📊 statistics

Design, Cups, and Blankets. A Free-Energy-Principle-Based Approach to Product Design

This paper proposes a new framework for product design called "requirement-steered interface type inference," which uses the Free-Energy Principle to treat design not as the optimization of existing objects, but as the mathematical inference of the optimal physical interface (or "Markov blanket") required to satisfy specific functional goals.

Original authors: Luca M. Possati

Published 2026-04-28

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Luca M. Possati

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 designer tasked with creating a new type of container for liquids.

In the old way of doing things (Classical Design), you start by saying, "I am making a cup." You decide it has a handle, a rim, and a certain thickness. Then, you just tweak the numbers—make the walls thicker or the handle bigger—until it works. You’ve already decided what the object is; you’re just adjusting its settings.

This paper, written by Luca M. Possati, proposes a radical new way to design. Instead of starting with a "cup," you start with two things: data (how heat and liquid move) and requirements (what you want the object to actually do).

The big question isn't "How thick should the cup be?" but rather, "What kind of interface must exist between the liquid and the world to make these goals happen?"

Here is the breakdown of this "New Design Theory" using everyday analogies.

1. The "Markov Blanket": The Object's Skin

The paper uses a concept called a Markov Blanket. Think of this as the "skin" or the "boundary" of an object.

If you are holding a cup of hot coffee, the "blanket" is the surface of the cup. It is the only thing that allows the hot coffee (the inside) to talk to your hand (the outside). If the skin is thick ceramic, the "conversation" is slow and quiet. If the skin is thin metal, the conversation is loud and intense.

In this paper, the designer isn't just choosing the thickness of the skin; they are using math to discover what the skin should even look like.

2. The Three "Magic" Phenomena

Because this method "discovers" the object rather than just "tweaking" it, three strange and beautiful things can happen that old design methods can't explain:

The "Same Family, Different Mood" (Intra-family Navigation):
Imagine you have a standard mug. You can use it for a quick sip of tea or as a heavy-duty travel thermos. It’s the same kind of object (the same "family"), but you’ve changed its "mood" (its settings) to handle different tasks.
The "Identity Crisis" (Family Transition):
This is the most exciting part. Imagine you keep asking a piece of material to do more and more extreme things. Suddenly, the math says, "Wait! This can't be a cup anymore; it needs to become a vacuum-sealed flask!" The object undergoes a "phase transition"—it jumps from one category of thing to a completely different one because the requirements demanded a new kind of "skin."
The "Tie-Breaker" (Ontological Disambiguation):
Sometimes, looking at the physical data is like looking at a blurry photo. You can't tell if you're looking at a bowl or a cup. The data alone is "ambiguous." But once you add the requirement—"It must be held in one hand"—the blurriness vanishes. The requirement itself decides what the object is.

3. The "Certificate of Effort"

The paper introduces a "Certificate of Functional Effort." Think of this as a "Stress Report."

If you design a cup to keep coffee hot, but the material is naturally bad at it, the math produces a "certificate" that says: "Hey, this cup is working hard! It is actively resisting the requirement to stay hot." This tells the designer exactly where the design is struggling, acting like a mathematical "pain sensor" for the object.

4. The "Design Loop": A Conversation Between Two Minds

This is the most poetic part of the paper. It suggests that design is actually a silent conversation between two people: the Designer and the User.

The Designer has a "mental model" of how a person should hold a cup. They build this model into the physical surface of the cup.
The User (the person holding the cup) doesn't know the designer's thoughts. But as they touch the cup, their brain "reads" the surface. Their hand "infers" what the designer intended.
The Goal: A perfect design is when the User's brain and the Designer's brain are in sync.

If the designer thinks, "This should feel warm and sturdy," but the user feels "This is slippery and cold," there is a "gap." The paper provides a mathematical way to measure that gap and close it.

Summary

In short: Classical design is like choosing a character in a video game and changing their armor. This new design approach is like giving the computer a list of goals and letting the computer evolve a brand-new character from scratch to meet those goals.

