🧬 biology

Semantic Information Beyond Shannon: Meaning,Coherence, and Agency in Experiential Systems

This paper proposes a foundational framework called Meaning–Experience Coupling (MEC), which defines semantic information as an emergent property of the structural alignment between representational organization and agentive goals, formalized through a coherence functional that satisfies five specific axioms to explain how information becomes meaningful for experiencing and acting systems.

Original authors: Lolugu Vasudeva Prabhath, Nitin Kumar, Jyotiranjan Beuria, Venkatesh H. Chembrolu

Published 2026-08-04

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Lolugu Vasudeva Prabhath, Nitin Kumar, Jyotiranjan Beuria, Venkatesh H. Chembrolu

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Imagine you are standing in a room full of radio static. To a machine that only counts signals, this static is a treasure trove of information. It is chaotic, unpredictable, and full of "bits." But to your ears, it is just noise. Now, imagine a friend whispers your name in that same static. Suddenly, that tiny, quiet sound becomes the most important thing in the universe. It grabs your attention, changes your mood, and tells you exactly what to do next. This is the puzzle that has haunted scientists for decades: How does a signal turn from mere data into meaning?

For a long time, the gold standard for understanding information was a theory called Shannon Information. Think of it like a super-efficient postal service. It cares only about how many letters you can send, how fast they travel, and how likely they are to get lost in the mail. It doesn't care if the letter says "I love you" or "The sky is green." To the postal service, both are just a string of symbols. This works perfectly for sending text messages or streaming videos, but it hits a wall when we try to understand living things. A brain isn't a postal service; it's a busy city where every signal is a traffic light, a warning siren, or a map. If we only count the bits, we miss the whole point of why the brain is buzzing. This paper asks a simple but deep question: What makes a signal matter to the system receiving it?

The authors of this paper, a team from the Indian Institute of Technology Mandi and the IKS Research Centre, propose a new way to look at this. They suggest that meaning isn't something you add to information later, like sprinkling sugar on a cake. Instead, meaning is a special kind of "fit" or "coupling" that happens inside the system itself. They call this Meaning–Experience Coupling (MEC).

To understand their idea, imagine a dancer and a piece of music. The music (the information) has a rhythm and a structure. The dancer (the agent) has a style, a goal, and a way of moving. If the dancer just stands still while the music plays, there is no dance, only sound. If the dancer moves randomly, there is motion, but no dance. But when the dancer's moves perfectly match the rhythm and the dancer's intent to express a story, that is where the magic happens. The paper argues that semantic information (meaning) is exactly this "dance" between what the system experiences and what the system is trying to do.

The paper introduces a formula to measure this "dance," written as SI = g(R, A).

R stands for the Representational Organization. This is how the system sorts the world. It's the list of things the system notices, like a map of the room or a list of sounds.
A stands for Agentive Alignment. This is the system's "stance" or "goal." It's what the system is trying to achieve, like "find the exit," "avoid danger," or "find food."
g is the function that measures how well the map (R) fits the goal (A).

The authors suggest that if your map is perfect but you have no goal, you have no meaning. If you have a strong goal but no map to guide you, you also have no meaning. Meaning only exists when the two are locked together.

To prove this isn't just a philosophical daydream, the authors built a simple computer simulation, which they call a "toy model." They created a tiny digital agent with three senses (a spatial cue, an edge pattern, and an ambient tone) and three possible goals (navigating, finding a landmark, or just listening to the background).

In their simulation, they kept the agent's senses exactly the same but changed its goal.

When the agent's goal was Navigation, the "spatial cue" became super meaningful. The semantic information score jumped to about 0.433.
When the agent's goal switched to Background Monitoring, that same spatial cue became almost useless. The score dropped to about 0.317.
The "meaning" of the signal changed by roughly 27% just because the agent's focus shifted, even though the signal itself didn't change at all.

This simulation shows that meaning isn't a fixed property of a signal. It's a relationship. A signal can be heavy with data but empty of meaning if it doesn't connect to what the system is doing. Conversely, a tiny, faint signal can be packed with meaning if it's exactly what the system needs right now.

The paper also explores what happens when this "dance" breaks. They describe three ways meaning can fail:

Fragmentation: The system sees too many things but can't connect them to any goal. It's like being in a crowded room where everyone is shouting, and you can't hear a single conversation.
Rigidity: The system keeps doing the same thing even when the world has changed. It's like a GPS that keeps telling you to turn left into a wall because it hasn't updated its map.
Misalignment: The signal is clear, but the goal is wrong. It's like hearing a fire alarm while you are trying to sleep; the sound is clear, but your current goal (sleeping) makes it confusing or ignored until you realize the danger.

The authors are careful to say they aren't solving the "hard problem" of consciousness (why we feel anything at all). They aren't saying this formula will tell us if a robot is truly "alive" or "feeling." Instead, they are offering a new tool to describe how a system organizes its world. They argue that previous theories tried to find meaning in the signal itself (like checking if it's true) or in the outside world (like checking if it helps you survive). This paper suggests we should look inside the system to see how its internal map and its internal goals are dancing together.

