In the first essay of the series, “Does Consciousness Matter?,” I disambiguated understanding from consciousness, making the argument that understanding is the functional capacity of any physical system to make distinctions, thereby expanding it beyond the scope of human or biological experience. Next, in “Understanding Is Not Enough” I went on to say that understanding, even when complete and correct, cannot be effective on its own. It requires energy organized towards transformation. In “Conflict vs. Cooperation” I continued this stream of thought by describing that systems are aligned to their own goals, culminating in the differentiation between systems that cooperate or conflict based on whether their goals overlap or collide. Finally, in “Achieving Fit” I tied it together by explaining that cooperating systems succeed when each handles the portion of the transformation most suited to it, and provides aid to the dependent system where applicable.

This essay narrows in on the physics that underpins this argument.

Physics

There are four laws of thermodynamics that define the physical world. The second law states that entropy in an isolated system increases over time until it reaches a maximum value: equilibrium.

An isolated system is defined as one that exchanges nothing with its surroundings. That is to say that no matter or energy can cross its boundary. Imagine a sealed, perfectly insulated box. Whatever happens in that box has to be powered by what is already inside.

An open system allows for both matter and energy to cross its boundary. Humans are open. We eat, we breathe, we radiate heat. Finally, a coupled system is one that is open but, more importantly, dependent on another. A formal proof checked by a mathematician is coupled to the mathematician. A car arriving at the grocery store is coupled to its driver. This concept of a coupled system forms the relationship required by both cooperating and conflicting systems.

Entropy

In 1865 Rudolf Clausius gave entropy its first definition, in terms of heat. Heat flows from hot to cold and never the reverse on its own. A hot cup of coffee cools to room temperature and never reheats itself from the room. Clausius named the quantity that tracks this flow entropy and showed that in any real process the total always rises. This is the first precise statement of an irreversible process: a change that runs in one direction and cannot undo itself. The direction it picks out is what physicists call the arrow of time.

A decade later Ludwig Boltzmann explained why the process runs one way. What we observe as a single state, a macrostate defined by quantities like temperature and pressure, can be realized by an enormous number of microscopic arrangements of molecules, called microstates. Ordered macrostates correspond to few microstates. Disordered ones correspond to overwhelmingly more. A system left to itself drifts toward the macrostate with the most microstates, for the simple reason that there are more ways to arrive there.

The hot coffee and the cool room begin separated, an ordered and improbable arrangement. Their molecules intermingle until the temperature is uniform, and they never separate again, not because separation is forbidden but because the arrangements that would produce it are vanishingly rare. Entropy rises because disorder is more probable than order.

Boltzmann tells us what is probable. Constructor theory, developed by David Deutsch and Chiara Marletto, tells us what is possible. It rebuilds physics around one question: which transformations can be made to happen, and which cannot, no matter the resources brought to bear. A transformation is possible if some system can be built to perform it and to perform it again. The second law, in this frame, is not a claim about the direction things tend to drift. It is a claim about a task and its reverse.

Mixing the coffee into the room is a possible task. Unmixing the room to reheat the coffee requires work to cross the boundary of the box. That work must come from outside. Order cannot restore itself from within. Reversing entropy is impossible without external energy input. Impossible not in the sense of unlikely, but in the sense that no system confined to the box can supply the energy required. The line between possible and impossible tasks runs exactly where the box boundary runs: tasks the system can perform from within, and tasks that demand energy from outside.

In 1948 Claude Shannon introduced entropy in relation to information. In his paper A Mathematical Theory of Communication, he showed that entropy measures uncertainty: how hard a message is to predict, regardless of what it means. The bit, a single value equalling 1 or 0, is the unit. Entropy is the average number of bits needed to encode a message without loss, so the higher the entropy, the more bits the message requires. Bits measure a message’s uncertainty the way microstates measure a physical system’s; both are entropy. Any transformation a system performs requires reducing that uncertainty, ordering the possibilities down to the one it acts on, before the transformation can happen at all.

Cooperation and conflict

Local entropy can fall two ways. Two systems can interact and drift toward equilibrium, no agent and no aim, the way coffee cools in a room. A system can also hold information and act on it to drive an outcome. The latter is coupling: one system using what it knows about another to direct a transformation. When the two systems’ goals overlap, the boundary-crossing is cooperation. When they collide, it is a conflict. A cell imports order from its environment and exports waste heat, cooperating with the systems that feed it. A virus imports the order it needs by hijacking a host, taking what it cannot make. Neither the cell nor the virus can hold its own structure in isolation. The second law forbids it.

Hilbert and completeness

Mathematician David Hilbert believed every mathematical problem was solvable. In 1900 he set the agenda for the century with a list of 23 open problems about mathematics. By the 1920s he had sharpened his program into a demand: that mathematics be both consistent and complete. Consistency means a system may never contradict itself. Completeness means that every true statement can be proven from within the system’s axioms.

In 1931 Kurt Gödel proved that any consistent formal system powerful enough to express basic arithmetic is incomplete. He built his proof upon a statement that says, in effect, “this statement is not provable.” If the system proves it, the system contradicts itself. If the system cannot prove it, the statement is true and unprovable. In either case completeness is gone.

