AI Glossary / AI and machine learning / Advanced
Dynamic epistemic logic
Definition
Dynamic epistemic logic studies how knowledge and beliefs change when information is communicated or events are observed. It provides formal models for reasoning about what different agents know before and after an information-changing action.
What it means
An organization can have the right information somewhere without everyone who needs it knowing it. Even when two people know a fact, each may be unsure whether the other knows it. Those differences can affect coordination.
Epistemic refers to knowledge. Dynamic refers to change. The field connects a description of agents' information with descriptions of events that alter that information, such as public announcements or private messages.
A simple public-announcement model assumes the announcement is truthful, trusted and publicly received. More elaborate models can distinguish who hears a message, who notices that someone else heard it, and whether a communication changes beliefs rather than knowledge.
The framework is not a claim that software understands information like a person. It is a formal way to examine the consequences of stated assumptions about information and communication.
How it works
The model starts with possible situations and relationships describing which situations each agent can distinguish. If an agent cannot tell whether a delivery has arrived, both arrived and not-arrived situations may remain possible for that agent.
An information event changes this structure. In a basic truthful public announcement, situations incompatible with the announced fact are removed. The resulting model can be checked for what each agent now knows.
Private or uncertain communications require richer event descriptions. Two agents may experience the same external event differently because one receives a message and the other does not.
The conclusions depend on the assumptions. If a real message can be delayed, misunderstood or ignored, a model that treats it as instantly public and trusted may give an unrealistically strong coordination guarantee.
Three practical examples
- Workflow design: analyze whether a handover makes both teams aware that a prerequisite has been completed.
- Multi-agent research: represent how agents' beliefs change after observations and communications during a shared task.
- Protocol analysis: examine the difference between sending a message, receiving it and knowing that receipt has been acknowledged.
Think of it this way
Think of a meeting where an organizer announces a room change to everyone present. Attendees learn the new room and can also recognize that the other attendees heard the announcement.
The analogy has a limit. Someone may have stepped outside or misheard. A logical model must either assume those possibilities away or represent them explicitly. The mathematics does not guarantee the quality of real communication.
Worked example
Comparing a public notice with a private message
Suppose a fictional two-person team must begin a task only after a machine is ready. Initially, neither Alex nor Sam knows whether readiness has been confirmed.
In one model, a trusted announcement of readiness reaches both, and both know that it reaches both. The update can support stronger shared knowledge than simply giving the information to one person.
In another model, only Alex receives a private readiness message. Alex now has information that Sam lacks. Sam's uncertainty has not disappeared merely because the organization has sent a message somewhere.
The exercise does not claim that an email creates common knowledge. It shows why assumptions about visibility and acknowledgment matter when translating a communication event into a coordination rule.
Illustrative case study
A fictional logistics team designs an automated handover between planning and dispatch. Its process diagram treats posting an update as equivalent to both teams knowing the update.
A review separates posting, delivery, acknowledgment and permission to proceed. The designers model which states remain possible for each participant after each event, then add a clear acknowledgment requirement for the critical handover.
The invented lesson is conceptual: communication has an information effect that should be specified. The organization need not deploy a specialist logic engine to benefit from distinguishing those events.
Common misconceptions and mistakes
- A message sent is knowledge shared. Receipt, trust and awareness of others' receipt may differ.
- Dynamic epistemic logic is ordinary database updating. It studies information and beliefs across agents, including what they know about one another.
- A formal proof makes the assumptions true. A proof establishes consequences within the model; the model must still match the process.
Business use
The field is most directly relevant to researchers and specialists working on multi-agent reasoning, communication protocols and formal verification. Business readers can use its central distinctions when examining handovers and approval workflows.
For a critical process, ask who observes each event, what they can infer and what acknowledgment is required before action. Keep communication failures visible rather than hiding them behind a single completed status.
People also ask
Is this about how a language model learns during training?
Usually no. It concerns formal changes in represented knowledge or belief after events.
Does every announcement have to be truthful?
Basic public-announcement logic uses strong assumptions; broader approaches can model other communication situations.
Is common knowledge the same as everyone knowing a fact?
No. It also involves higher-order knowledge about others knowing, recursively, under the formal definition.
Check your understanding
An automated system sends a private completion message to one of two teams. Why should the workflow not assume both teams know completion occurred?
Answer: The second team may have received no information. The communication model must represent who received the message and any acknowledgment, rather than infer shared knowledge from sending alone.
Related terms
- Epistemic modal logic
- Multiagent systems
- Belief revision
- common knowledge
- formal verification
Learn next
Read Epistemic modal logic. Draw one handover as separate events for sending, receiving, acknowledging and acting, then identify the assumptions between them.
More in AI and machine learning
- Dominance-based rough set approach - The dominance-based rough set approach, or DRSA, analyses decisions involving ordered criteria and ordered outcome classes.
- Double delta feature - A double delta feature is the change of a change: it measures how quickly a measurement is speeding up or slowing down, rather than how fast it is simply...
- DPLL algorithm - The DPLL algorithm is a complete search procedure for deciding whether a propositional-logic formula in conjunctive normal form can be satisfied.
- Eager learning - Eager learning builds a general predictive model from training examples before new prediction requests arrive.
- Early stopping - Early stopping is ending a model's training at the point where it performs best on fresh data, before it starts memorizing the training set.
- Einsum notation - Einsum notation is a compact way of writing operations on multi-dimensional arrays by naming their axes, so sums, products, transposes and contractions are...
Sources
Stanford Encyclopedia of Philosophy, Dynamic Epistemic Logic: https://plato.stanford.edu/entries/dynamic-epistemic/. Checked 12 October 2026. The workflow examples and explanatory analogies are original.
Previous: Dynamic Conditional Correlation Next: Dynamic factor