AI and Knowledge: The Moral Imagination of the Academy

Good morning, good afternoon, and good evening, colleagues. It is a great honor to join this global dialogue convened by UNESCO and the IAU on AI and Knowledge: Tensions Between the Common Good and Commodification. My remarks today explore how AI is reshaping the political economy, ethics, and purpose of higher education, and what responsibilities universities bear as stewards of knowledge in a rapidly transforming world.

The 4Is if Knowledge Production

To understand how AI is altering the terrain of knowledge, we must first recall that knowledge production has always been structured by power. It is not an even field, but a global system shaped by what I call the 4Is: institutional, intellectual, ideological, and individual dynamics.

First, institutional dynamics define the architecture of knowledge itself. Universities, academies, private laboratories, and digital platforms form the sites where knowledge is produced. Their governance, resources, and alliances vary enormously, from Silicon Valley’s corporate research clusters to publicly funded universities in Africa, Asia, or Europe and these conditions determine what is studied and who participates.

Second, intellectual dynamics govern the movement and legitimacy of ideas. Theories and methodologies circulate widely, yet they are not equally valued. In much of the world, Western epistemologies often set the norms of scientific legitimacy, shaping how ideas are validated and reproduced.

Third, ideological dynamics reveal that knowledge is never neutral. It reflects political and economic structures, whether Marxist, neoliberal, or postcolonial, and each paradigm encodes a worldview that influences AI’s design and deployment.

Finally, individual dynamics remind us that scholars themselves are embedded in systems of privilege and exclusion. Their identities, including nationality, gender, race, class, among other social inscriptions, continue to structure access to funding, mobility, and recognition within global research networks.

These interlocking forces remind us that the knowledge ecosystem is already stratified, and AI is entering this world not as an equalizer, but as an accelerator of both opportunity and inequality.

In addition, we must attend to emergent structural constraints that condition how the 4Is play out under AI. For example, the growing “compute divide” between well-resourced institutions and those lacking hardware intensifies institutional and individual disparities. Studies show that academic-only research teams are underrepresented in compute-intensive work, revealing how the lack of infrastructure restricts the epistemic possibility of those outside elite labs. Similarly, the de-democratization of AI literature traces how limited access to GPU clusters and cloud credits means that capacity, not merit, often dictates which scholars are visible. This deepens institutional inequality by ensuring that powerful knowledge infrastructures remain concentrated.

Moreover, the growing literature on epistemic injustice in AI shows how algorithmic systems can degrade the credibility and voice of marginal knowledge traditions. The concept of generative algorithmic epistemic injustice underscores the fact that AI can amplify testimonial prejudice, suppress interpretive resources, and enforce access inequality, mechanisms by which collective knowledge systems become skewed toward dominant paradigms. When AI adopts a “view from nowhere,” it risks hermeneutical erasure, the gradual erosion of non-Western epistemologies. In that sense, the 4Is are undergirded not just by material power, but by epistemic technologies that invisibilize certain forms of knowledge in the name of universality.

Thus the 4Is function not merely as categories but as interactive fault lines: institutional resources ground epistemic capacity, intellectual canons gate meaning, ideological frameworks legitimize certain designs, and individual identities mediate access. AI introduces a new vector of acceleration through compute, algorithmic authority, and infrastructure governance, but does so within a preexisting architecture of inequality.

Features and Implications of AI

AI’s rise represents not merely a technological revolution but a profound epistemic shift, transforming how research is imagined, conducted, and disseminated across disciplines and borders.

First, in research conception and design, large-scale data analytics and simulation tools are reshaping discovery itself. A Science Advances study of 15 million PubMed abstracts found that roughly 13.5 percent of 2023 biomedical papers show signs of LLM assistance, suggesting a measurable shift from intuition to computation.

Second, the processes of writing, peer review, and validation are undergoing rapid transformation. Generative AI accelerates translation, drafting, and review, yet raises new issues of originality, attribution, and trust. Editors now encounter “prompt injection” and AI-generated referee reports. Scholarly bodies advocate documented disclosure logs that disclose what model was used, when, and for what task to safeguard originality and trust.

