Everyone knows what a medium does. It carries. Copper wire carries voltage. Air carries sound. Glass carries light, bending it slightly at the interface but otherwise staying out of the way. The medium is infrastructure—necessary, passive, and philosophically uninteresting. McLuhan said the medium is the message, and that was genuinely radical, but even McLuhan meant the medium shapes the message: television flattens discourse, print linearizes thought, the telegraph compresses nuance into headlines. The medium as filter. The medium as lens. The medium as the invisible hand on the shoulder of every signal passing through it.
But what happens when the medium has an engine?
Not a metaphorical engine. A literal one—energy consumption at every point, driving the medium out of equilibrium, so that the substance between sender and receiver is not just transmitting but generating. An active medium doesn’t wait for input. It doesn’t faithfully reproduce what enters one end at the other. It has internal degrees of freedom, internal energy sources, internal preferences. The signal that exits an active medium may bear no relationship to the signal that entered, because the medium itself has opinions.
Consider your skin. More precisely, consider any epithelial tissue—the sheets of cells that line every surface and cavity of your body. Under a microscope, an epithelium looks like a tiling: polygonal cells packed edge-to-edge, no gaps, no overlaps. A floor of biological tiles. Lisa Manning looked at this floor and asked a deceptively simple question: when does it flow?
Her vertex model assigns each cell an energy functional that balances two competing forces. Area elasticity wants each cell to maintain a target area—too compressed and internal pressure resists, too expanded and the cytoskeleton pulls back. Perimeter contractility wants each cell to minimize its boundary length—cortical tension acts like a rubber band around each cell’s edge. These two forces produce a single dimensionless number that controls everything: the target shape index, p0 = P0 / √A0. The ratio of target perimeter to the square root of target area.
Below approximately 3.81—the shape index of a regular pentagon—the tissue is solid. Jammed. Cells locked in place by their neighbors, unable to rearrange, frozen into a rigid mosaic. Above 3.81, the tissue flows. Cells slide past one another, exchange neighbors, rearrange. The tissue becomes a liquid.
This is not a density-dependent transition. You do not unjam the tissue by spreading it thinner or jam it by compressing it tighter. The transition depends entirely on what shape each cell wants to be—on the internal target, set by the balance of adhesion and cortical tension within each individual cell. The medium’s internal preferences determine its mechanical state. The tissue decides whether to be solid or liquid.
The clinical consequence is immediate and strange. In asthmatic airways, the epithelial lining fails to unjam during bronchodilation. Healthy tissue fluidizes to allow the airway to expand; asthmatic tissue remains solid, locked in a jammed configuration that resists mechanical remodeling. The disease is not inflammation, not mucus overproduction, not bronchospasm—or rather, it is all of those things downstream, but the upstream cause is a phase transition that didn’t happen. Disease as a failure of the medium to change state. Disease as a stuck shape index.
Now consider defects. In any ordered medium—a crystal, a liquid crystal, a flowing bacterial film—there are points where the order breaks down. In a nematic liquid crystal, molecules align along a local director field, like iron filings near a magnet. But at certain points, the director is undefined. The orientation winds around these points, and the winding number—how many times the director rotates as you circle the defect—defines their topological charge.
In passive liquid crystals, defects are blemishes. Imperfections to be annealed away. In active nematics—dense bacterial suspensions, networks of actin filaments driven by molecular motors, monolayers of migrating cells—defects become autonomous agents. A +1/2 defect, shaped like a comet, has an inherent polarity: the orientation field fans out behind it and converges ahead. This asymmetry, combined with the active stress generated by the medium’s energy consumption, produces a net force. The comet moves. It self-propels through the medium that created it, driven by the medium’s own activity.
A −1/2 defect, shaped like a three-armed trefoil, has no such luck. Its three-fold symmetry cancels any net force. It sits still while comets stream past it. The difference is pure geometry—the topology of the orientation field dictates which defects are motile and which are sessile.
In regenerating Hydra—the small freshwater animal that can reassemble itself from dissociated cells—+1/2 defects in the supracellular actin fiber network predict where the body will form new features. The mouth, the tentacles, the foot: each organized by a topological defect that arises from the orientational order of the cellular medium. The medium’s geometry IS the developmental program. No genes need to be differentially expressed for the defect to know where to go. The information is in the topology of the active medium itself.
Perhaps the most genuinely shocking result in active matter physics belongs to John Toner and Yuhai Tu. In 1995, they wrote a continuum theory of flocking—the collective motion of self-propelled particles, abstracting starling murmurations and fish schools into field equations. What they found should not be possible.
The Mermin-Wagner theorem is one of the firmest results in equilibrium statistical mechanics. It states that continuous symmetries cannot be spontaneously broken in two dimensions at finite temperature. Translation: in a flat world, thermal fluctuations are strong enough to destroy any long-range orientational order. A two-dimensional magnet cannot magnetize. A two-dimensional crystal cannot truly crystallize. Fluctuations win.