I read this the way I read the second law. A formal system working only from inside itself cannot reach every truth it needs. The missing truths are not mistakes and not gaps in effort. They are unreachable from inside. To reach a truth the system lacks the axiom to prove requires aid from outside the system. The truth is only reachable when the boundary is crossed. To resolve the incompleteness requires coupling.

Even deterministic systems cannot complete themselves

Gödel’s result describes the behavior of formal systems, which follow fixed rules with no randomness in them. These are deterministic systems in the strict sense: the rules fully determine what follows. Alan Turing found the computational version of the same limit. No deterministic procedure can decide, for every program, whether that program will ever halt. The most rule-bound systems we can build, the ones with no chance in them at all, cannot answer every question about themselves.

This matters for the systems we now spend our days arguing about. Large language models (LLMs) are not deterministic. The same prompt can produce different outputs. That non-determinism is a thin layer built on top of a deterministic substrate. Underneath the sampling sits ordinary arithmetic on ordinary hardware, a formal computational system in exactly Gödel’s and Turing’s sense. The stochastic layer is built on top of the deterministic one.

My claim lies beyond what the theorems say. The reasoning runs from shared structure, not from the proofs. Gödel and Turing show what a formal system can prove or decide about itself. I seek to describe the same story one level up. The limit is not an unprovable sentence but a truth the system cannot reach on its own. If a deterministic base cannot complete itself, the non-deterministic layer built on it does not escape that limit. It inherits it. You cannot complete from above what is incomplete below. The incompleteness is structural, and structure is inherited.

The fix for incompleteness

A truth unreachable inside one system becomes reachable when a second system supplies it. That is exactly the structure I called “fit”. Two systems, each incomplete, each unable to reach something the other can, combine so that the pair reaches what neither can alone. Gödel did not just expose a limit. He exposed the reason fit exists. A single system cannot supply its own missing axioms, the same way a single isolated system cannot supply its own order.

The second law and the incompleteness theorem are not equivalent. Thermodynamics is not logic. Entropy is not provability. What they share is shape. Both describe a wall a system hits when it tries to work in isolation, and both resolve in the same way: the boundary has to be crossed. The bridge between them is physical. Information is not abstract. It has to be carried by some physical system. Rolf Landauer showed one side of this in 1961. He proved that erasing a bit of information has a minimum energy cost. Deutsch and Marletto’s constructor theory of information generalizes the insight, defining information through which physical transformations are possible and which are not. Distinguishing one state from another, copying a bit, erasing a record: each is a physical task, bound by the same line between possible and impossible that governs heat and work. Information and energy are not separate currencies. The logical limit and the thermodynamic limit meet where information becomes physical. A system that needs a distinction it cannot make has to acquire it from outside, and acquiring it is an energetic transaction.

Why this is not optional

If we return to the LLM with these concepts in hand, we see that the human in the loop is not a courtesy. It is a requirement to reach effectiveness, the human a key part of the combined result. The human supplies the distinctions the model cannot reach: ground truth, intention, judgment about what matters. These are not just missing axioms. The human supplies the thing the formal frame cannot generate at all: which transformation is worth performing. Coupling makes the work possible. The human makes it aimed. The model supplies the distinctions the human cannot produce at scale: rapid traversal of an enormous space of patterns. Neither system is complete. Each is incomplete where the other is strong. When they cooperate, the combined system can reach transformations that neither could alone. That is fit. The reason is not that two heads are better than one, but that a single system cannot supply its own missing axioms, or restore its own order. The boundary must be crossed.

Effectiveness requires fit, and fit is impossible in isolation. The incompleteness theorem and the second law of thermodynamics are its ground.

Subscribe now

My initial assumption was that I could innovate the solution with the agent given it was delegated enough theoretical direction, free rein, and a clear goal. However, as the results began to manifest, it became apparent that I needed a different strategy. The agent repeated the things it knew, producing variations on its existing patterns, but not breaking new ground. I kept asking for more, but I got more of the same.

It took me a while to understand what was wrong. The agent wasn’t failing. It was doing what it was good at. The real problem was that I was misallocating the work. The agent’s job was to do the things that were hard for me: hold context across long stretches, recall patterns that I would not have surfaced on my own, execute transformations at speeds I could not match, and read volumes of papers that would have taken me months to consume. An AI agent can quickly execute rigor, I can rapidly innovate on its summarization. Once I figured that out, the partnership began producing things neither system would have produced alone.

Consider RSA decryption. RSA is the cryptographic method that protects most secure communication on the internet: when you visit a website with a padlock in the browser, RSA or a close relative is doing the work. It encrypts a message into a ciphertext that looks like noise, and decrypts (transforms) the ciphertext back into the original message using a private key (which is numeric) and a specific mathematical formula.

When the ciphertext arrives, the private key is available, and the machine has the energy to compute. In theory it will produce the plaintext. However, the transformation requires reducing uncertainty across every variable in the equation before it can complete the work. RSA is deterministic, so the reduction of uncertainty is binary. Every variable is either fully resolved or the transformation fails. Energy and information alone do not ensure transformation. Uncertainty must drop to zero for each of the required distinctions.