Third, in publication and dissemination, algorithmic visibility increasingly determines impact. Altmetric data show that scholarly attention once monopolized by X (Twitter) is migrating to Bluesky, revealing how platform politics influence research diffusion.

Fourth, authorship and attribution have become contested territories. Policies differ globally; some academic systems now require disclosure of AI use, while others debate whether algorithms can be co-authors. The 2025 Bartz v. Anthropic settlement confirms authorship and credit as legal frontiers. South Africa’s ASSAf code now treats AI as a non-author and mandates transparent disclosure of its use.

Finally, the opportunities and risks of AI coexist in a fragile balance. Collaboration, translation, and innovation flourish alongside surveillance, bias, and monopolization of AI infrastructure. While AI broadens collaboration and translation, it also intensifies environmental and infrastructural inequalities as training large models consumes hundreds of thousands of liters of water and centralizes compute capacity in a few regions.

So while AI expands capacity, it also reshapes control, prompting us to examine how power and resources flow through this new political economy of knowledge.

To deepen that point, we can draw on critiques from fairness research that connect representational harm to epistemic foundations. Several scholars show how dominant ML fairness approaches often ignore the epistemic assumptions embedded in data and models, thereby reinforcing structural injustice. By not interrogating those assumptions, AI risks formalizing inequities in inference itself rather than merely offsetting them with fairness constraints. Further, algorithmic practices such as automated benchmarking and evaluation mirror existing knowledge gaps: The Algorithmic Construction of Epistemic Injustice (SSRN) demonstrates how LLMs reflect dominant knowledge taxonomies in data ingestion and evaluation, further entrenching hermeneutical bias.

Moreover, the phenomenon of techno-linguistic bias documented in some studies reveals how most AI language systems support only a handful of global languages. Only 2–3 % of the world’s languages are materially supported, and many design decisions assume universal conceptual structures that fail to translate non-Western worldviews. This kind of linguistic homogenization erases epistemic difference in favor of majoritarian norms, an invisible form of colonial logic.

These insights compel us to see the features of AI not as mere tools or risks, but as epistemic interventions, which alter how knowledge is validated, attributed, and disseminated. Recognizing that helps us frame reform not as patchwork, but as structural redesign.

AI and Global Knowledge Hierarchies

AI both mirrors and magnifies the world’s knowledge hierarchies, deepening what I have elsewhere called epistemic divides—the structural imbalances between those who produce knowledge and those who consume it.

To begin with, institutional funding patterns increasingly favor marketable, technology-oriented research, privileging countries with strong R&D investment and private-sector collaboration. A Digital Science survey shows that China produced over 40 percent of global AI research in 2024, surpassing the US and EU combined and redirecting where agendas and citations originate.

Moreover, disciplinary and regional gatekeeping persists, with English remaining the lingua franca of science. The “open-to-read ≠ open-to-train” debate, as some have argued, exposes how open-access research can be exploited for training proprietary models without consent or credit, an irony that turns openness into extraction.

Additionally, technological infrastructures, including compute capacity, cloud access, and data repositories, define who can meaningfully participate in AI research. Public agencies are attempting to adapt. For example, the NIH now rejects grant applications written largely by AI and caps per-investigator submissions to preserve human review integrity. Yet dependency on corporate clouds and closed APIs continues to shape who can participate and who remains excluded.

At the same time, diasporic and transnational scholars play a crucial bridging role in this evolving landscape. New African platforms such as NITheCS and the African Research and Innovation Hub seek to pool data and computing resources for continental research sovereignty, challenging North–South asymmetries and expanding the space of global collaboration.

In the end, the result is a paradox: AI amplifies both innovation and inequality, creating a new digital epistemic divide. As some scholars warn, AI can erode the trust that underpins peer review and intellectual exchange if authorship and accountability remain opaque. The divide now separates those who design and direct AI from those who must depend on it, a hierarchy of knowledge that risks hardening into permanence.

