AI in Missionary Practice
Voice, Orality, and Accessibility
~6 min read
Text is only one interface to language. That fact can disappear inside discussions of generative AI because the modern AI boom arrived through chat boxes. For many communities, however, speech is the dominant medium of daily communication. For Deaf communities, sign languages possess their own linguistic structures. For people with visual, motor, cognitive, or literacy-related access needs, a text-first ministry can create barriers even when the written content is accurate.
Artificial intelligence can help because it increasingly works across modalities. Speech recognition can convert audio to text. Text-to-speech can produce listening versions. Vision systems can describe images. Language models can simplify or restructure content. Sign-language technologies can assist production and revision. The opportunity is real. Accessibility should not be inferred from format conversion alone.
Mission Beyond Text
The Ciyawo translation project is useful because it treats oral work as a primary translation environment rather than merely reading a written translation aloud at the end.1
Oral-first workflows can better match communities where Scripture is heard, discussed, memorized, and shared through speech. AI can support transcription, segmentation, search, audio cleanup, comparison, and later text production.
The missionary mistake would be to assume that “digital resource” means “written resource on a screen.”
In some contexts, the most valuable AI investment may be speech recognition or searchable audio rather than another text generator.
Synthetic speech can make enormous libraries available in audio quickly. This is useful for commuters, low-literacy users, people with visual impairments, and oral learners.
Voice choice matters. Pronunciation of names matters. Theological terms matter. In smaller languages, synthetic voices may sound foreign or unnatural.
The Church should therefore measure accessibility through actual users, not through feature availability.
Automatic captioning can improve access to sermons, lectures, and videos. It can also introduce embarrassing or theologically significant errors, especially around names, biblical terms, accents, and low-resource languages.
Public captions should be reviewed proportionally to consequence. A short informal livestream may tolerate minor mistakes. A training library used for years deserves stronger correction.
Sign Languages and Identity
Wycliffe and DOOR International have publicly described several AI-assisted sign-language technologies: Chameleon for camera-based capture and avatar workflows, Lava under development, and Avodah Connect for sign search and replacement tasks.2
These are documented deployments or development efforts. They are not proof that AI has solved sign-language translation or that communities will accept every avatar-based approach.
Sign languages are languages, not visual encodings of nearby spoken languages. Quality therefore depends on linguistic competence, signer identity, natural expression, regional variation, and Deaf-led evaluation.
A technically accurate avatar may still be experienced as unnatural. A system can preserve semantic content while losing features of signing style, embodiment, or community identity.
Synthetic representation can also create a real benefit in sensitive environments. An avatar may allow signed material to be distributed without showing the identity of the original signer.
That can protect people. It can also change the relational quality of the message. The security benefit should therefore be weighed with community input rather than treated as an automatic improvement.
Synthetic sign-language avatars raise questions beyond accuracy. Signers communicate identity through body, face, timing, and style. An avatar can protect a real signer’s identity in a sensitive context but may also reduce the perceived humanity or trustworthiness of the message.
Different communities may judge the trade-off differently. The correct architecture is therefore not universal. Security, naturalness, cost, and community acceptance should all enter the decision.
Sign-language AI must be evaluated as language technology under Deaf-led linguistic judgment, not as visual encoding of nearby spoken languages. 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.
Accessibility Is a Workflow Property
AI can rewrite text into simpler language. This can support readers with different literacy levels, language learners, or cognitive access needs.
Simplification can also remove theological precision, tone, ambiguity, or important rhetorical structure. A simplified Bible study is not automatically an accessible Bible study. Someone competent in both content and audience needs to evaluate what was lost.
Accessibility systems can process highly sensitive information. Speech-to-text may send audio to a cloud provider. Meeting transcription may capture pastoral disclosures. Assistive features should not be exempted from data classification because their purpose is good.
The best architecture may differ by user. Local processing can improve privacy in some contexts. Cloud systems may provide much stronger recognition for another user. Accessibility and security need to be evaluated together.
Christian accessibility is not primarily a branding commitment to inclusive technology. It is rooted in the conviction that people should be able to participate meaningfully in the life, teaching, worship, and mission of the Church.
Technology can remove barriers that were never spiritually necessary. That is exactly the kind of efficiency Christians should celebrate. The standard, however, is not “we produced audio,” “we added captions,” or “we generated an avatar.”
The standard is whether people can actually understand, participate, respond, and exercise agency.
Accessibility is demonstrated by users, not inferred from output format. A resource can technically contain captions and still be inaccessible. Captions may be inaccurate, badly timed, or written at a reading level inappropriate for the audience. An audio version may exist but use pronunciation that makes key terms difficult to understand. An avatar may sign accurately enough for one region and poorly for another.
Accessibility therefore belongs to the whole production workflow rather than a final conversion step.
AI can reduce the cost of producing variants, which should allow ministries to test more options with actual users. The economic gain should be spent partly on evaluation.
Access is not merely consumption. Christian resources should enable participation in teaching, worship, leadership, and mission.
A person using text-to-speech should be able to navigate Scripture references efficiently. A Deaf student should be able to discuss theological concepts in a language capable of carrying them. A learner with cognitive disabilities should not receive only simplified devotional material while deeper teaching remains inaccessible.
AI can widen the forms in which theological content is expressed. The measure is participation, not the number of generated formats.
Organizations accumulate accessibility debt when they produce large libraries in inaccessible formats and promise to fix them later.
Generative AI can help remediate existing material, but the better strategy is accessible-by-design production. Structured headings, alt text, clean transcripts, semantic documents, and multiple modalities make later adaptation easier.
AI should reduce accessibility debt rather than multiply content faster than it can be made usable.
AI should help remediate inaccessible archives without encouraging production to grow faster than accessibility review. 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.
Co-Design and Participation
Recent reviews of generative AI and disability identify promising uses in personalized support and augmentative communication while also emphasizing uneven evidence and the underrepresentation of some disability groups.
Another body of research warns that technology can reproduce ableist assumptions when disability is framed only as a deficit to be repaired.
Christian accessibility should begin from participation and dignity. Disabled Christians are not merely recipients for whom nondisabled technologists make accommodations. They should help decide what useful technology looks like.
Co-design is therefore more than a user-experience technique. It is an agency principle.
“Ask users for feedback” is weaker than co-design. Feedback occurs after major decisions are made. Co-design gives affected people meaningful influence over what problem is being solved and which trade-offs are acceptable.
A Deaf-led team may prioritize natural signing and community identity differently from a hearing engineering team. People with visual impairments may prefer a simple semantic document over a visually rich interface retrofitted with labels.
AI lowers prototyping cost, making iterative co-design easier. Ministries should take advantage of that.