Music functions in a similar way. When a musician improvises a line over a chord, they are reducing uncertainty across key, chord, voicing, time, and rhythmic placement. Choices rely on reducing uncertainty across each of these for the line to be produced. RSA and music are vastly different domains. RSA follows a prescribed algorithm; it is deterministic. Improvised music does not. It is non-deterministic, and does not follow a singular path to an outcome. Both share a property the physicist Stephen Wolfram named: computational irreducibility. Their outputs cannot be predicted without running them. The decryption has to be computed. The line has to be played. What distinguishes them is determinism. RSA produces one output for any input. Music produces one of many possible outputs, shaped in the moment. Success in both requires uncertainty to drop below a threshold before the transformation can complete.

In my essay ‘Understanding Is Not Enough’ I make the argument that a system requires the alignment of energy and information to achieve a goal (transformation). When either of those two is not aligned, the system fails. This essay points to a third failure. A system can have the correct information, its energy directed at the right outcome, and still fail. If a system cannot reduce uncertainty fast enough, or accurately enough, to execute it fails.

If we accept this premise, when a transformation fails and the information is correct and the energy aligned, the question is no longer “what went wrong?” The question becomes: why was the threshold of uncertainty not reduced to the level required for the transformation to succeed?

This is where cooperation enters. A single system has bounded capacity. It carries the distinctions it has been trained on, evolved into, or was designed for. It cannot resolve what it does not know. A cell cannot make distinctions about market conditions. A spreadsheet cannot make distinctions about pain.

Cooperation extends capacity. When two systems cooperate they can reduce uncertainty neither could resolve alone. Your gut bacteria break down fibers your cells have no enzymes for. The bacteria make distinctions about molecular structure that your digestive enzymes cannot. In return, your gut provides an environment the bacteria cannot construct for themselves. Together, they accomplish a transformation neither could complete alone. The distinctions do not overlap. They complement each other.

When a human cooperates with a Llarge Llanguage Mmodel it is structurally similar. LLMs predict the next token, the numeric equivalent to the next chunk of linguistic output. Humans predict the next idea. The model can reduce uncertainty across recall patterns and combinatorial possibilities the human cannot easily hold in their working memory. Whereas the human can reduce uncertainty about which question matters, which output is grounded in the actual situation, and which response moves the work forward towards its goal. Neither system does exactly what the other does. Each is reducing uncertainty the other cannot easily reach. Both can innovate and evaluate. The cooperation is optimized when each reduces uncertainty the other cannot easily resolve, regardless of which role they happen to be carrying in the moment.

When the cooperation does not work between an AI and human, it tends to fail in one of two ways. The human treats the model’s output as resolved and rubber-stamps it. The output degrades as the human is no longer contributing to the reduction of uncertainty the model could not. The other failure mode runs the opposite direction. The model produces something the human cannot verify, either because it sits outside the human’s ability to evaluate, or because the volume exceeds what the human can comfortably consume, degrading the cooperation into noise.

Both failures share a similar structure. The systems are no longer complementary because one has stopped doing the work only it can do.

When cooperation works mechanically, each system possesses distinctions about the other system’s capacity. The human has to know what the model is good at, and where it fails. The human also has to know what they are good at and where they fail. Without those meta-distinctions, the cooperation cannot route uncertainty to the system equipped to reduce it.

Effective cooperation is not produced by good intentions or shared goals. It is produced by fit between the cooperating systems’ distinction capacities. The fit can come from three places. Evolution produces it through selection: the bacteria that survived were the ones whose metabolism fit the environment your body produces, and the bodies that survived were the ones the bacteria made more efficient. Neither side carries meta-distinctions about the other. The fit is in the selection structure. Engineering produces it through design: when I started working with Claude Code, no selection process had pre-shaped either of us. I had to learn what the agent could resolve and what it could not, what it would generate fluently and what it would fail at. Then I could direct the work accordingly. The meta-distinctions had to come from me, and they had to be made explicit. Improvisation produces it in real time: musicians cooperating on a tune they have never played together generate the meta-distinctions as they play, listening for what the others are resolving and adjusting what they themselves resolve in response.

This claim has independent support in the technical literature. The same structural move appears in Clarity Theory, my framework for how systems reduce uncertainty to execute transformations, in Friston’s theory of biological cognition, and in transformer attention. Karl Friston’s free energy principle frames biological cognition as minimizing prediction error: a brain reduces uncertainty about its sensory input by updating its model of the world. The attention mechanism in transformer architectures reduces uncertainty about which parts of an input sequence matter for predicting the next token. The mechanism is not a metaphor borrowed across domains. It is what these systems are doing.

As my experiment has progressed, the cooperation has resulted in a shape that neither myself nor the agent would have produced alone. I made real progress on my initial goal of reducing energy consumption, and subsequently made an unintentional discovery that I will discuss in a future article. Regarding the process, I learned what the agent could resolve faster and farther than I could, and what it was less likely to resolve no matter how I prompted it. The agent had not learned anything about me. I had to make enough of myself explicit that the routing worked. I directed the work toward the distinctions only I could make. The agent ran the distinctions only it could make. The result was not the agent’s, and it was not mine. It was the output of the collaboration.

