Belief-Desire-Knowledge Map¶
Mental-state model — instantiates Other-Agent State Model Calibration
Lays out what another agent probably believes, wants, knows, lacks, fears, and expects as an explicit set of hypotheses, each carrying a confidence level.
Belief-Desire-Knowledge Map is the standing analytical artifact of the archetype: it decomposes another agent's hidden state into distinct columns — what they believe, what they want, what they know (and don't), what they fear, what they expect — and records each entry as a labelled hypothesis with a confidence attached. Its defining discipline is separation: it refuses to let "they believe X," "they want Y," and "they don't know Z" collapse into a single undifferentiated impression of the other person, because those three call for completely different responses. Unlike the live reflection of an active-listening exchange or the experiential framing of an empathy map, this is a structured mind-model you can hold up, argue about, and revise — a decomposition, not a conversation and not a mood board.
Example¶
A product manager is about to pitch a risky platform migration to a skeptical VP of Engineering, and drafts a Belief-Desire-Knowledge Map of her before the meeting. Believes: that the last migration ran a year over (confidence: high — she said so publicly). Wants: to protect her team's roadmap from another disruption (high) and, less certainly, to be seen backing something ambitious (low). Knows: the technical case in detail. Doesn't know: that the new approach is incremental and reversible (medium — this may be the crux). Fears: being blindsided in front of her own reports (medium).
Laid out this way, the pitch rewrites itself. The map says the binding issue is not whether the VP believes in the platform but a specific knowledge gap — she thinks this is another big-bang migration — held at only medium confidence. So the PM front-loads the reversibility story and, because that hypothesis is uncertain, opens by asking rather than asserting. The map didn't tell her what the VP thinks; it told her which hypothesis, at which confidence, the whole meeting turns on.
How it works¶
- Name the boundary first. The map is headed with whose state it models and for what decision — a specific counterparty in a specific context, not a general psychological profile.
- Separate the columns. Belief, desire, knowledge, fear, and expectation are filled independently; the point is to stop one from masquerading as another.
- Write each cell as a hypothesis. Not "she is risk-averse" but "she wants to protect the roadmap" — a claim that could be wrong and could be checked.
- Tag confidence. Every cell carries high / medium / low, so the reader can see instantly which parts of the model are load-bearing guesses.
Tuning parameters¶
- Category set — the classic belief/desire/knowledge triad, or an extended set adding fear, incentive, and expectation. More columns capture more but dilute focus.
- Confidence scale — a coarse high/medium/low or a finer graded band. Finer scales invite false precision on what are, after all, guesses about a mind.
- Evidence strictness — whether each cell must cite the signal it rests on. Strict sourcing curbs projection but slows the map down.
- Refresh cadence — a one-off pre-meeting sketch or a living document revised after each interaction. The living version is more accurate but only if someone actually maintains it.
When it helps, and when it misleads¶
Its strength is that separation prevents the most common modelling error — reading a single behaviour (a missed deadline, a terse reply) as one global fact about the person — by forcing the observer to ask whether it reflects a belief, a competing desire, a knowledge gap, or a constraint. Confidence tags then keep the team honest about how much of the plan rests on speculation.
Its failure mode is that a written map hardens into fact: the hypotheses lose their question marks and the confidence tags get quietly ignored, so a convenient guess becomes an assumed truth. It is especially prone to the fundamental attribution error — filling the "desire" and "belief" columns with dispositional stories ("she just doesn't want change") while ignoring the situation the other agent is actually in[1]. And a tidy map can be run backwards, assembled to justify a read of someone the actor already holds. The discipline that guards against this is to date the map, keep every cell flagged as a revisable hypothesis, and lower confidence the moment the other agent does something the map didn't predict.
How it implements the components¶
modeled_agent_boundary— the map's header fixes whose state is being modelled and for which decision, keeping it a bounded working model rather than open-ended profiling.hidden_state_hypothesis_set— the belief / desire / knowledge / fear / expectation columns are the set of live hypotheses, held side by side so no single one prematurely wins.uncertainty_and_confidence_annotation— each cell's confidence tag marks how much weight it can bear and flags where alternative explanations remain open.
It does not elicit the evidence itself (that comes from the Active Listening Loop and the Perspective-Taking Interview), adversarially test the model for wishful thinking (that's the Counterparty Model Red Team), or govern whether the model is legitimate to hold (Consent and Privacy Boundary Checklist) — it structures and grades the hypotheses.
Related¶
- Instantiates: Other-Agent State Model Calibration — the map is the archetype's core hypothesis artifact that other mechanisms feed, test, and act on.
- Consumes: the evidence gathered by elicitation mechanisms such as the Active Listening Loop and Perspective-Taking Interview.
- Sibling mechanisms: Empathy Map with Evidence Marks · Counterparty Model Red Team · Active Listening Loop · Consent and Privacy Boundary Checklist · False-Belief Check · Perspective-Taking Interview · Prediction and Surprise Log · Role-Reversal Simulation · Interaction After-Action Review · Stakeholder Hidden-Constraint Board
Editorial Notes¶
Form Classification¶
Form family: Representation, Specification & Plan
Rationale: The mechanism externalizes hypotheses about a specific actor's beliefs, desires, knowledge, fears, expectations, and confidence in a structured map, so its operative form is an information representation.
Nearest alternative: Analysis, Modeling & Optimization — Analytic judgment populates the cells, but the practitioner deploys and revises the map rather than a computational estimator.
Review outcome: Adjudicated after independent review; high confidence.
Origin Attribution¶
Primary origin: Cognitive Science
Origin pattern: Convergent development
Present-day reach: Multi-domain
Rationale: Explicit decomposition of another agent into belief, desire, knowledge, and expectation hypotheses belongs to cognitive science's information-processing models of mind.
Related originating lineages:
- Computer Science & Software Engineering — BDI-style agent architectures made such state decompositions operational in artificial-agent design.
- Philosophy — Philosophy of mind and action supplied the propositional-attitude categories of belief, desire, and intention.
- Psychology — Social-cognitive and theory-of-mind research supplies empirical methods for inferring others' hidden states.
Review resolution: Cognitive science is the agreed primary lineage through mental-state representation and theory of mind. Computational BDI agents, philosophical intentional-state vocabularies, and psychology developed materially distinct inputs that converge in the Encyclopedia artifact.
Attribution caveat: The artifact deliberately combines psychological mentalizing, philosophical intentional states, and computational BDI-agent formalisms.
Encyclopedia synthesis: The exact catalogued form synthesizes established practice rather than reproducing a single standard historical label.
Review outcome: Reconciled after independent review; high confidence.
References¶
[1] Ross, L. "The Intuitive Psychologist and His Shortcomings: Distortions in the Attribution Process". Advances in Experimental Social Psychology 10, 173–220 (1977). Describes observers' tendency to overattribute behavior to dispositions while underweighting situational constraints. registry ↩