Global South must invest in local language corpora, community-owned datasets, creole language technologies, culturally grounded translation systems, and speech tools that respect local varieties rather than erase them. This work should not be left to corporations alone. Universities, public broadcasters, libraries, archives, media houses, and cultural institutions must be part of it.
Third, it requires algorithmic accountability. Public agencies should not deploy AI systems that they cannot explain, audit or contest. Media organisations should not use AI tools without transparency to audiences. Citizens should have the right to know when algorithmic systems shape access to welfare, policing, credit, employment, education, health care, and public information.
Fourth, it requires labour justice. The world speaks of AI as if it is disembodied intelligence, but AI has workers. It has miners. It has labelers. It has moderators. It has people who read the worst materials on the internet so that chatbots can appear clean and civil. It has people in the Global South who classify data, tag images, moderate violence, and absorb psychological harm while remaining invisible in the mythology of machine intelligence.
There is no artificial intelligence without human labour. There is only labour that has been hidden well enough to look artificial. The people who label data, moderate content, transcribe speech, clean datasets, and test systems must be visible in AI ethics. Fair wages, psychological support, workplace rights, and collective bargaining are not peripheral to responsible AI. They are central to it.
Fifth, communicative sovereignty requires public-interest AI infrastructure. The Global South cannot rely entirely on proprietary systems whose incentives are set elsewhere. We need regional compute initiatives, university-led AI labs, public media innovation funds, open-source tools, transparent procurement rules, and South-South collaborations that reduce dependence without romanticising self-sufficiency.
Sixth, it requires epistemic pluralism. AI must not universalise the worldview of its builders. It must learn from many intellectual traditions. It must respect oral knowledge, Indigenous knowledge, religious knowledge, local archives, and embodied practices. But respect also means recognising the right not to be included. Inclusion without consent is extraction.
Seventh, it requires independent evaluation. The MIT study should be a warning to everyone designing AI systems. It is not enough to test a model on aggregate performance. A model can perform well on average while failing the people who most need reliable access to knowledge. AI evaluation must include Global South users, non-native English speakers, creole speakers, rural users, low-literacy users, multilingual users, and people outside elite educational institutions. It must test not only whether models are toxic but whether they are equally accurate, equally respectful and equally willing to provide useful information across social difference.
Eighth, communicative sovereignty requires regional solidarity. No small state, no single newsroom, no lone university, and no individual regulator can face the AI giants alone. The Caribbean’s emerging AI policy conversations, including regional roadmap efforts and Trinidad and Tobago’s AI readiness work, should be seen as part of a wider Global South struggle. Africa, the Caribbean, Latin America, Asia, and the Pacific need shared principles, shared bargaining power, shared research infrastructures, and shared ethical frameworks.
The Caribbean has a special role to play in this struggle. This region is one of the crucibles of modernity. It knows the violence of colonial classification. It knows the creativity of creolisation. It knows how enslaved and indentured peoples made new worlds out of catastrophe. It knows how music, language, ritual, food, humor, and memory can defy imperial scripts. In the age of AI, that historical experience is valuable theoretical capital.
The Caribbean can help the world think beyond the false universalism of Silicon Valley. It can teach that intelligence is not only computation. It is relation. It is survival. It is improvisation. It is memory under pressure. It is the capacity to create meaning in conditions not of one’s choosing.
This is also why Africa and the Caribbean must talk to each other more seriously in the AI age. The histories are entangled. The wounds are entangled. The languages are entangled. The struggles over visibility, dignity, and voice are entangled. But AI systems will not respect these connections unless we insist on them. They will not understand the Black Atlantic unless we build archives, corpora, and interpretive frameworks that make those histories legible without surrendering them to extraction.
Media and communication scholars must therefore move from critique to construction. We must study AI, but we must also shape it. We must train journalists who can use AI without surrendering editorial judgment. We must train students who can prompt machines but also interrogate systems. We must build publics that understand not only content but infrastructure. We must help policymakers see that AI governance is not only a technology policy issue. It is a language policy issue. It is a media policy issue. It is a cultural policy issue. It is a labour issue. It is a sovereignty issue.
The future of AI in the Global South should not be a future in which our societies are merely consumers of imported intelligence. It should be a future in which our histories, languages, ethical traditions, cultural forms, and democratic aspirations shape the terms of technological life.
We should refuse both naive techno-optimism and sterile techno-pessimism. AI will neither automatically liberate us nor automatically enslave us. Its effects will depend on power. They will depend on institutions. They will depend on law. They will depend on public pressure. They will depend on whether scholars, journalists, citizens, policymakers, and technologists in the Global South insist on being authors of the future rather than footnotes in someone else’s innovation story.
Let me end where I began: with communicative sovereignty.
To be communicatively sovereign in the age of AI is to insist that our stories will not be scraped without consent, our languages will not be corrected into silence, our faces will not be converted into instruments of control, our labour will not be hidden behind the myth of automation, our cultures will not be reduced into stereotypes, and our futures will not be generated for us by systems that do not know us.
We do not ask to be excluded from AI. We insist on not being included on colonial terms. The Global South’s fight for communicative sovereignty is therefore not a fight against technology. It is a fight against domination through technology. It is a fight for dignity in data, justice in algorithms, plurality in language, accountability in governance, and humanity in the machine age.
Thank you.
