Bias, Cultural Power, and Digital Colonialism

Artificial intelligence can distribute capability more widely while concentrating infrastructure in fewer hands.

Both realities matter. The concern is larger than biased outputs. It includes representation, data control, infrastructure, theological authority, and the ability to exit.

Bias Is More Than Offensive Output

Language AI performs unevenly. High-resource languages benefit from larger corpora, stronger benchmarks, more user feedback, and commercial incentives.

A model described as multilingual may still serve some communities poorly. Inclusion therefore requires deliberate investment rather than assuming general model improvement automatically reaches every language.

Bias discussions often focus on obviously prejudiced text. Cultural power can operate through subtler patterns: which topics receive detailed answers, which languages receive strong performance, whose institutions are treated as authoritative, and which categories the system assumes are universal.

A polite output can still reproduce structural imbalance. Benchmarks define what counts as success. If low-resource language systems are evaluated only through automatic similarity metrics, community judgments about naturalness or register may remain invisible. Evaluation should include criteria that matter to actual users.

Who Controls the Data and Infrastructure?

Mission organizations may hold Scripture corpora, dictionaries, recordings, oral histories, and translation notes in languages with little other digital representation.

Possession does not automatically establish unrestricted moral permission for machine-learning reuse. Communities may have collective interests in how language and cultural data are repurposed.

Ask who created the material, under what expectations, who benefits, whether the community can refuse new uses, and whether local institutions can access what is built from it.

Representation without governance is not the same as agency. A local institution may control its content while depending entirely on an external model provider for inference.

This may be acceptable. Agency does not require owning every layer. The key is knowing where control ends, what alternatives exist, and whether dependence creates unacceptable vulnerability.

Theological Power Travels Through Corpora

Which dialect enters the corpus? Which orthography becomes standard? Whose pronunciation is treated as normal? A system may technically support a language while sounding elite, foreign, regional, or religiously marked.

Local speakers need authority to evaluate whether the system belongs in their linguistic environment.

If theological AI draws primarily from English-language North American or European sources, it may speak as though one digital tradition is generic Christianity.

The solution is not to reject Western theology. It is to make sources visible, broaden the corpus, and strengthen local theological production and governance.

A retrieval system can make diversity explicit by labeling source tradition, geography, date, and author rather than blending everything into one answer.

The user can then see when the system is drawing mainly from one denominational or regional stream.

Transparency is often more realistic than pretending to construct a perfectly neutral corpus.

Agency, Colonialism, and the Right to Refuse

The Latin American and Caribbean Catholic episcopal body CELAM has developed its own regional pastoral and theological reflection on artificial intelligence, including dignity, community, technocratic power, and the need for inculturation.¹

Its position does not represent every Latin American Christian. Its existence demonstrates that Christian reflection on AI is already being produced from within the region rather than merely exported to it.

Use the term diagnostically, not as a synonym for foreign technology. It becomes useful where several features combine: External priorities; Extraction of data or knowledge; Durable dependency; Weak local control; Externally imposed norms; Poor ability to refuse or exit. The positive goal is not simply technological inclusion. It is capability plus agency. AI may give underserved Christian communities access to tools historically beyond their resources. The strongest projects also increase their ability to evaluate, govern, modify, maintain, and reject those systems.

Build in ways that redistribute agency along with capability. One of the clearest tests of dependency is whether a community can leave.

Can data be exported? Can another system be substituted? Can the organization continue basic ministry without the platform?

Poor exit capacity turns convenience into leverage. Digital colonialism becomes a more precise diagnostic when it asks who can say no.²

African, Asian, Latin American, Indigenous, and diaspora Christians should appear as thinkers, builders, and governors rather than only beneficiaries. The long-term test is what grows around the tool. The ministry may gain speed while losing understanding, or it may use speed to create margin for better judgment and relationship. Success should therefore be evaluated over time through competence, accountability, resilience, and service rather than through the first impressive demonstration.

PART VI

Footnotes

1. CELAM is a Latin American and Caribbean Catholic episcopal body; these works demonstrate regional Christian agency but are not presented as a consensus position for all Latin American Christianity. See Consejo Episcopal Latinoamericano y Caribeño (CELAM), La inteligencia artificial: Una mirada desde América Latina y el Caribe; Consejo Episcopal Latinoamericano y Caribeño (CELAM), Inteligencia artificial y desarrollo humano integral ↩

2. Kwok, “Data Colonialism and Indigenous Languages in AI: A Critical Review of Existing Initiatives and Their Struggles with Data Sovereignty.” ↩