If the knowledge economy is to remain a global commons rather than a hierarchy, we must confront how AI reconfigures access, authorship, and authority before these inequities become the foundation of the next intellectual order.

Beyond these observable patterns, we should reflect on how capital accumulation in AI research reshapes priorities. Several studies demonstrate how patenting and venture funding skew research toward commercially exploitable outcomes, often at the expense of public-interest questions. These financial pressures amplify institutional biases: resource allocation follows market logic, not epistemic or social need. This intensifies epistemic divides by rewarding the commercially viable rather than the socially essential.

Decolonization and Reclaiming Epistemic Integrity

The diffusion of AI reopens long-standing questions about epistemic authority; about whose knowledge counts, whose is rendered invisible, and who gains or loses agency in shaping the future of reason itself.

To begin with, colonial legacies continue to shape epistemic hierarchies. AI does not emerge in a vacuum; it inherits the architectures of power built into older systems of classification and extraction. The same hierarchies that once governed empire now govern data—who collects it, who labels it, and whose labor remains unseen. Decolonizing AI therefore means interrogating the economic and institutional conditions that reproduce global dependency while masking it as technical progress.

Moreover, we must confront the growing reality of algorithmic colonialism. Despite its claim to universality, AI still reflects narrow linguistic and cultural assumptions. Most large models are trained on Western or state-curated corpora that standardize certain worldviews and erase others. Algorithmic colonialism operates through both the data and the logic of design, embedding invisible hierarchies into systems that appear neutral. Addressing it requires confronting structural bias at its source by diversifying data provenance, valuing translation, and creating accountable pathways for inclusion in model development.

In this sense, achieving epistemic justice becomes a question not only of fairness but of sovereignty over meaning itself. Communities must retain the right to define how their knowledge and data are represented and used. Ethical AI depends on relational accountability, encompassing consent, context, and co-creation, rather than extraction. Epistemic justice calls for a shift from treating knowledge as raw material to recognizing it as a living relationship grounded in responsibility and respect.

Additionally, the challenge of plural governance invites us to rethink the ethics of global standard-setting. Efforts to regulate AI often rely on universalist principles that obscure the diversity of moral reasoning across societies. True plural governance begins by acknowledging multiple ethical traditions, such as Ubuntu’s emphasis on interdependence, Indigenous reciprocity, Confucian harmony, and creating deliberative spaces where these values shape technical standards. Instead of one global code of ethics, we need intersecting frameworks of care that balance innovation with accountability to human and ecological communities.

Finally, democratization of knowledge remains central to any just future for AI. Behind every model lies human labor, including data collectors, annotators, researchers, whose contributions are rarely acknowledged. Democratizing knowledge means expanding participation and recognition across the entire research chain. Open infrastructures, equitable access to data and computing power, and fair attribution systems are essential if AI is to reflect, rather than erase, the multiplicity of human creativity.

Decolonizing knowledge, then, is not a fancy metaphor. It is a practical and moral imperative, to ensure that AI becomes a space of dialogue rather than domination, and that the future of knowledge mirrors the diversity of the world itself.

Building on this imperative, emerging scholarship identifies novel forms of epistemic injustice specific to generative AI systems. For instance, a recent taxonomic study argues that generative hermeneutical erasure occurs when an AI’s standard conceptual vocabulary suppresses local or minority epistemologies by simply failing to represent their categories. This is not just omission, it is active conceptual marginalization. Similarly, the notion of algorithmic profiling as a source of hermeneutical injustice shows how personalization systems can deprive individuals of interpretive resources necessary to understand their own experiences, thus further narrowing epistemic agency. By bringing these analytical lenses, we strengthen the case that decolonizing knowledge must be oriented not just to inclusion, but to epistemic resilience, structurally resisting erasure by design.

Envisioning a Plural AI Future

The trajectory of AI, whether it consolidates control or diversifies participation, remains an open question, and one that demands collective stewardship.