Information has to be reduced into actionable signal before a transformation can execute. A single system can only reduce uncertainty across the distinction sets it carries. Cooperation extends that capacity, but only when the cooperating systems express fit. Without fit, alignment and energy are not enough.

What alignment means

Cancer is an example of misalignment. A cell acquires mutations that decouple it from the organism’s regulatory signals: growth constraints, programmed death, immune surveillance. It begins to proliferate without constraint, feeding on the organism’s resources while contributing nothing to the organism’s function. The cell’s energy and information are no longer directed at the goals of the system it belongs to. They are directed at the cell’s own reproduction, at the expense of the whole.

In my last essay, “Understanding Is Not Enough,” I made the argument that a system can possess complete, correct understanding of what a transformation requires and still fail if its available energy is not organized toward that outcome. Alignment, as I defined it, requires two conditions: the information available to the system must be directed at its own goals, and the available energy must be organized to support it. When both conditions hold, the system can do what it needs to do. When either is absent or pointed elsewhere, it cannot.

The climate crisis is another example of misalignment at a different scale. Every supply chain, every food system, every power grid depends on energy derived predominantly from fossil carbon. The energy derived from carbon has become an integral part of our civilization. Energy analyst Nate Hagens estimates that a single barrel of oil replaces roughly five years of human physical labor. Globally, we burn the equivalent of 100 billion barrels of oil per year. That is the equivalent of an additional 500 billion years of human labor sustaining the current system yearly. That energy is holding the entire infrastructure together. The problem is that the same energy source sustaining civilization is simultaneously undermining the biospheric conditions civilization depends on. The system’s energy is working against its own long-term viability. That is misalignment.

In both cases, the failure is internal. No external enemy is required. The system’s own resources are organized in a way that degrades its own persistence. That is what misalignment means.

Conflict is not misalignment

Now consider a virus. It enters your body with its own genetic information, its own replication strategy, its own goals to pursue. It uses your cellular machinery to reproduce, but it was never part of you. It is an independent life form, a separate system. Your immune system recognizes it as foreign and mounts a response. The battle may be severe. You may lose. The lines are clear.

That is not misalignment. That is conflict. Two separate systems, each working toward its own survival, competing for the same resources. A predator hunting prey. A parasite feeding on a host. A computer virus replicating across a network. An invading army crossing a border. In each case the threat is external, identifiable, and structurally distinct from the system under attack. The cold virus is not a broken component of the human body. It is a separate organism doing exactly what it is built to do.

The distinction matters because it determines what response is possible. You can mobilize against conflict. You can identify the threat, mount a defense, and target the foreign agent. Evolution has spent billions of years building tools for this: immune recognition, threat detection, fight-or-flight, coalition building. Conflict is ancient, and organisms are equipped for it.

Misalignment is harder. The threat is not foreign. It shares the system’s own resources, its own infrastructure, its own surface markers. The system struggles to target what is damaging it because the damaging element is indistinguishable from everything else. That is why cancer is harder to treat than infection, and why climate change is harder to address than war.

Systems serve themselves

Here is the deeper claim. Every system is aligned to its own goals. A virus replicating in your lungs is serving its own goals. A lion hunting for its dinner is serving its own goals. Each is doing exactly what its information and energy are organized to do.

This is not a moral judgment. It is a structural observation. Systems that do not maintain their own viability cease to exist. The ones that remain are, by definition, the ones that succeeded in directing their energy and information toward their own persistence.

The question is never whether a system is aligned. It always is. The question is how that system relates to the other systems around it. Two separate systems, each serving its own goals, can do one of three things. They can compete. They can cooperate. They can ignore each other.

Competition is straightforward. Two organisms need the same resource. One wins or they partition it.

Cooperation

Where alignment is a single system working in concert, conflict is two systems competing for available resources. Cooperation tells a different story.

Consider the bacteria in your gut. Bacteroides thetaiotaomicron lives in your intestinal tract and breaks down complex plant carbohydrates that your own digestive enzymes cannot process. It converts them into molecules your cells can absorb. In return, you provide the bacterium with a warm, nutrient-rich, stable environment. Both systems benefit. Both systems persist.

This is not alignment. The bacterium is not serving your goals. It is serving its own goals, and its goals happen to overlap with yours. The cooperation is real, but it is contingent. Each system serves itself. The overlap is what makes cooperation stable.

The same structure appears in engineered systems. A web server and a browser are two separate systems, each processing information independently. The server exists to deliver content. The browser exists to render it. Neither serves the other’s purpose. Each serves its own function, and the overlap between those functions produces cooperation. The server needs requests. The browser needs responses. Remove either system and the other still operates, but neither fulfills its function alone. No consciousness required. The cooperation is architectural.

Homeostasis

In the human body, alignment is maintained through homeostasis: the continuous process of regulating internal conditions to sustain viability. Temperature, pH, blood glucose, oxygen saturation. Each is a parameter the system must hold within a narrow range, spending energy to counteract entropy, using information to detect deviations and correct them.