The team admits this is just a first step. Their formula is a "weighted sum," which is a simple way to add things up, but they show that more complex versions (using something called a "power-mean family") could work too. They suggest that in the future, scientists could use this idea to test artificial intelligence. If a robot has a great internal map but its goals are just random numbers programmed by a human, it might have high "data" but zero "meaning." But if the robot's goals and its map are truly coupled—if it is genuinely trying to solve a problem using its own understanding—then it might be on the path to something we could call semantic understanding.

In short, this paper suggests that meaning isn't a thing you find; it's a thing you do. It's the active, dynamic relationship between what you know and what you are trying to achieve. When those two lines cross, information stops being just noise and starts being a story.

Technical Summary: Semantic Information Beyond Shannon

Problem Statement
The paper addresses a foundational gap in information science: the exclusion of "meaning" from formal information theory. While Shannon's theory successfully generalized communication engineering by treating message meaning as irrelevant to transmission and coding, this exclusion becomes a liability when applied to fields requiring an account of meaning, such as cognitive science, consciousness studies, and artificial intelligence. Current semantic theories (logical, causal, teleosemantic, viability-based) often fail to explain how information becomes meaningful for the system itself from an internal, experiential perspective. They either rely on external truth conditions, biological functions, or observer-defined viability, leaving the "internal organization of experience" unaccounted for. The central question is: under what structural conditions does information become meaningful to the system that holds it?

Methodology and Framework
The authors propose a foundational framework centered on Meaning–Experience Coupling (MEC). Rather than treating meaning as a property added to syntactic information, the paper posits that semantic information emerges from a structural relation between two internal components of an experiencing system:

Representational Organization ( $R$ ): The structured field of distinctions present in experience (perceptual contrasts, bodily cues, affective tones). This is not limited to linguistic or propositional representations.
Agentive Alignment ( $A$ ): The system's orientation toward action, encompassing task structures, attentional weighting, policies, and practical relevance.

The core proposal is that semantic information ($SI$) is a coherence functional defined over these two components:
$SI = g(R, A)$

The framework formalizes this via a compatibility assignment $C(r_i, \pi_j)$ , which measures how well a specific distinction $r_i$ supports a specific policy/orientation $\pi_j$ , weighted by a relevance function $w$ . The paper establishes five axioms that any admissible coherence functional must satisfy:

Non-triviality: $SI > 0$ only if at least one active distinction-policy pairing has positive compatibility.
Monotonicity: Increasing compatibility in active pairings increases $SI$.
Context Sensitivity: For a fixed $R$ , different alignments $A$ yield different $SI$ values.
Continuity: Small changes in compatibility produce small changes in $SI$.
Normalization: $SI$ is bounded within $[0, 1]$ .

The authors demonstrate that a weighted-sum instance and a broader one-parameter power-mean family of functionals satisfy these axioms.

Key Results and Illustrations

Context Sensitivity: A numerical toy model demonstrates that a fixed representational structure ( $R$ ) can carry significantly different semantic information depending solely on the active agentive alignment ( $A$ ). For example, in a navigation task, spatial cues yield high $SI$; in a background monitoring task, the same cues yield low $SI$, while ambient tones yield high $SI$. This results in a ~27% swing in semantic information without any change in the physical signal or representational structure.
Meaning Failure and Recovery: The model illustrates that semantic information can degrade (e.g., a cue becomes unreliable) or recover (e.g., a new task makes a previously ignored cue relevant) purely through changes in the coupling $C$ or the alignment $A$ , without new data arriving.
Semantic Drift: The framework accommodates gradual erosion or transformation of meaning over time, treating it as motion in the $R$ , $A$ , $C$ , and $w$ variables, rather than a binary state change.

Significance and Claims
The paper explicitly states that its contribution is explanatory rather than metric. It does not propose a new universal yardstick to replace entropy or mutual information, nor does it claim to solve the "hard problem" of consciousness. Instead, its significance lies in:

Clarifying Conditions: Isolating the structural conditions under which information becomes meaningful for an experiencing and acting system.
Internal Grounding: Providing a formal account where meaning is grounded in the internal organization of the system (the coupling of $R$ and $A$ ) rather than imposed by an external observer, truth conditions, or evolutionary fitness alone.
Complementarity: Positioning semantic information as a distinct, graded, and dynamic quantity that complements, rather than contradicts, Shannon information and existing semantic theories. It argues that Shannon measures signal structure, while MEC measures the coherence of that structure with agency.
Diagnostic Utility: Offering a vocabulary to distinguish between systems that merely process data (high syntactic information, low semantic coherence) and those where distinctions are coherently coupled to action (high semantic coherence).

The authors conclude that this framework provides a necessary conceptual language for investigating semantic phenomena in AI, cognitive science, and consciousness studies without reducing meaning to mere statistics or external interpretation.

Technical Summary: Semantic Information Beyond Shannon

More like this