At the outset, the paradox of openness and enclosure captures one of AI’s most troubling contradictions. Even when AI is presented as open, through open source, open data, or translation tools, it often reinscribes exclusivity. The infrastructure of translation, hosting, and interface remains proprietary. Without democratized compute and control, openness can easily become a Trojan horse: open data feeding closed, centralized systems that deepen dependency rather than dismantle it.

Equally pressing is the ongoing crisis of access. Deep learning has become so resource-intensive that only elite institutions and firms can sustain the demands of large-scale models. This escalating “compute divide” limits who can meaningfully contribute to frontier AI research. What began as a scientific challenge has evolved into a structural one, where capacity, not creativity, determines participation.

Furthermore, new asymmetries are emerging as power in AI becomes concentrated in a handful of corporate and national actors who control both the infrastructure and the agendas of research. This consolidation raises urgent questions of sovereignty: who sets the norms, who builds the platforms, and whose questions ultimately count? The dominance of a few players risks converting once-open fields into gated domains, narrowing the very horizons of discovery that AI was meant to expand.

At the policy level, global initiatives have begun to offer counterweights. The UNESCO Recommendation on the Ethics of AI and new higher-education open-science frameworks represent vital footholds for shared governance. Scholars emphasize that these multilateral instruments must go beyond aspirational language; they should foster accountability, inclusion, and legitimacy in how technologies are designed, distributed, and assessed.

Ultimately, we confront the core dilemma: will AI evolve into a truly global commons, an infrastructure shaped by many, or into a privatized knowledge marketplace that deepens dependency? Current trajectories reveal a persistent tug-of-war: technologies that enable shared access (cloud, open models, federated systems) coexist uneasily with forces of exclusivity (closed models, licensing, compute scarcity). The outcome will determine whether knowledge is shared or hoarded, whether AI amplifies human plurality or entrenches digital hierarchy.

In short, the question before us is not only what technology can do, but who decides what it should do, and for whom.

To deepen this analysis, recent discussions of the compute divide highlight how uneven access to computational infrastructure reshapes the landscape of AI research. As large models demand immense resources, industry laboratories increasingly dominate the frontier, setting the technical and ethical parameters that others must follow. This shift narrows the range of participants and perspectives, creating a cycle where innovation and governance become concentrated in the same few hands. Rather than a mere question of resource allocation, this dynamic exposes the need for reflexive mechanisms that hold both systems and institutions accountable for the assumptions they encode. Emerging proposals for epistemic audits, systematic evaluations of how data, design choices, and performance metrics reflect social hierarchies, seek to make these structures visible. Such audits invite deeper scrutiny not just of what models produce, but of how they construct meaning, whose voices they privilege, and whose experiences they omit. In doing so, they point toward the possibility of a more plural and transparent AI infrastructure capable of sustaining collective participation rather than monopolized authority.

Ethics and the Knowledge Commons

At its heart, this debate is about conscience, the moral imagination of the academy in an algorithmic age. AI challenges not only what we know, but how we know, and to what ends that knowledge is used. Universities, long the custodians of critical inquiry, must now decide whether they will remain stewards of knowledge or become spectators to its commodification.

To begin, we must safeguard knowledge as a universal public good. Beyond national and corporate boundaries, “open science” requires thoughtful guardrails. Openness cannot mean appropriation; open access must include consent, attribution, and traceable links to original authors so that models amplify rather than extract human labor. Protecting the public nature of knowledge ensures that AI strengthens, rather than supplants, the human contribution that underpins all scholarship.

Equally important is the need to promote AI literacy that integrates ethics and civic reasoning. Frameworks such as UNESCO’s Recommendation on the Ethics of AI and various regional and national policies and codes offer templates for responsible research practice across regions. Yet AI literacy must extend beyond technical competence by cultivating civic responsibility, moral reasoning, and epistemic humility, embedding these values at the core of global curricula. Education, in this sense, becomes both a safeguard and a form of resistance to unreflective automation.

