Who Counts as a Caregiver? A Dual-Axis, AI-Enhanced Model of Support in Asian Familistic Contexts

Research context & significance

Research questions & objectives

• Build a dual-axis, AI-enhanced model of caregiving engagement that integrates identity recognition, love-duty negotiation, individual readiness, collective family readiness, and predictive risk into one decision-support framework — so that support reaches caregivers at the right time, in the right way, without eroding dignity or adding to their burden.

Research approach

1. A · Input / Context

   Caregiver narratives and conversational text. Support group messages. Family chat logs (with consent). Validated scales (ZBI, PHQ-9, GAD-7). Multilingual inputs from FDWs — Singlish, Bahasa Indonesia, Tagalog, Burmese. The voices that surveys miss.

2. B · Method / Process

   LLM-based risk detection and readiness inference. Two-tier readiness architecture: Brown's six macro identity stages + Transtheoretical Model micro states. Love-duty language pattern detection. Deterministic safety gate for crisis. Human-in-the-loop dashboard for community nurses and social workers. AI as listening infrastructure — never as authority.

3. C · Evaluation / Validation

   Compare 4 conditions: survey-only (current standard) vs. AI-enhanced continuous monitoring, human-only case management vs. integrated human-in-the-loop model. KPIs incl. readiness-aligned uptake, time-to-uptake after risk detection, concordance with validated clinical scales, false alarm rate, dignity and equity audit, and caregiver-reported experience. Success is measured not only in efficiency but in whether caregivers feel seen, respected, and supported in a way that fits their life and their values.

Expected outcome

A validated dual-axis model and 5-layer decision-support architecture that helps frontline workers reach invisible caregivers, time support to readiness rather than crisis, and measure what matters: dignity, equity, and genuine care benefit. A policy brief for MOH and Healthier SG. A practical tool co-designed with social workers, case managers, and caregivers themselves.

Progress since last update

• Conceptual synthesis explored across caregiving science, health services research, social psychology, migrant labour studies, and human-AI interaction
• Research gap identified and articulated: most models treat risk, identity, readiness, and family context as separate problems. Dual-axis model specified. 2-tier readiness architecture developed (Brown's macro stages + Transtheoretical micro states)
• 5-layer computational architecture proposed with deterministic safety gate and human-in-the-loop dashboard. 7-paper research agenda with dual-domain deliverables mapped

Tangible outputs & evidence

• Paper 1 draft: Conceptual synthesis, dual-axis model, 2-tier readiness architecture, 5-layer system architecture, objective function, and 7-paper agenda

Key result

• The goal is not service uptake for its own sake, but net care benefit of timely, readiness-aligned, dignity-preserving, equitably distributed support.
• A caregiver who declines help is not a failure of uptake – it is a signal to listen differently.

What should we notice?

• The research gap is not another caregiver burden survey or another chatbot. It is joint decision-making across risk, identity, readiness, and family context. AI is not the intervention – it is the listening infrastructure that makes support receivable.

Next steps

1. 1

   Finalise the integrated framework overview and objective function for Paper 1

   15 Oct 2026

2. 2

   Draft the 2-tier readiness architecture and 5-layer system specification for Paper 1

   31 Oct 2026

3. 3

   Complete Paper 1 manuscript for supervisor review

   30 Nov 2026

Success by next forum: Complete draft of Paper 1 + Operationalisation blueprint for Paper 2

Risks & blockers

• Access to multilingual caregiver data – particularly FDW languages and Singlish – may be challenging
• Ethical approval for conversational data collection requires careful design to protect vulnerable populations. AI safety concerns in clinical contexts require robust safeguards from day 1.
• Balancing the Computer Science slant Social Work depth in Paper 1 requires careful calibration.

Support needed

• Seeking supervision on CS rigour and balance to social work depth here. Confirmation that the integrated framework architecture meets milestone expectations

Possible Research Grants (Proposed)

• NCSS Transformation Sustainability Scheme
• IMDA SG Digital Scholarship (Postgraduate)