Is Physics Really in Charge? Ellis vs. Carroll

There is a particular flavor of argument that sounds, at first pass, like philosophy—the kind of thing that reasonable people wave away as semantics—but which quietly underpins some of the most consequential disagreements in contemporary science. The question of whether high-level phenomena genuinely cause things, or merely describe things that were always going to be determined by lower-level physics, is one of them.

In a recent clip from his Theory of Everything podcast, Curt Jaimungal puts this question directly to cosmologist and mathematician George Ellis, framing it as a defense of physicist Sean Carroll's reductionist position. What follows is a concentrated, surprisingly concrete exchange about one of the harder problems in the philosophy of science: downward causation.

The setup is worth understanding clearly, because the stakes are easy to miss.

The Reductionist's Strongest Case

Carroll's position—which Jaimungal attempts to articulate in the conversation—is a version of what philosophers call eliminative reductionism, or more charitably, explanatory reductionism. The argument runs roughly as follows: when we say a thermostat "causes" a heater to switch on, or a Python script "causes" electrons to rearrange themselves in a semiconductor, we are using convenient shorthand. If you pushed hard enough on any of those descriptions, you would bottom out at microphysics. The thermostat is atoms. The Python is electrons. The decision your brain made this morning is neurons firing in patterns governed by electrochemical gradients. There is no additional causal work being done by the higher-level description; it's just a more legible map of the same territory.

This is a genuinely powerful position. It has driven enormous scientific progress. Reducing the complexity of life to chemistry, and chemistry to physics, produced germ theory, molecular biology, and the entire pharmaceutical industry. Carroll is not making an amateur's mistake here; he's defending a methodological framework with an extraordinary track record.

Jaimungal frames it precisely: "all of what was just said would just still be accounted for by the microphysics entailing the macrophysics."

Ellis's response is to reject the framing rather than fight on its terms.

What "Only Physics" Actually Means

Ellis's core move is to distinguish between enabling and deciding. Physics, he argues, enables outcomes. It does not select them. "If I tell you Maxwell's equations," Ellis says in the exchange, "here's Maxwell's equations. So what does that do? Doesn't do anything. I give you Newton's laws of motion. Tell me what Newton's laws of motion will cause. Doesn't do anything. It only does something in a context."

This is not a mystical claim. It's a claim about the structure of causal explanation. A law of physics describes what is permissible; it does not specify what happens. The context—the arrangement, the constraints, the goals encoded in the system—is what determines which of the physically permitted outcomes actually occurs.

The thermostat is his first example, and it's a good one precisely because it's so stripped down. A thermostat has a sensor, a target temperature, a comparator, and an effector. When the measured temperature diverges from the target, a signal closes a circuit, a heater activates, molecules start moving faster. Ellis's point: the physics did not decide that 20 degrees is wrong and 40 degrees is right. You decided that, by setting the dial. The physics faithfully executed the instruction. "The physics isn't deciding anything," he says. "The physics is the servant, not the master."

The Carroll camp would respond that "you" setting the dial is itself a physical event—neurons, muscles, mechanical action—and that the target temperature is encoded in the physical structure of the thermostat. Everything folds back into physics. Ellis would say: yes, but the content of the target—why 40 and not 20—is not determined by physics. It comes from somewhere else in the causal hierarchy.

This is where the argument gets genuinely interesting, and where neither side fully dispenses with the other.

Constraints, Hierarchies, and the Three Things Higher Levels Do

Ellis extends his case through what is, frankly, an underappreciated concept in mainstream science communication: the role of constraints in physical law. A pendulum's bob swings at a frequency determined by the length of the pendulum arm. That length is a macrolevel constraint. Change it, and you change the behavior of every particle in the bob. The constraint is causally real—it does actual work—and it is set at a higher level of organization than the particles it governs.

From here, Ellis scales up to biological systems. Every truly complex system, he argues, is modular and hierarchical—a claim that software engineers and cell biologists would both recognize instantly. You build complexity by decomposing it, solving the simple parts, and reintegrating. The modular structure is not incidental; it is what makes complexity possible at all. And crucially, those modules interact vertically: higher levels constrain, select, and modify what lower levels do.

The biological examples he reaches for are vivid. Gene regulatory networks respond to high-level organismal conditions and instruct molecular machinery to produce specific proteins. During embryonic development in fruit flies, positional signals cascade across a sheet of initially identical cells, switching specific genetic circuits on or off at specific locations. The result is a body plan—backbone, limbs, nervous system—arising from what is, at the molecular level, a series of context-dependent instructions. Cells that could have become anything (Ellis uses the term pluripotent) become something specific because the higher-level system told them to.

He adds a detail that I find clarifying in its concreteness: apoptosis, programmed cell death. Human fingers are separate because the cells between them were systematically destroyed during fetal development. The organism, at a systems level, executed a deletion. The physics of cell death is well understood. What is not explained by the physics alone is why those particular cells were targeted. That required higher-level information—developmental signals, positional encoding—that does not exist at the cellular level.

"What higher levels do in any complicated system is they create, modify, or destroy lower level elements," Ellis says. "And that's the core of a huge amount of biology."

The Organizational Analogy and Its Limits

Ellis rounds out with a corporate analogy—Boeing as a modular hierarchical structure—where hiring, training, and termination processes select, modify, and remove employees in ways that parallel biological creation, modification, and destruction. It's a clarifying illustration, though it does introduce a complication Ellis doesn't address directly: organizations are themselves made of humans who have intentions, goals, and agency in ways that thermostats and pendulums do not. Whether the analogy illuminates the biological case or smuggles in additional assumptions is a question worth sitting with.

This points to a genuine tension in the downward causation literature. The concept spans an enormous range of phenomena—from simple mechanical constraints (pendulums) to feedback control systems (thermostats) to developmental biology to conscious human decision-making—and it's not obvious that the same causal logic applies uniformly across that range. Philosophers of science including Jaegwon Kim have raised serious objections to strong versions of downward causation, arguing that if physical events are causally closed, higher-level causes are either reducible to lower-level causes or are epiphenomenal. Ellis is aware of this literature; he has written extensively on it. But the clip doesn't give him room to address it, and Jaimungal's channeling of Carroll, however good-faith, is not the same as Carroll actually in the room.

What This Debate Is Actually About

There is a version of this disagreement that is purely terminological—a dispute about which level of description we want to call "causal." If that were all it was, the stakes would be low.

But Ellis's position, taken seriously, has consequences. If higher-level organization genuinely selects among physically permitted outcomes—if context is causally real—then explaining a system requires understanding its organizational structure, not just its substrate. That has implications for how we do biology, how we think about neuroscience and consciousness, and how we approach the design of complex technologies. Collapsing everything to microphysics isn't wrong, exactly. It may just be systematically incomplete.

Carroll's reductionism is not the villain in this story. It is a precise and productive tool. The question Ellis is pressing is whether it is the only tool, or whether there is causal work happening at higher levels of organization that it structurally cannot see.

That question does not have a settled answer. What it has is better and worse ways of being asked—and a thermostat, of all things, turns out to be a surprisingly good place to start asking it.

Priya Sharma is Buzzrag's science and health correspondent. She was previously a neuroscience PhD student and still reads methods sections for fun.