💻 computer science

Beyond the Spell: A Dynamic Logic Analysis of Misdirection

This paper introduces a dynamic epistemic logic framework that formally models both verbal and visual misdirection, illustrated through the French Drop magic trick, and provides a sound and complete axiom system for the logic.

Original authors: Benjamin Icard, Raul Fervari

Published 2026-07-23

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Benjamin Icard, Raul Fervari

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are a detective trying to solve a mystery, but the suspect isn't just hiding the truth; they are actively painting a fake picture of reality right in front of your eyes. This is the world of misdirection, a concept that sits at the crossroads of psychology, magic, and computer science. To understand how it works, we need two basic tools. First, there's belief: what you think is true in your head. Second, there's observation: what your eyes actually see. Usually, these two match up perfectly. If you see a cat, you believe there is a cat. But what happens when someone tricks your eyes? What if you see a cat, but it's actually a dog, or worse, what if you see nothing at all, but you are tricked into believing a cat is there?

This is where Dynamic Epistemic Logic comes in. Think of this as a super-precise map for tracking how information changes in people's minds. It's like a video game engine for logic, where every time someone speaks or does a gesture, the "world" of beliefs and observations updates instantly. While scientists have spent years mapping out how people lie with words (verbal misdirection), they haven't really built a map for how people lie with their eyes (visual misdirection). This paper asks a simple but tricky question: Can we use this logical map to explain how a magician tricks your brain into seeing things that aren't there, and why that feels so different from just being told a lie?

The Magic of Logic: Decoding the French Drop

In this paper, authors Benjamin Icard and Raul Fervari build a new logical tool called DLM (Dynamic Logic of Misdirection). Their goal is to create a formal system that can track not just what people believe, but exactly what they see, and how a magician can mess with both at the same time.

The Two Ways to Trick a Brain

The authors start by sorting misdirection into two main categories, using a classic magic trick called the French Drop as their test case. Imagine a magician holding a coin.

Dissimulation (The Art of Hiding): This is when the magician hides the truth. In the French Drop, the magician secretly keeps the coin in their left hand (palming it) while making it look like they are moving it. They are "dissimulating" the coin's true location. They aren't showing you a fake coin; they are just blocking your view of the real one.
Simulation (The Art of Faking): This is when the magician creates a false reality. The magician makes a gesture that looks exactly like they are dropping the coin into their right hand. They are "simulating" a transfer that never happened.

The paper argues that while we have good logical models for verbal lies (like a politician saying "I am honest" when they aren't), we haven't had a good way to model these visual tricks until now.

The New Logic: Seeing vs. Believing

The core innovation of DLM is how it treats observation. In many old logic systems, if you see something, you automatically believe it. But the authors point out that this isn't always true. You can have atomic observation (seeing a shape without knowing what it is) and epistemic observation (seeing a shape and recognizing it as a specific object).

Think of it like this:

Atomic Observation: You see a blurry red blob in the water. Your eyes register "red," but your brain hasn't decided if it's a fish, a toy, or a stain.
Epistemic Observation: You look again, focus, and your brain says, "That is a fish!"

The French Drop works by manipulating these stages. The magician uses a fake gesture (simulation) to force the audience's atomic observation (seeing a hand move) to instantly become a false epistemic observation (believing the coin moved). At the same time, they use a hidden hand (dissimulation) to stop the audience from seeing the truth.

Modeling the Trick

The authors use their new logic to break down the French Drop step-by-step:

The Setup: The audience sees the coin in the left hand. They believe it's there.
The Move (Simulation): The magician pretends to move the coin to the right hand. The logic shows that this action updates the audience's "world." Suddenly, in their minds, the coin is in the right hand. Their eyes have been tricked into "seeing" the transfer.
The Reveal: The magician opens the right hand (empty) and then the left hand (full). The audience is shocked.

The paper proves that this shock, or surprise, happens because of a "mismatch." The audience's brain had a strong belief (the coin is in the right hand) that was completely contradicted by the new visual evidence (the coin is in the left hand). The logic shows that this surprise is stronger when the audience was sure of the wrong thing, rather than just unsure.

What This Logic Can and Cannot Do

The authors are careful to state what their model does. They have created a sound and complete system, meaning their rules are consistent and cover all the cases they intended to model. They successfully showed that:

Visual simulation (faking an action) logically implies visual dissimulation (hiding the truth). If you fake a coin moving to the right, you are automatically hiding the fact that it stayed in the left.
You can distinguish between a "genuine" action (showing the truth) and a "bogus" action (showing a lie).
The model captures the specific feeling of surprise in magic tricks.

However, the paper also notes what it doesn't do. It doesn't try to explain why humans are naturally bad at spotting these tricks (that's a psychology question). It also doesn't claim to solve every type of deception in the world. For instance, they mention that if a magician is clumsy and fails, the logic might need to be adjusted. They also admit that their current model assumes the magician always succeeds in the trick, which is a simplification.

The Bigger Picture

By turning magic tricks into logical equations, Icard and Fervari aren't just trying to teach magicians how to do better tricks. They are building a foundation for understanding how information flows between people. This could help in the future to model complex situations like military deception, where soldiers might use camouflage (dissimulation) and fake tanks (simulation) to confuse an enemy. It could even help in AI planning, where a computer agent needs to figure out how to mislead another agent to achieve a goal.