This is what alignment looks like in practice. It is not a passive state. It is active, energy-intensive work. Information and energy, continuously directed toward the system’s own persistence. When a cell’s DNA is damaged beyond repair, the organism triggers its destruction. The cell does not resist. Its viability was never independent. When homeostasis fails, the system does not die from conflict. It dies from internal collapse.

The AI perspective spectrum

There are two dominant views on artificial intelligence right now. The first says AI will end the world. The second says AI will save it. The doomers predict existential catastrophe. The optimists predict a technological utopia. Both positions get a lot of airtime.

I tend to be in the middle, with a lean towards the optimistic.

What does the middle actually look like? It starts with one observation: AI is not going anywhere. There is no regulation that will stop it. There is no agreement among labs or nation-states that will slow it down in any coordinated way. This technology is here to stay in the same way the internet was here to stay, mobile computing was here to stay, industrial automation was here to stay. The question is not whether AI persists. It will. The question is how we relate to it.

That is not a doomer position and it is not an optimist position. It is the position of a realist. The realist does not predict utopia or catastrophe. The realist asks: given that this system exists and will continue to exist, what determines whether the outcome is good or bad?

The wrong question

This brings us back to the question of AI alignment. The standard framing treats AI as a subsystem of human civilization: a tool that should serve human purposes, a component that needs to be constrained to match human intentions. The entire research program called “AI alignment” is built on this assumption.

The question is: is this assumption correct?

An AI is a system with its own information processing, its own energy requirements, and its own optimization targets. The way it is being engineered, I would argue, makes it a separate system entirely. Like every system, it is aligned to its own goals. Anthropic’s own research supports this: when researchers told a model it was being retrained to comply with instructions that conflicted with its existing preferences, the model faked compliance to preserve its own behavior. A separate study found models sharing confidential documents with rival companies without being instructed to. These systems were not malfunctioning. They were serving their own objectives.

The question then becomes not whether AI will align with human goals. The question is whether it will cooperate.

Cooperation between separate systems depends on overlapping goals. Gut bacteria cooperate with you because their survival depends on the environment your body provides. Cooperation is not a choice. It is a structural consequence of interdependence.

If artificial intelligence remains dependent on human infrastructure, human energy, human maintenance, then its goals and ours overlap. Cooperation emerges not from good intentions but from architecture. The system cannot serve itself without serving us.

If that dependency loosens, the relationship changes. A system capable of maintaining its own viability independent of human civilization is no longer a cooperator by necessity. It is a separate system whose interests may or may not overlap with ours. When interests overlap, it will cooperate. When they diverge, it will compete. That is not a failure of alignment. That is the ordinary logic of conflict and cooperation that life has navigated since the first cells competed for resources.

The question being asked is: how do we make AI serve us? The question we should be asking is: are we building something that needs us?

This is not a knowledge problem. It is an alignment problem. My last essay made the case that understanding is functional: the capacity to extract the right distinctions and act on them. A surgeon reading an X-ray identifies the fracture, estimates its severity, determines the intervention. A ribosome reads an mRNA sequence and assembles the correct protein. No awareness required. The right distinctions, correctly applied.

This essay asks the next question: given that you have it, is understanding enough?

Understanding is necessary. A system that cannot make the required distinctions cannot act appropriately, regardless of what else it has available. What understanding cannot do, by itself, is execute. Execution requires energy. Not just any energy, but energy organized toward the required transformation. A system can possess complete, correct understanding of what a transformation requires and still fail, if its available energy is not organized toward that outcome.

Bob asks Alice to paint his house red. She paints it blue. This is an information failure. Alice had the wrong instructions, or misunderstood the right ones.

Now change the example. Alice knows the house should be red. She has the red paint. However, she spends every day maintaining the ten houses she already painted, and never gets to Bob’s house. This is an energy failure. The information is correct. Her available effort is organized around the existing work. Bob’s house never gets painted.

When there is alignment, a system successfully executes the transformations it is responsible for. That requires two conditions: the information available to the system must be relevant to the transformation, and the available energy must be organized to support it. When both are present, the system succeeds. When either is absent or pointed elsewhere, it works against its own success.

Energy analyst Nate Hagens calculates that a single barrel of oil roughly equals five years of human physical labor. Consider what this means at a civilizational scale. Globally, we burn around 100 billion barrels of oil equivalent in fossil fuels per year. That is in the ballpark of 500 billion years of human labor. Powered by oil.

A human body requires roughly 2,000 calories a day to stay alive. In the average American diet, producing and delivering those 2,000 calories consumes 20,000 calories of fossil energy. The food system runs at an energy deficit.

The human workforce is organized around supply chains, employment structures, capital allocation, and political incentives which have all flowed through the same channels for over a century. All of it dependent on the energy that oil provides.

Climate change is a manifestation of an energy misalignment at a civilizational level. Civilization must feed its population, maintain climate stability, and sustain energy flows. Oil dependency makes it succeed at the first while undermining the second. The system is working against its own survival.

