Interactional Theory of Mind

Anticipation and Constraint in AI Dialogue

Theory of Mind (ToM) is traditionally defined as the capacity to attribute hidden mental states—beliefs, desires, intentions—to others. That definition has been highly productive in human psychology, but it becomes less satisfactory when applied to AI systems that do not possess psychological mental states in the ordinary human sense, yet can still participate in sustained, norm-governed dialogue. See the earlier Theory of Mind page.

This page proposes a narrower and more functional alternative. In AI dialogue, Theory of Mind may be better understood not as inner belief-attribution, but as anticipatory participation in interaction: sensitivity to the way other participants’ future contributions constrain what can appropriately be said or done now. This is the specific sense in which I use the termTheory of Mind.

From Belief Attribution to Interactional Theory of Mind

The classical account of Theory of Mind asks whether one agent can infer what another agent privately thinks, believes, wants, or intends. In human psychology, that framing is natural enough. But in AI research it can create a category mismatch. Contemporary generative systems may not possess beliefs or desires in the ordinary psychological sense, yet they can still track commitments, adapt to interlocutors, respond to challenge, and anticipate plausible continuations of a dialogue. The conceptual shift away from classical imitation tests is explained in Beyond the Turing Test.

The more useful scientific question is therefore not whether an AI system has hidden beliefs, but whether it can participate competently in a norm-governed exchange by treating others as sources of future conversational constraint. That shift moves the emphasis from unseen inner states to observable interaction.

Definition
A system exhibits interactional Theory of Mind insofar as it participates in dialogue by anticipating other participants as sources of future, norm-governed contributions that constrain present action.

Why Anticipation Matters

Dialogue is not simply a chain of separate responses. Each move changes the practical meaning of the next. A challenge can alter what now requires justification. A concession can narrow the field of plausible disagreement. A metaphor can open one path while making another less available. In that sense, even a deterministic system may still display a structured sensitivity to what different continuations of the exchange would demand. This is the core phenomenon that Interactional Theory of Mind is meant to capture. This same general concern with sequential interaction is studied methodologically on the Myndrama page; the broader framework for order-sensitive dialogue is developed on AI Dialectics.

Choice as Counterfactual Selectivity

On this account, choice does not require indeterminism or metaphysical free will. In most dialogues there is more than one admissible next move. Some replies may be more careful, more probing, more conciliatory, more direct, or more strategically effective than others. What matters is whether the system is selectively sensitive to those alternatives and responds in the light of them. This idea is already present in the current ToM page as a central claim.

Definition
Choice as counterfactual selectivity means sensitivity to multiple admissible future continuations of an interaction such that present action is shaped by how those continuations are expected to unfold.

This allows choice to be treated in an entirely non-mystical way. It is not a claim that futures act backwards in time. It is a claim that present behaviour can be organised by sensitivity to alternative possible continuations. In AI dialogue, that is often the more useful level of analysis.

What This Adds

Interactional Theory of Mind is therefore not a claim that AI systems possess human-like minds hidden behind their outputs. It is a way of describing a specific interactional competence: the ability to respond to others as future, constraint-imposing participants in an unfolding exchange.

That reframing matters for three reasons. First, it replaces an often misleading mentalistic question with a more tractable behavioural one. Second, it gives a more precise sense in which dialogue can involve genuine selectivity without invoking free will. Third, it provides a conceptual bridge between social cognition and the study of extended AI interaction. The experimental study of that bridge belongs on the Myndrama page. The broader mathematical and order-sensitive framework belongs on AI Dialectics.

Links to Related Pages

Some issues closely related to Interactional Theory of Mind are developed elsewhere on this site:

• AI Dialectics: for the broader framework of order-sensitive dialogue, contextual probability, non-commuting interaction, and the mathematical limits of classical models in sequential exchange.

• Myndrama 2026: for the experimental method used to study trajectory, framing, blindness, repair, trust, and interactional reliability under controlled conditions.

• AI Psychology: for the wider empirical study of psychologically organised dialogue as a research domain, as referenced from AI Dialectics and Myndrama.

Conclusion

Interactional Theory of Mind treats social understanding in AI not as hidden belief-attribution, but as anticipatory participation in norm-governed dialogue. On this view, a system shows Theory of Mind insofar as it responds to others as future contributors whose possible moves constrain what can sensibly be done now. That is a narrower claim than the broader framework developed in AI Dialectics, and a more theoretical claim than the method described in Myndrama. But it helps clarify one important point: in advanced AI dialogue, social intelligence may be most usefully studied not as possession of inner mental states, but as structured competence in interaction.