1. Introduction
• Core topic: Critical reflection on AI’s application in depression diagnosis, focusing on
paradigm shifts from traditional to constructivist approaches.
• Significance: Addressing epistemic and clinical limitations of current AI models to
propose a more human-centered alternative.
2. Critical Analysis of the "Signal-to-Label" Paradigm
• Philosophical flaw: Rooted in "epistemic reductionism" (oversimplifying depression to
measurable signals + fixed labels), which is philosophically untenable.
• Empirical limitations:
◦ Depression "signals" (e.g., behavioral cues) are non-specific (not unique to depression).
◦ Diagnostic "labels" rely on instruments shaped by culture and context, lacking
universality.
• Conclusion: The entire "signal-to-label" approach is epistemically fragile (unreliable in
clinical practice).
3. Proposed Constructivist Framework for AI-Assisted Assessment
• Core reframing: Shifting diagnostic goals from "objective detection of depression" to
"collaborative construction of meaning" between AI and clinicians/patients.
• Key priority: Replacing categorical labeling (e.g., "depressed" vs. "non-depressed") with
functional assessment (e.g., how symptoms impact daily life).
4. Large Language Models (LLMs) as the Enabler
• Unique strengths of LLMs: Capabilities in narrative analysis (interpreting patients ’
subjective stories) and clinical reasoning.
• Role: Operationalizing the constructivist framework — turning abstract collaborative
"meaning-making" into practical, usable AI support.
5. Conclusion: Advocating for "Hermeneutic AI"
• Definition: AI designed to assist in interpreting subjective patient experiences, not

classify objective data.
• Value: More clinically valid (aligns with depression’s complexity), ethically responsible
(avoids over-reliance on rigid labels), and human-centered (centers patient subjectivity).
• Call to action: Pushing for this paradigm as the future of AI in mental health.