In short, this paper takes the "magic" out of a magic trick and replaces it with a clear, step-by-step logical map. It shows that when a magician tricks you, they aren't just playing with your eyes; they are hacking your brain's update system, and now, we have a new language to describe exactly how that hack works.

Technical Summary: Beyond the Spell: A Dynamic Logic Analysis of Misdirection

Problem Statement
Misdirection is defined as the intentional action of causing misrepresentation in an agent or group of agents. While verbal misdirection (e.g., lying, bluffing, half-truths) has been extensively investigated from a formal perspective using epistemic logic, visual misdirection has received little attention. A critical gap exists in the formal treatment of visual misdirection, particularly regarding the distinction between dissimulation (hiding the truth) and simulation (showing false facts), and the resulting inaccurate observations. Existing frameworks often conflate observation with belief or fail to model the dynamic shift from non-epistemic (atomic) observation to epistemic observation induced by deceptive visual actions.

Methodology
The authors introduce DLM (Dynamic Logic of Misdirection), a dynamic epistemic logic designed to model both verbal and visual misdirection. The methodology involves:

Observational Epistemic Models: The semantic foundation extends standard Kripke models with a specific valuation for observation symbols. Unlike previous work on attentive agents where observation implies attention, DLM distinguishes between:
- Atomic Observation ( $o_a p$ ): A non-epistemic, causal perception where an agent visually perceives $p$ without necessarily forming a belief about it (Dretske's "simple seeing").
- Epistemic Observation: A stronger notion defined as atomic observation coupled with the belief that one is observing ( $B_a o_a p$ ).
- Belief ( $B_a \phi$ ): Represented via a standard doxastic modality with Euclidean, transitive, and serial accessibility relations.
Action Models with Postconditions: DLM adapts standard Dynamic Epistemic Logic (DEL) by utilizing action models that include postconditions. This allows the logic to update not only the agents' beliefs but also the truth values of their observation atoms.
- Verbal Actions: Modeled as announcements ( $tell^+_a$ for truthful, $tell^-_a$ for deceptive). These update accessibility relations but leave observation atoms unchanged.
- Visual Actions: Modeled as showing ( $show^+_a$ for genuine, $show^-_a$ for deceptive). These update both accessibility relations and the observation atoms (postconditions), effectively changing what the agent "sees" and subsequently believes they see.
Axiomatization: The paper provides a sound and complete axiom system (OL for the static fragment and DOL for the full dynamic logic). The system includes reduction axioms to eliminate dynamic modalities by transforming them into static formulas, relying on the properties of the action models.
Case Study (The French Drop): The logic is applied to model the "French Drop" magic trick. The trick is decomposed into:
- Initial State: The magician holds a coin in the left hand ( $l$ ); the audience is uncertain.
- Method (Simulation): The magician performs a visual action $show^-_a(r \land \neg l)$ , simulating a transfer to the right hand. This action updates the audience's observations to $o_b r$ and $o_b \neg l$ , converting their atomic observation into a false epistemic observation.
- Dissimulation: Simultaneously, the magician conceals the coin (palming), preventing the audience from observing $l$ .
- Revelation: The magician performs $show^+_a(l \land \neg r)$ , revealing the truth and correcting the audience's observations.

Key Contributions and Results

Formal Distinction of Modalities: The paper formally distinguishes between verbal and visual modalities and between genuine and misleading actions. It explicitly models the transition from atomic (non-epistemic) observation to epistemic observation via visual actions.
Simulation vs. Dissimulation: The authors define Simulation ($Sim$) as an active process of presenting false information (verbal or visual) to induce a wrong representation. Dissimulation ($Dis$) is defined as the withholding of information (hiding the truth) or the failure to perform a revealing action.
- Result: The logic proves that successful simulation of $\neg p$ implies the dissimulation of $p$ (Corollary 30), provided the addressee successfully adopts the false belief/observation.
Stronger Notions of Belief and Observation: The framework defines "strong epistemic observation" and "strong belief" as attitudes supported by genuine verbal or visual actions, distinguishing them from weak beliefs that may lack justification.
Modeling Surprise: The paper demonstrates that DLM can capture "mismatch-based surprise." This occurs when an agent's epistemic observation (e.g., observing the coin is in the right hand) is suddenly contradicted by a new visual action revealing the opposite reality (the coin is in the left hand), creating a logical inconsistency between expectation and observation.
Completeness: The paper establishes that the axiom system DOL is sound and strongly complete with respect to the class of Euclidean and transitive relational models, noting that seriality is not preserved under all dynamic updates (a known characteristic of DEL with postconditions).

Significance and Claims
The authors claim that DLM constitutes a novel application of Dynamic Epistemic Logic to the specific domain of misdirection. The significance of the work lies in:

Filling the Gap: Providing the first formal treatment of visual misdirection and inaccurate observations, contrasting with the extensive literature on verbal deception.
Clarifying Concepts: Using logical tools to clarify the interaction between non-epistemic and epistemic seeing, and between simulation and dissimulation.
Foundational Step: The paper positions itself as a starting point for modeling different forms of misdirection. It does not claim to be a comprehensive theory of all deception but rather a formalization of established conceptual distinctions.
Future Directions: The authors suggest that the framework can be extended to model collective misdirection (accomplices), intentional aspects of deception (using causal logic or STIT), and strategies of misdirection via epistemic planning. They acknowledge limitations, such as the loss of seriality in models after updates, and propose exploring alternative model classes or preservation properties in future work.