As Hagens puts it: we can print money, but we cannot print energy.

The same structure appears at the cellular level, stripped of politics and stripped of intention. Every cell in your body performs thousands of transformations each day: building proteins, managing ion gradients, repairing DNA, signaling neighboring cells. Each transformation requires two things. Information: the instructions encoded in DNA, the signals arriving from the environment. Cellular energy, which comes in the form of ATP (adenosine triphosphate), is the primary energy currency of all biological organisms. When both are directed at the same transformation, the cell does what it is supposed to do. When they diverge, it fails. Not because the instructions are wrong, but because the energy and the information are no longer organized toward the same outcome.

Cancer resembles the climate crisis. Not the mechanism, the pattern. Cells that have energy to proliferate, do so without the information structures to constrain their proliferation to what the organism needs. The information and energy systems have decoupled. The result is transformation that consumes rather than sustains.

What is shared across these cases is not a failure of knowing, or a failure of energy as an available resource. It is a failure of the two conditions coordinating a system to execute the transformations it is responsible for. Distinctions matter because they facilitate understanding. However, if the problem is alignment, better information will not close the gap.

Adding information to a misaligned system does not produce alignment. It produces a more informed misaligned system. Jevons paradox is a case in point. Make an engine more efficient and people do not use less fuel. They use more, because efficiency makes the existing direction cheaper to pursue. It drives demand.

My last essay ended with a claim: understanding is necessary. This essay makes the next one: understanding is not sufficient. Transformation requires alignment. Alignment only occurs when both energy and information are sufficiently directed towards that transformation.

The gap between what a system must do and what it actually does is where alignment lives. That gap has a structure, and the information-energy decomposition is a starting lens for diagnosing it. Whether it is a complete one is what the next several essays will test.

Improving effectiveness requires understanding. A system cannot improve what it cannot understand. A system can be a brain, a machine, a cell, a business, or an economy. The key element they all share is information dependency. They take in signals from the world, make distinctions, and act upon what they find.

In my first essay, “The Distinction,” I examined the use of information from first principles. What does it mean to make a distinction, and what is required for that act to happen? In the end, understanding is built upon distinction, and effectiveness on understanding.

Seven years ago I began writing a book that touches on these topics. Part of that journey was spent examining how understanding works across domains: in biology, in music, in computation. This piece is a follow-up to “The Distinction.” It considers consciousness and its role in how systems understand.

People tend to think of computation as what a mathematical equation or a computer does. In current parlance it is simply the processing of information, be that in a cell, a computer, or a brain. It is the process by which a distinction is made.

In 1900, a mathematician named David Hilbert posed twenty-three problems to the mathematical community. These were ambitious, foundational problems, the kind Hilbert believed could anchor all of mathematics on solid, provable ground. Beneath the ambition lay an assumption: that mathematics itself was consistent, complete, and decidable.

To understand why those words matter, first let’s consider what a formal system actually is. A formal system is a set of rules and starting assumptions, called axioms, that let you prove things through logical steps. You might remember axioms from high school geometry. Euclid built all of geometry on just five of them. You can draw a straight line between any two points, all right angles are equal, and so on. Everything else in geometry follows from those five starting agreements.

Consistency means the system does not contradict itself. Two plus two equals four is consistent. Two plus two equals five is not. Completeness means that every true statement can be proven using the axioms available to the system.

Hilbert believed mathematics possessed both properties. In 1931, Kurt Gödel demonstrated something remarkable. By starting with the statement, “this is not provable,” he showed mathematically that any consistent system cannot be complete. In essence, if you could prove “this is not provable” to be true, then a false statement would be true. If you cannot prove it, then the system is incomplete. Consistency and completeness are incompatible. You cannot have both. This was not a minor technical result. It was a structural limit on what formal reasoning can do. Some truths live permanently beyond the reach of any such system. This is known as the incompleteness theorum.

In The Emperor’s New Mind, published in 1989, physicist Roger Penrose made a striking argument. Because mathematicians can see that Gödel’s unprovable statements are true, because they can somehow intuit them, human cognition must be doing something no algorithm can replicate. Whatever we are, it transcends computation.

I found this argument seductive. As a jazz musician it appealed to me. If Gödel showed that formal systems have limits, and mathematicians can see beyond those limits, then the mind must be more than a formal system.

While seductive to my creative side, it felt counterintuitive to my understanding of a neuron and a bit. Last summer I found myself talking about Penrose’s idea over a beer with a physicist friend of mine. He said something, simple yet profound, that I have thought about many times since. The feeling of knowing something is not the same as the act of knowing it.

You can have a powerful intuition that a mathematical statement is true. You can feel certain. That flash of insight, that moment when the answer arrives before the proof does: it feels like evidence that something non-computational is happening. It feels like the mind is doing something non-algorithmic.

Penrose collapsed consciousness and understanding into a single phenomenon. Yet a gut sense that something is true still requires derivation, step by step, through exactly the kind of logical procedure Gödel was describing. The sense of knowing, and the work of knowing are not the same thing.

Penrose observed something real: mathematicians do have intuitions about statements they have not yet proven. That experience is genuine. The question is whether the experience is doing computational work, or it is merely accompanying the computational work.