In addition, the academy must adopt transparent governance for data stewardship. Environmental impact should become a standard reporting metric; after all, a single training run can consume as much water as a small campus. Disclosure of such costs is vital to ethical accounting. Transparency must also address emerging vulnerabilities, such as prompt injection, the deliberate manipulation of AI inputs to distort scholarly or policy outputs, which threatens the credibility of digital knowledge systems.

Furthermore, transregional collaboration offers a path toward equity and balance. Pan-regional initiatives in Africa and Asia, for instance, demonstrate how shared computing and open-data infrastructures can reclaim collective agency in global research. Such partnerships ensure that local priorities inform global innovation agendas, transforming participation from dependency into co-creation.

Finally, universities and multilateral organizations must be empowered as stewards of the digital commons. Building public AI infrastructure and inclusive data standards provides a counterweight to corporate enclosure. These institutions can uphold epistemic diversity as the hallmark of global scholarship and renew the academy’s historic mandate: to safeguard knowledge as a public trust, open to scrutiny and shared for the common good.

To sustain the moral mission of higher education, universities must act not as passive consumers of technology but as architects of ethical and inclusive knowledge futures. The conscience of the academy lies not in its algorithms but in its ability to align intelligence with integrity, to ensure that innovation remains accountable to humanity.

Reclaiming Intelligence and Responsibility

As we arrive at the end of this reflection, I return to the central question: how can education guide intelligence with wisdom, connect knowledge with justice, and ensure that technology serves humanity? The answer lies not in the capabilities of machines but in the commitments of the human mind, its capacity for discernment, empathy, and moral courage.

To begin, reclaiming the human is our first imperative. AI’s extraordinary power to simulate cognition compels us to re-examine what cannot be simulated, namely our capacities for empathy, imagination, and ethical judgment. Several scholars remind us that moral agency is not computational but relational; it is exercised in care, dialogue, and responsibility. To reclaim the human, then, is to affirm that intelligence is more than information processing; it is the practice of discernment within community, the ability to know not only what can be done but what should be done.

Second, we must reframe knowledge itself. As AI becomes a producer of text, data, and even hypotheses, the university’s role must be to safeguard judgment, the capacity to ask questions that no algorithm can pre-empt. Philosophers of science call this the cultivation of “technologies of humility”: a deliberate practice of reflexivity about what we do not know and about the consequences of what we create. Inquiry must remain a human craft, grounded in doubt, curiosity, and the courage to think beyond automation.

Moreover, the pursuit of knowledge must advance justice. Ethical governance is inseparable from social justice. Research on algorithmic accountability exposes how bias is encoded not only in code but in commerce, policy, and social structure. Governance, therefore, must move beyond technical audits toward structural reform, embedding transparency, equity, and participatory oversight at every stage of AI development. In this sense, justice is not an accessory to design; justice is design.

At the same time, AI challenges us to expand the epistemological imagination. It invites us to widen the very meaning of intelligence. Cognitive diversity, from Indigenous relational epistemologies to ecological systems thinking, offers alternative grammars of knowing that resist the narrowing tendencies of algorithmic logic. Some scholars describe this as an “ecology of knowledges,” an approach that values plurality over hierarchy. Education’s great task is to weave these diverse epistemologies into a global conversation rather than condense them into a single algorithmic canon.

Finally, we must build an ethic of sustainability and stewardship. Every algorithm carries a material footprint. Training one large-language model can consume millions of liters of water and immense amounts of energy, as documented in numerous studies. An ethic of stewardship demands that universities lead by example, developing greener computing, transparent reporting of resource use, and curricula that link digital innovation with ecological responsibility. Sustainability is not merely environmental; it is epistemic, about how we care for both the planet and the knowledge that sustains it.

In closing, let us remember: AI will not determine our future; our values will. The true measure of its success will not be speed or precision, but whether it enlarges humanity’s capacity for truth, justice, and shared flourishing. That is the conscience of the academy, and our collective responsibility.

Thank you.

Please support the research and time that goes into preparing these essays by subscribing to the paid version of the Substack.