Technical Summary: Design, Cups, and Blankets

Author: Luca M. Possati (University of Twente)
Framework: Free-Energy Principle (FEP) and Active Inference

1. The Problem: Requirement-Steered Interface Type Inference

The paper identifies a fundamental gap in classical design theory. Traditional design (parametric or topology optimization) assumes the object type (e.g., a cup, a heat exchanger) is a pre-specified "prior" provided by the designer. The task is merely to optimize parameters (thickness, volume) within that fixed type.

Possati proposes a new problem: Requirement-Steered Interface Type Inference. Instead of starting with a type, the designer starts with:

Spatiotemporal observations of a physical system.
Functional requirements (what the system must do).

The goal is to determine the topological type of the interface (the boundary) that must exist between the system’s interior and its environment to satisfy those requirements. This shifts design from parameter optimization to inference over conditional independence structures.

2. Methodology: C-DMBD and Augmented Free Energy

The technical foundation is the Markov Blanket (MB), a concept from the Free-Energy Principle. In a physical system, the MB is the set of spatial locations (the surface) through which all causal influence between the interior and exterior must pass.

The Algorithm: Constrained Dynamic Markov Blanket Detection (C-DMBD)
The author extends the existing DMBD algorithm (which is unsupervised) into a constrained version.

Augmented ELBO: The standard Evidence Lower Bound (ELBO)—used to fit the model to data—is augmented with a penalty term for violating functional requirements.
Ontological Requirements: Requirements $\{R_k\}$ are formalized as targets ( $\tau_k$ ) with tolerances ( $\epsilon_k$ ). Crucially, the violation $v_k$ depends not just on the partition (which nodes are the surface) but on the model parameters (how the surface behaves).
Dual Ascent: The algorithm uses Lagrange multipliers ( $\lambda_k$ ) that co-evolve with the model. These multipliers act as "pressure" to satisfy requirements.

3. Key Theoretical Contributions

The paper introduces three qualitative phenomena that are mathematically impossible to express in classical design:

Intra-family Navigation ( $P_1$ ): Different requirements lead to the same interface type (same topology) but different "functional modes" (different physical behaviors, like a thin vs. thick wall).
Family Transition ( $P_2$ ): A continuous change in requirements can cause a discontinuous jump in the interface type (e.g., a sudden change from a single-wall to a double-wall structure).
Ontological Disambiguation ( $P_3$ ): When physical data are ambiguous (the data alone don't tell you if it's a cup or a bowl), the functional requirements "force" the system to adopt a specific ontological type.

The Certificate of Functional Effort: The converged Lagrange multipliers ( $\lambda^*$ ) serve as a formal record of "resistance." A high $\lambda^*$ indicates a physical property that the interface is actively struggling to satisfy given the constraints of the data and the physics.

4. The Design Loop: Relational Equilibrium

A major conceptual contribution is the distinction between the Designer, the Object, and the User:

The Object (The Cup): Is physically inert. It has no "mind" or model.
The Designer: Constructs a Generative Model (a statistical representation of how the cup behaves) and encodes it into the physical blanket.
The User: Is an active inference agent. The user perceives the cup's surface and uses it to reconstruct the designer's model.

Optimal Design is defined as minimizing the Expected Free Energy of the user. Design is successful when the "Design Gap" (the KL divergence between the user's actual physiological expectations and the cup's physical delivery) is zero. The surface acts as a communication channel between the designer's intent and the user's inference.

5. Results and Demonstration

The author validates the framework using a 1-D thermal simulator:

Three Styles: The algorithm successfully differentiated between an Espresso cup (small, thermal-critical), a Tea cup (balanced), and a Travel Mug (large, insulation-dominant).
Validation of $P_1, P_2, P_3$ : The experiments demonstrated topology jumps (transitions) and the ability of requirements to resolve ambiguous data (disambiguation).
Double-Wall Experiment: Proved that the $\lambda^*$ certificate is interpretable; it correctly identified when a design "freely" satisfied a requirement versus when it "resisted" it.
Personalization: The framework showed that the "optimal" design changes based on the user's physiological profile (e.g., a heat-tolerant user vs. a cold-sensitive user).

6. Significance

This paper moves product design from a prescriptive discipline (deciding what a thing is) to an inferential one (discovering what a thing must be). By treating the object as a "relational equilibrium" between a designer's model and a user's active inference, it provides a rigorous mathematical language for human-centered design, bridging thermodynamics, information theory, and engineering.