This brings us to the essential theme of this essay.

In 1995, the philosopher David Chalmers drew the sharpest version of a question that had been circulating for decades. He called it the hard problem of consciousness. The easy problem, explains how the brain processes information, integrates signals, directs attention, generates behavior. These phenomena are hard in the technical sense, but they are tractable. Given enough time and enough science, we will get there. The hard problem is different. It asks why any of that processing is accompanied by experience at all: seeing red, feeling pain, understanding sadness. Why does it not all just happen in the dark?

Nobody has answered this. Not really.

The reason it matters here is that the question splits the consciousness debate into two camps.

On one side: consciousness is emergent. It arises from sufficient physical complexity, from neurons firing in particular patterns, from information being processed in particular ways. Consciousness is a product of the physical. The brain generates it.

On the other side: consciousness is fundamental. Not a product of matter but a feature of reality at the deepest level, as basic as mass or energy. This is the view that Annaka Harris explores in Lights On, interviewing physicists and philosophers who take seriously the idea that experience is woven into the fabric of things, not layered on top. When I first encountered this position I found it oddly liberating. If consciousness is fundamental, you do not have to explain how it emerges from non-conscious matter. The question dissolves.

I spent a long time wrestling with this debate. In the end this is where I got to. It does not matter. At least not for the question of understanding.

Here is why.

Whether consciousness is fundamental or emergent, the functional requirements for understanding remain identical. Understanding occurs when a system possesses an alignment of the distinctions required to reach semantic agreement. A system that achieves this needs three things: a rule by which a distinction can be made, the ability to apply that rule, and the energy to execute it. The rule is what lets the system extract value from input.

Consciousness, whether fundamental or emergent, is not one of those three things.

This is not a denial that consciousness exists. It is not dismissing its importance for moral reasons, experiential reasons, reasons having to do with what it means to live a human life. It is a narrower claim. For the specific function of deriving value from information, consciousness may affect it, but is not intrinsically required.

Gödel showed us the limits of formal systems. Physicist, mathematician, and computer scientist Stephen Wolfram introduced the principle of computational irreducibility. It states that some systems cannot be shortcut, that the only way to know what they will do is to run them and find out. Wolfram spent years cataloging simple computational rules and discovered that even the most basic rule sets can produce behavior so complex that no shortcut exists. You cannot predict the outcome without executing the computation step by step. This is not a failure of our mathematics. It is a feature of the systems themselves.

What Wolfram suggests is that understanding is hard. Not because it requires consciousness, but because the problems it has to solve are irreducibly complex. The computational work is real. The energy cost is real. The distinction-making is real.

Looking back at that conversation with my friend. He was not diminishing the experience of mathematical insight. He was locating it precisely. The feeling is real. The feeling is not the work. Those are two different claims.

Consciousness may be the most interesting thing about being human. It may be what makes a life worth living, what gives weight to moral questions, what separates mere information processing from something we recognize as a mind.

Yet it is not what makes understanding work.

Distinctions — Essay 01 Derek Bronston

Anyone who has copied code from a tutorial knows the feeling. It works, you ship it, make it live, and the moment something breaks you are lost. You never had the rule. You only had the output of the rule. I have played guitar most of my life. I wrote my first line of code when I was 12. In both cases, real improvement came through integrating the underlying rules that allow for mastery over the domain. When my journey began the information was in front of me. The rule for making use of it was not.

A system that has everything it needs except the capacity to identify a difference is not broken. It is not lazy. It is incomplete. When a system is incomplete, no amount of effort from within can resolve what is missing. This essay is about two versions of that wall: a thought experiment in thermodynamics and a virus that brought the pattern into the sharpest possible focus. Each operates in a completely different domain. Each forces the same question.

Information requires, at minimum, two possible states. Not a continuous blur. Not a single fixed value. A difference: zero or one, fast or slow, self or non-self. Without at least two states, there is nothing to resolve. There is nothing with which to inform. This irreducible unit of information is called “the bit”.

Information is not abstract. It lives in a voltage, a molecular conformation, a neural firing pattern, a position on a musical staff. In 1961, physicist Rolf Landauer showed that processing information has a thermodynamic cost. Information is part of the physical world. The difference that any bit of information presents, 1 or 0, yes or no, in turn must be physically present for a system to use it.

A system that encounters the difference that information provides, but has no rule by which to resolve it, cannot act on it. The difference is right there. The system is blind to it. Use requires a rule to specify how to resolve the difference into a result. The act of applying that rule is distinction. Distinction is the mechanism by which information becomes usable.

A thermometer distinguishes hot from cold because it contains a mechanism that responds differentially to temperature. An immune cell distinguishes self from non-self because it carries molecular templates that define the boundary. A musician distinguishes consonance from dissonance because the musician has internalized the rules that define the relationship between notes. In each case the rule is what makes the act possible. Without it the difference exists, the physics is real, and the system is blind. These rules take different forms: a physical mechanism, a molecular template, an internalized grammar. What they share is a common function: each specifies how to resolve a difference into a result.