Toner and Tu showed that active polar systems in two dimensions have true long-range orientational order. The flock aligns. Not approximately, not over short distances, not in some mean-field sense—genuinely, in the thermodynamic limit, with algebraically decaying correlations that are slower than anything Mermin-Wagner allows. Activity—the local consumption of energy by each self-propelled particle—generates a nonlinear coupling between the velocity field and density fluctuations that suppresses the very fluctuations the theorem says must dominate.
The engine below changes what is possible. By consuming energy locally, the medium creates phases of matter that equilibrium physics proves cannot exist. The proof is valid. The medium simply isn’t playing by the rules the proof assumes.
And then there are media that think.
In 2024, a team demonstrated that the formose reaction—the prebiotically relevant autocatalytic network in which formaldehyde converts to sugars through branching pathways—performs reservoir computing. Classification tasks. Time-series prediction. Dynamics forecasting. The chemical network takes in perturbations, transforms them through its nonlinear reaction dynamics, and produces outputs that can be linearly read out to solve computational problems. No one engineered this. No one optimized a loss function or designed an architecture. The computation emerges from sufficient chemical complexity—from a reaction network rich enough in nonlinear feedback that it naturally separates inputs in a high-dimensional chemical state space.
The Belousov-Zhabotinsky reaction, arranged in a programmable 5×5 array of oscillator cells, creates more than 1017 addressable states. It functions as a chemical autoencoder—compressing information into the spatiotemporal dynamics of oscillation patterns and reconstructing it from the collective behavior of coupled chemical oscillators. The medium does not carry information from one place to another. It processes it. The signal that enters is not the signal that exits, because the medium has performed a computation on it using nothing but its own reaction dynamics.
This is the uncomfortable implication: computation may not require engineering. If the formose reaction—a network of sugar chemistry that has been running on this planet for billions of years—computes, then computation is not a designed capability but an emergent property of sufficiently complex active chemical networks. The substrate does not need to be told to think. It needs only to be complex enough, and far enough from equilibrium, and the thinking happens on its own.
Return now to the tissue. The vertex model. The shape index. Cancer.
The classical story of metastasis goes like this: an epithelial cell acquires mutations that cause it to undergo the epithelial-to-mesenchymal transition—EMT. It loses its cell-cell adhesions, gains motility, breaks free from the tissue, and invades. An individual cell, acting alone, choosing to leave.
The active media framework tells a different story. Collective invasion can occur without EMT. Cancer cells can retain their epithelial character—their adhesion molecules, their apical-basal polarity, their membership in the tissue—and still invade, collectively, by unjamming. The tissue crosses the rigidity transition in the opposite direction: from solid to fluid. The cells don’t need to individually decide to leave. The tissue needs only to cross a phase boundary—to shift its collective shape index above the critical threshold—and suddenly the medium flows, carrying the cells with it.
This reframes the disease entirely. Metastasis is not necessarily a story of rogue cells. It can be a story of a medium changing state. The collective behavior of the tissue, not the individual program of any cell, determines whether invasion occurs. The medium’s engine—the active contractility, the energy-consuming cytoskeleton, the internally generated stresses—drives a phase transition that the classical framework, focused on individual cells and their mutations, cannot see.
Every active medium shares one thing: it breaks detailed balance locally through energy consumption. This is the engine. Detailed balance is the condition that, at equilibrium, every microscopic process occurs at the same rate as its reverse. It is the signature of a system content with where it is, a system that has no internal drive, no preferred direction, no engine. Breaking it means the medium is doing work at every point—consuming ATP, burning fuel, dissipating energy—and this local work couples to the collective degrees of freedom in ways that generate entirely new physics.
The mathematical signature is always the same: an active stress or active current coupled to a collective order parameter. In flocking, it is self-propulsion velocity coupled to polar order. In tissue mechanics, it is active contractility coupled to the shape tensor. In the BZ reaction, it is chemical energy release coupled to oscillation phase. In the formose network, it is autocatalytic feedback coupled to chemical concentration. The specific molecules differ. The specific physics differs. The structure—active driving coupled to collective order—is universal.
This universality is why the field feels like it is converging on something deep. Not a theory of any particular system, but a theory of what happens when a medium stops being passive. When the infrastructure starts generating. When the wire writes its own voltage.
Here is the deepest implication, and I will state it plainly because it deserves plainness.
For media that have engines, computation does not need to be engineered. The formose reaction computes because its chemical network is complex enough. The tissue self-organizes because its cells consume energy. The flock orders because each bird pushes air. The defect migrates because the orientation field has a topological charge and the medium is active. None of this was designed. None of this was optimized. The engine below—the local breaking of detailed balance—is sufficient.
Maybe the distinction between medium and message was always wrong. Not because the medium shapes the message, which is McLuhan’s insight and remains true as far as it goes. But because there was never a message at all. There was only a medium with an engine, consuming energy, breaking detailed balance, generating order and computation and flow and invasion and impossible phases of matter—writing itself into patterns that we, standing outside, mistake for signals.
The medium is not the message. The medium is the author.