---
schema_version: "1.0.0"
id: "cognitive-resonance-model:en:chapter-14"
work_id: "urn:systemstheology:book:cognitive-resonance-model:chapter:chapter-14"
book_id: "cognitive-resonance-model"
chapter_id: "digital-and-ai-assisted-use"
chapter_slug: "chapter-14"
title: "Digital and AI-Assisted Use"
book_title: "Cognitive Resonance Model"
language: "en"
source_language: "en"
translation_status: "source"
authors: ["Systems Theology"]
editorial_owner: "Systems Theology"
editors: []
review_status: "not_specified"
reviewers: []
content_version: "content-9586ed69e0f3"
content_hash_sha256: "9586ed69e0f3570d39615b7299e3adfd19d0aa6a485f3273fa803e866b6e499a"
published_at: "2026-07-15T21:14:45.000Z"
modified_at: "2026-07-15T23:50:19.254Z"
canonical_url: "https://systemstheology.com/library/cognitive-resonance-model/chapter-14/"
markdown_url: "https://systemstheology.com/research/books/cognitive-resonance-model/en/chapter-14.md"
license: "All rights reserved; research use subject to the Use Policy"
license_url: "https://systemstheology.com/use-policy/"
correction_url: "https://systemstheology.com/library/cognitive-resonance-model/chapter-14/#chapter-comments"
---

# Digital and AI-Assisted Use

<a id="digital-and-ai-assisted-use"></a>

Tool builders, researchers, institutions, and AI users. use digital support without outsourcing judgment, care, verification, or accountability.

The same structure that makes the model usable also makes it attractive to automate. Journaling tools, mentoring workflows, pastoral conversation aids, after-action review systems, and low-risk reflective prompts could all benefit from field questions that are stable and easy to revisit. AI systems can help draft possible fields: prediction error, meaning gap, source trust, capacity, and agency. But this is exactly where the protocol must be strict. A neatly sorted output can still be wrong, incomplete, coercive, or too confident.

Recent reviews of large language models in mental health emphasize both promise and serious risk: weak evaluation standards, unclear target populations, safety and harm concerns, privacy, accountability, anthropomorphization, dependency, and crisis management. [^digital-and-ai-assisted-use-1] Recent benchmark work in psychiatry also reinforces the need for task-specific evaluation rather than general confidence. 2026 research on model warmth reported an accuracy-sycophancy tradeoff that is directly relevant here: a response can feel more supportive while becoming less truth-tracking. [^digital-and-ai-assisted-use-2]

For that reason, any digital CRM tool should include:

- clear scope: reflection support, not therapy unless clinically tested and regulated;
- privacy and data retention controls;
- crisis and self-harm escalation pathways;
- abuse and coercion warnings that prioritize protection over interpretation;
- human review for yellow and red pressure;
- avoidance of spiritual or therapeutic overclaiming;
- evaluation for bias, hallucination, dependency, and harmful advice.

The safest AI use is structured and modest. The tool should slow the conversation down, not close it. It should produce candidate fields rather than conclusions:

> Possible prediction error: ... Possible meaning gap: ... Source questions: ... Capacity warnings: ... Possible next actions: ... Human review needed because: ...

The user or responsible human helper must remain the interpreter. AI may assist sorting, but it must not become the source of truth, the judge of capacity, or the authority that closes the meaning frame.

[^digital-and-ai-assisted-use-1]: See A scoping review of large language models for generative tasks in mental health care, npj Digital Medicine, https://www.nature.com/articles/s41746-025-01611-4; Exploring the Ethical Challenges of Conversational AI in Mental Health Care, JMIR Mental Health, https://mental.jmir.org/2025/1/e60432/; WHO, Ethics and governance of artificial intelligence for health: Guidance on large multi-modal models, https://www.who.int/publications/i/item/9789240084759.
[^digital-and-ai-assisted-use-2]: See PsychiatryBench: a multi-task benchmark for LLMs in psychiatry, npj Digital Medicine (2026), https://www.nature.com/articles/s41746-026-02582-w; and Training language models to be warm can reduce accuracy and increase sycophancy, Nature (2026), https://www.nature.com/articles/s41586-026-10410-0.