Lying as Protocol Manipulation: Predictive Modeling Without Theory of Mind
Abstract
This paper develops a naturalistic account of lying as a specialized form of deception grounded in predictive modeling and protocol manipulation rather than in a dedicated Theory of Mind faculty. We distinguish lying from deception, argue that lying requires participation in a norm-governed system of information exchange, and clarify the sense in which intentionality can be understood without explicit belief representation. Drawing on developmental patterns, minimal biological modeling, and computational reasoning, we propose that modeling other agents may be continuous with modeling physical environments. On this view, lying emerges not from a discrete cognitive module for mental-state attribution, but from sufficiently rich predictive modeling of a communicative protocol. Recursive sophistication and belief-language may enhance this capacity, but they are not conceptually required for its basic form.
1. Introduction
Does lying require representing another person’s beliefs? A dominant assumption in philosophy and cognitive science holds that lying depends on attributing mental states and intentionally manipulating them. On this view, to lie is to represent that another agent believes P, to know that P is false, and to assert P in order to alter that belief.
This paper explores an alternative hypothesis. Lying may not require a distinct faculty for modeling minds. Instead, it may require only a sufficiently sophisticated predictive model of a communicative protocol. The “mind” of another agent need not be represented as a separate ontological object; it may function as part of a learned causal structure governing interaction.
Under this account, lying is the strategic manipulation of a norm-stabilized system of information exchange for the purpose of reward optimization. Belief attribution, where present, may function as a compressed abstraction over predictive regularities rather than as a necessary architectural component.
2. Deception and Lying: A Structural Distinction
A crucial distinction must be drawn between deception and lying.
Deception can be defined broadly as causing another agent to form or maintain a false representation, regardless of the method used. Deception can occur through camouflage, distraction, feigned behavior, silence, or environmental manipulation. It does not require language, assertion, or participation in shared truth norms.
Lying, in contrast, is narrower and structurally constrained. Lying involves communicating false content within a shared representational system governed by expectations of truthful use. It presupposes:
The norm in question need not be moralized or explicitly codified. It may consist simply in a stable statistical regularity: signals are typically used to convey accurate information. Lying exploits that regularity. It is therefore best understood as a specialized form of deception that operates through manipulation of a norm-governed communicative protocol.
3. Minimal World Modeling: The Ant and the Shampoo Barrier
To clarify what is meant by “modeling” without invoking symbolic reasoning or belief attribution, consider a minimal biological case.
An ant is placed inside a circular boundary formed by a thin ring of shampoo. The shampoo functions as a chemical barrier. Upon first contact, the ant does not immediately cross. Instead, it samples the substance with its sensory apparatus and registers it as aversive.
The ant then begins walking along the inner circumference of the ring. It traces the boundary, moving laterally as if searching for a discontinuity. Only after exploring the perimeter and failing to discover an opening does it alter its strategy and cross the barrier.
This behavior does not require symbolic reasoning or explicit representation of “barriers” as concepts. What it demonstrates is something more basic: persistence of internal state across time, sensitivity to environmental regularities, and strategy modification based on accumulated sensory input. The ant samples, encodes, explores alternatives, and updates behavior.
The example illustrates a minimal form of modeling: constructing an internal structure from sensory interaction that guides future action. Social modeling may be continuous with this environmental modeling, differing in complexity rather than in kind.
4. Communicative Protocols as Learned Structures
A communicative protocol can be defined as a structured mapping between:
In ordinary interaction, honest signaling stabilizes coordination. If signal S in context C reliably predicts outcome O, agents can align behavior efficiently.
Lying occurs when an agent learns this structure and deliberately exploits it. The agent need not represent “belief” as an abstract object. It need only learn a predictive mapping of the form:
Signal S in context C → Predicted response R → Expected reward trajectory.
If emitting a false signal S' yields a more advantageous predicted trajectory, the agent may select it. The communicative norm, typically truthful signaling, functions as a stable background regularity that makes such exploitation possible.
5. Intentionality Without Belief Representation
Lying is ordinarily described as an intentional violation of truth norms. On the present account, “intentional” does not require reflective awareness of rule-breaking or explicit representation of another’s belief state.
Intentionality can instead be understood as counterfactual protocol competence:
Norm violation, in this sense, is structural rather than moral. The agent need not represent the norm as a rule; it need only recognize that the system’s typical operation can be predictably redirected by emitting a nonstandard signal.
Thus, intentional lying reduces to strategic deviation within a learned predictive system.
6. Two Architectures of Deceptive Behavior
Architecture 1: Mentalistic Model
On a traditional account, the agent explicitly represents:
This supports explicit recursive reasoning and counter-deception.
Architecture 2: Protocol-Based Model
On the protocol-based account, the agent learns:
The internal states of other agents function as predictive variables within this learned structure. They need not be represented as propositional attitudes. What is modeled is the dynamic execution of the protocol itself.
Both architectures may converge behaviorally. The claim defended here is not that mentalistic modeling never occurs, but that it is not conceptually required for basic lying.
7. Simulation and Recursive Depth
A common objection holds that higher-order deception requires recursive belief attribution (“I think that she thinks…”). However, recursion may be understood more minimally as iterative simulation of protocol execution.
An agent may internally simulate:
If I emit S, the other will likely respond with R. If R occurs, I can follow with T, leading to outcome U.
Such nested simulations do not require representing beliefs as beliefs. They require only the capacity to project multi-step trajectories within a structured interaction system. Increased recursive depth may reflect scaling in predictive capacity rather than the activation of a qualitatively distinct cognitive module.
8. Developmental Emergence of Lying
Children begin producing simple denials around ages two to three, but stable, strategically maintained lying typically emerges later. This developmental delay may reflect the gradual enrichment of social predictive models rather than the sudden activation of a Theory of Mind module.
Children must learn:
Through repeated interaction, these regularities are compressed into increasingly efficient predictive structures. Lying emerges when the child can simulate deviations within this structure and anticipate their downstream effects.
9. Belief as Predictive Compression
Belief states may function as compressed predictive summaries of behavioral regularities. To say “the other believes X” may simply encode the expectation that the agent will act in ways consistent with X being true.
Such compression is explanatorily powerful and cognitively efficient. However, it need not correspond to a distinct architectural module. The predictive machinery may operate at a lower level, with belief-language serving as an abstraction over that machinery.
On this view, modeling a communicative protocol does not require positing a separate ontological category of “mind.” It requires tracking structured regularities in interactive systems.
10. Lying as Specialized Protocol Manipulation
Lying can therefore be understood as:
The strategic exploitation of a norm-stabilized communicative protocol by transmitting false content in order to redirect predicted behavioral trajectories.
It differs from general deception because it operates within a shared representational system governed by expectations of truthful use. Its defining features are:
Explicit belief representation and recursive mental-state attribution may enhance this capacity, particularly in complex human societies. However, they are not conceptually necessary for its basic form.
11. Conclusion
Modeling begins at minimal levels, as illustrated by simple organisms adapting behavior through sensory interaction. Social modeling may be continuous with environmental modeling, differing primarily in dimensionality and structural richness.
Lying emerges when an agent acquires a sufficiently integrated predictive model of a communicative protocol and can strategically deviate from its stable regularities. What is required is not a discrete Theory of Mind module, but the capacity to simulate and manipulate structured interaction systems.
Whether higher-order mentalizing constitutes a qualitative discontinuity or merely increased predictive depth remains open. Nonetheless, lying can be coherently framed as specialized deception achieved through protocol manipulation, with explicit belief modeling functioning as a powerful but optional abstraction layered atop more fundamental predictive processes.