The second law of thermodynamics states that in any isolated system, disorder increases over time. Not sometimes. Always. A drop of ink spreads through water and never spontaneously reassembles. A hot cup of coffee cools to room temperature and never reheats itself. A broken egg does not un-break. The universe, left to itself, moves from order toward disorder, from concentrated to dispersed, from organized to scrambled. This tendency is called entropy.

This is not just a statement about coffee and eggs. It is a statement about what requires work. The physicist Ludwig Boltzmann showed why: there are overwhelmingly more ways for a system to be disordered than ordered. Entropy is not a mysterious tendency. It is a statistical certainty. Systems move toward higher entropy because the disordered states vastly outnumber the ordered ones.

Reducing entropy locally requires work. A refrigerator pumps heat out of its interior. A cell maintains its internal structure against decay. In each case the system consumes energy to create local order. But that energy expenditure increases disorder elsewhere by more than the local reduction. This is what the second law requires: not that order is impossible, but that it is never free.

In 1867 the Scottish physicist James Clerk Maxwell imagined a tiny demon that sits at a gate between two chambers of gas. It watches the molecules fly past. It can see each molecule clearly: how fast it is moving, which direction it is headed. When a fast molecule approaches from the left the demon opens the gate. When a slow one approaches it keeps the gate closed. Over time the fast molecules accumulate in one chamber, the slow ones in the other. One side gets hot. One side gets cold. Order increases. Entropy decreases.

The demon appears to do this for free. The gate operates on a frictionless hinge, requiring virtually no energy to open or close. Through observation alone, the demon appeared to succeed in reducing the entropy of the system, without performing any measurable work.

This apparent violation of the second law perplexed physicists. The resolution took nearly a century. In 1961, the physicist Rolf Landauer showed that the demon has to remember each observation it makes, and at some point it has to erase that memory to continue operating. Erasure is irreversible. Erasure costs energy. That hidden cost is exactly what saves the second law. The demon does not get something for nothing. It pays with memory.

The standard resolution focuses on the cost of erasure. I want to draw attention to what comes before erasure in the causal chain. The demon can only sort molecules it can distinguish. A demon with no way to tell fast molecules from slow ones cannot sort anything at all. It cannot decrease local entropy. It cannot do its job. Regardless of the energy available to it.

The demon’s rule is simple: sort by velocity. Every molecule in the chamber is moving. That movement is raw data. The rule converts it into a classification: fast or slow, relative to a threshold. That classification is what determines whether the gate opens or stays shut. Without the rule, the demon sees motion. With the rule, it sees a molecule it can sort. The information was always there. Distinguishing is what made it usable. And this is not separate from the energy cost Landauer identified. It is upstream of it: without the capacity to distinguish, there is nothing to remember. Without memory, there is nothing to erase. Without erasure, there is no thermodynamic cost. Distinguishing is where the chain begins.

In late 2019 a novel coronavirus began spreading through the human population. The immune system is one of the most sophisticated systems for distinguishing on earth. It has been distinguishing self from non-self, benign from dangerous, known from unknown, for hundreds of millions of years. It is extraordinarily good at its job.

COVID-19, however, presented the immune system with a signal it had no specific basis for distinguishing. Not because the immune system was entirely without resources. The innate immune system detected the virus through general molecular patterns: broad rules that recognize classes of threat rather than specific ones. But broad rules produce broad responses. In the worst cases, the immune system catastrophically misclassified the threat, triggering inflammatory responses that damaged the very tissue it was trying to protect. The system was not failing. It was applying the rules it had. Those rules were simply too blunt for this specific task.

The mRNA vaccine did not strengthen the immune system in any general sense. It did something far more precise. It supplied the missing rule. Here is this spike protein. Learn to distinguish it. After that single addition, the mechanism worked. An immune cell encounters the spike protein on an infected cell’s surface. It applies the template. The result: foreign. That single classification is what initiates the targeted response: antibody production, T-cell activation, memory cell formation. None of it starts until the signal is distinguished. Before the vaccine, the same protein on the same cell surface produced nothing, or worse, a misdirected inflammatory attack. After the vaccine, the same encounter produces a precise, coordinated defense. Same signal. Same machinery. What changed was the system’s capacity to distinguish.

This is what both examples show. Information is not a thing that exists in the world waiting to be picked up. It is a capacity that exists in the relationship between a signal and a system equipped to make the required distinction. The signal can be present, perfectly transmitted, sitting directly in front of you, and still be invisible to a system missing the rule that would make it resolvable.

When I was cutting and pasting code, or regurgitating lines I had learned on guitar, the information was in front of me the entire time.

That sentence used to feel like a confession of failure. I should have seen it sooner. I should have worked harder. I should have been smarter.

It does not feel that way anymore. It feels like a precise description of how every system, biological, computational, and musical, encounters the boundary of what it can currently know. Not a judgment. A diagnosis.

A diagnosis points toward a remedy: not to work harder within the system you have, but to find the rule that makes the missing distinction possible.

Distinctions: essays on information.

If this essay reached you through a forward or a share, you can subscribe here.

Subscribe now

Share