When people talk to a chatbot about goals, conflict, or identity, they leave traces: word choice, hedging, what they emphasize, what they avoid. Psychological analysis in this context does not mean diagnosing disorders. It means using language technology and structured reflection to surface recurring patterns that might help someone understand themselves.
EgoQuest, my conversational personality research project, sits in that space. This post explains the idea at a conceptual level. Proprietary probe design, scoring, and calibration are not published here; see /egoquest for redacted research documentation.
Why conversation instead of forms
Questionnaires fix the vocabulary. Respondents choose among pre-written options. That reduces noise for psychometrics but removes context: how someone explains a decision, how tone shifts under pressure, how they describe relationships.
Conversational interfaces invert the tradeoff. Language is richer and messier. Models must handle ambiguity, topic changes, and cultural variation in how people express emotion or assertiveness. The research challenge is extracting stable signals from unstable text without overclaiming.
Layers that typically appear in this kind of system
Most serious pipelines combine several approaches rather than a single black-box label:
- Linguistic features: syntax, sentiment, formality, pronoun use, and domain-specific markers (carefully curated, not raw forum dumps republished without consent)
- Transformer classifiers: fine-tuned encoders that map utterances or short windows to trait hypotheses
- Retrieval grounding (RAG): typology references and exemplar language so model outputs stay tied to documented concepts instead of free association
- Multi-source fusion: reconciling signals from different frameworks when they partially agree or conflict
- Confidence gating: withholding or softening conclusions when evidence is weak
None of these steps is magic. Each introduces bias from training data, label noise, and cultural blind spots. Transparent uncertainty is part of responsible design.
Structured reflection vs passive inference
Passive inference alone risks telling users what they are without room to disagree. A healthier pattern pairs inference with reflective prompts: questions that invite the user to confirm, reject, or nuance a hypothesis.
Psychology research has long distinguished self-report from observer rating; conversational AI blurs the line because the system is both observer and interlocutor. EgoQuest treats dialogue as collaborative sense-making, not a verdict delivered after three messages.
Limits and responsibilities
- Not clinical. Language patterns are not DSM criteria. Do not use conversational typing for triage or treatment decisions.
- Not deterministic. People change by context, mood, and growth. A profile is a snapshot, not destiny.
- Data ethics. Training on public text with self-reported labels inherits label error. Releases should minimize republication of identifiable content.
- Cultural humility. Models trained predominantly on English forum data will miss nuance in Caribbean English, code-switching, and local norms of politeness or directness.
Connection to EgoQuest
EgoQuest applies these principles to multi-framework personality exploration: Jung/MBTI dichotomies, Enneagram, OCEAN, Socionics, and temperament, unified in one chat experience. The EGO metric names the conceptual structure that holds those lenses together without reducing a person to a single score.
The EgoQuest page lists everything I have published about it.