Why Science Needs Its Own Version of the Void

Science runs on two forces, and only one of them gets trained into researchers. Law means explicit reasoning, measurement, repeatability and clear reporting. It's what most people picture when they picture science. Chaos is the unmapped gap where a question hasn't yet become something a method can handle. That part rarely makes it into a syllabus. The argument here is that good research needs both, and that the field has gotten very good at teaching only one.

What the Law side demands

The Law side is familiar. Claims should survive deliberate reasoning rather than rest on a vibe. Measure what you can, prefer numbers, and write the work so another team could reproduce it. Keep a plan and stick to it, because ad hoc changes mid-experiment quietly wear down validity. Use a control arm so you can see the counterfactual, and change one thing at a time. Report in standard formats, with familiar paper structure, familiar graphs and familiar terms, so that whoever runs the meta-analysis later doesn't have to decode you.

This framework isn't arbitrary. Before Francis Bacon and the Scientific Revolution, natural philosophy was rich, but its practitioners kept circling back. They were trapped by their own preconceptions and couldn't pool findings beyond individual minds and texts. Lawfulness was the fix: shared procedures that let one person's result become everyone's starting point. It's also teachable. Procedures can go in books, they can be articulated, and they can be tested. That's why so much of science education, from elementary school through graduate training, has been built around them.

Where Chaos comes in

Chaos is the other master. The word comes from the Greek "χάος," usually rendered as "the yawning void." In myth it was the first being, a pseudo-sentient gap between earth and sky where things come into existence, both comprehensible and unspeakable. The author's sense of the word isn't defiance of order. It's the primordial state that comes before the world gets repackaged into something familiar and digestible.

Musashi, rationality and the nameless virtue

The argument leans on Eliezer Yudkowsky's Twelve Virtues of Rationality, which closes with a virtue that has no name, and on Miyamoto Musashi's The Book of Five Rings. Musashi's point is about intention. When you strike, your intent has to carry through to cutting the enemy in the same movement. A parry or a feint that doesn't finish the cut accomplishes nothing. The author maps this onto reasoning: every step must cut through to the correct answer, and your map has to carry through to reflect the territory. Insisting you followed proper form doesn't help if the answer is wrong.

The sharpest part of the essay is a parable. Suppose your idea of rationality is to believe the Great Teacher, who says the sky is green, while you look up and see blue. If you decide rationality means trusting the teacher, you lose the chance to find your mistake. Naming a principle doesn't rescue you either, whether you call it "the map that reflects the territory" or "Bayesian decision theory." You find errors by comparing your description against what it failed to capture, not against itself. Musashi's version of the same idea is that after years of practice under strict constraints, you glimpse the center, and the techniques become one technique. He called that the Way of the Void.

The gap in junior alignment work

The author's test is informal and pointed. When junior AI alignment researchers describe their work, do they engage with the void? Not the word, which would just become another piece of jargon, but the substance. Too often the answer is a glazed look, a search for inspiration, a hand-hold in the dark.

The author argues that many of these researchers do have access to the unknown but hide it, embarrassed to be lost. Some retreat into lawful but sterile work, measuring statistical correlations between benchmarks while waiting for their own Johannes Kepler to walk in and make Tycho Brahe's data add up. Others attach their confusion to a "passion" for interpretability and stare into the sand, hoping patterns will cohere. The essay's complaint isn't that these researchers are careless. It's that they're keeping the most useful part of their thinking out of sight.

Why science sits closer to madness

The essay makes an unexpected claim: science aligns with a certain kind of madness far more than engineering does. Those building apocalyptic gadgets, the author says, deserve the title "mad engineer" rather than "mad scientist." The reason isn't that good science invites madness. It's that good science demands harsh, unfiltered contact with parts of reality not yet absorbed into our worldview. That contact is uncomfortable, and the author treats the discomfort as the job itself.

What to do about it

For novices, the advice is to engage with Chaos so you get better at cutting through it. A lifetime spent staring into the dark without drawing anything into the light is probably wasted. A lifetime spent avoiding the unknown and refusing to name it is wasted for certain. If your environment punishes that kind of thinking, keep it private, but don't drop it from your notes or your own reasoning.

For senior researchers, the obligation is more explicit. Talk openly about ignorance. Build research settings where wordless, confused hunches are acceptable to chase. Don't sanitize your process into a story where the answer was obvious all along. The essay suggests textbooks should highlight the failures of the models they present and the uncertainty that remains.

The risk on both sides

The author acknowledges the failure mode. Done poorly, embracing Chaos becomes worship of the unknown, which is the opposite of pursuing truth. Mysticism in a lab coat helps nobody. The fix is to treat the void as the object of inquiry, not a substitute for method. Meanwhile, the essay notes that science is routinely flattened into laboratorism and a checklist that starts with "make a hypothesis." Restoring room for uncertainty in papers, seminars and everyday conversation is the correction the author proposes.

My read is that the test is simple enough to run this week. In your next group meeting, ask what the team doesn't understand yet, and give that question more than thirty seconds. If the answer is a shrug, the lawful machinery is running well, but the research may have stopped looking for its next question.