Beyond the equations, physics forces three questions: what is reality made of, where does mind sit within it, and on what grounds can we know it? These four books each offer one answer.
2026 · Book Recommendations · Issue 18
Physics is often treated as a calculating tool, but every time it reaches its limit, the question it asks is a worldview question. Deutsch argues that reality is comprehensible — not through the accumulation of observations, but through good explanations that are hard to vary, whose reach is in principle unbounded. Rovelli says the world's substrate is not made of "things" but of a web of events and relations — even space is granular, and time does not exist at the most basic level. Feynman asks what a "law" really is, and anchors it back to the one and only judge: guess — compute — compare. Penrose lines up mathematics, physics, and mind as three worlds, pressing on why the physical world obeys a Platonic mathematical layer, and why understanding cannot be exhausted by any algorithm.
| Book | Author | Year | The One Thing This Book Nails |
|---|---|---|---|
| The Beginning of Infinity | David Deutsch | 2011 | Knowledge has no fixed boundary — what drives it isn't observation but "hard-to-vary" good explanations, whose reach is in principle infinite |
| Reality Is Not What It Seems | Carlo Rovelli | 2014 | The world's substrate isn't objects but events and relations; space is granular, and time does not exist at the most basic level |
| The Character of Physical Law | Richard Feynman | 1965 | What a "law" is and why we trust it — the answer is the language of mathematics, plus guess–compute–compare as the only judge |
| The Emperor's New Mind | Roger Penrose | 1989 | Mathematics, physics, and mind are three real worlds; understanding is not an algorithm, and consciousness may need undiscovered new physics |
We usually assume the foundation of science is observation: watch enough, and the regularities surface on their own. Deutsch judges this "empiricism" to be fundamentally wrong. Observations never speak for themselves — the same data can house countless mutually contradictory explanations. What actually drives knowledge is explanation: a conjecture about the unseen mechanism behind the scenes. And what separates good from bad is not how well it fits the data, but a surprisingly simple test — being hard to vary.
"Seasons come because a winter goddess grieves" can explain anything, and therefore explains nothing — its details can be swapped out at will without affecting the conclusion. "Seasons come from the tilt of the axis" has every part bolted down: move one piece and the whole thing collapses, and that is precisely why it has force. The parts of a good explanation interlock and cannot be shuffled around freely — this is what divides it from myth, conspiracy theory, and after-the-fact storytelling. What explains everything explains nothing.
From this Deutsch draws a bold worldview: the reach of a good explanation has, in principle, no ceiling. An explanation built to solve the problem in front of you often governs far beyond it — the laws Newton wrote for the planets equally govern cannonballs and tides. So the growth of knowledge has no natural ceiling; he calls this "the beginning of infinity." It comes with a principle of optimism: all evils stem from a lack of knowledge, and "anything that is not forbidden by the laws of nature is achievable, given the right knowledge."
This is not blind optimism. Deutsch's formula is: "Problems are inevitable, but problems are soluble." Each solution uncovers new, deeper problems — progress is not converging on some endpoint but moving from one problem to a better one. This puts humans in a singular position: as beings that carry and create explanations, we are not insignificant bystanders but key nodes wherever knowledge can reach.
Deutsch treats "good explanation" as a near-universal master key, yet is vague about how to judge "hard to vary" precisely — in edge cases it looks more like after-the-fact taste than an operable standard. The book is enormously wide-ranging but also enormously self-assured; its claims about the quantum multiverse and about Popper strike many peers as overreaching.
Deutsch's "hard to vary" is a sharp blade for incident post-mortems. After an outage, the root cause a team most readily accepts is usually the kind that explains everything — "too much load," "an occasional blip," "that machine is flaky": it fits any outage equally, and so explains nothing. To try next week: when writing the post-mortem, run each candidate root cause through a "hard to vary" test — if this explanation holds, what other phenomena must it predict, and what must it rule out? The more specific and the more falsifiable the predictions, the harder the explanation. The one that collapses if you change a single detail is the real cause; the one that holds no matter how you change it, delete.
Rovelli is one of the founders of loop quantum gravity, and this book is an intellectual history running from Democritus's atoms all the way to the frontier of quantum gravity. It has a single throughline: our intuition about "what reality is made of" is overturned by physics again and again. At every step, the world looks less like "a box stuffed with things."
The first blow comes from relations. In quantum mechanics, a particle's properties are not something it possesses on its own, but something that shows up only at the moment it interacts with something else — this is Rovelli's own "relational interpretation." An electron is not a tiny ball that is always somewhere with a definite velocity; it is a string of interaction events, and between events it has no definite "look." The world is therefore not a collection of objects but a collection of events.
The second blow comes from space itself. Loop quantum gravity says space is not a smooth, continuous, infinitely divisible stage; it has smallest grains — at the Planck scale, space is a discrete "spin network," and cutting finer has no meaning. Space is not a container in which things sit; space itself is a thing, a web woven from the grains of the gravitational field.
The third blow is the most counterintuitive: in the most basic equations, the variable of time disappears. The passage of time we experience is, Rovelli argues, a thermodynamic and perspectival effect — arising from our blurred, macroscopic observation of the world, not from a feature of its substrate. Put the three blows together and what remains is: no static things, no continuous space, no flowing time — only a web of relations woven from interacting events — and we are inside that web, not outside it.
The book bets its whole worldview on loop quantum gravity, which still has no experimental evidence; whether it or rivals such as string theory is right is far from settled. Narrating it as "the truth of reality" is more Rovelli's faith than established fact. The poetic prose occasionally lets beauty paper over leaps in the argument.
Rovelli's "the world is events, not objects" is strikingly isomorphic with the Buddhist ideas of dependent origination and non-self: the "self," an "emotion," a "problem" — things we treat as fixed objects — are all processes that ceaselessly arise and pass. To try next week: pick something you are gripping tightly — an identity label, a recurring anxiety — and during one sitting, turn it from a noun into a verb, asking "in which interactions, at each moment, is it being regenerated?" When "I am anxious" is seen as "anxiety is, right now, arising from these conditions," the gripping hand loosens an inch — this is the first-person version of Rovelli's relationality.
These are Feynman's seven 1964 lectures at Cornell, and the subject is not any specific law but the very fact of "law" itself. Starting from gravitation, he hammers again and again on a fact at once fascinating and unsettling: deep down, nature obeys mathematical regularities — astonishingly simple ones, far simpler than the task at hand requires. A formula written for the planets governs, with equal precision, the tides, the galaxies, and the falling apple — Feynman captures this unity as "nature uses only the longest threads to weave her patterns."
He gives special attention to the relationship between mathematics and physics. Why must nature speak in math? He does not pretend to an answer; he only notes that you cannot bypass mathematics and "translate it into plain words" to truly understand physics — math is not an ornament on physics, it is the only language in which physics can speak clearly about itself. This is a humble worldview: the world can be understood, but understanding it costs you the price of learning a foreign tongue.
On quantum theory he offers his famous confession: "I think I can safely say that nobody understands quantum mechanics." This is not for effect. His point is that at the most basic level, nature's behavior (probability, uncertainty) violates all everyday intuition, and what we can do is not to picture it in familiar images but to accept the equations, be quiet, and compute — and it comes out right every time. The world is under no obligation to let you "make sense" of it.
Finally he reveals the engine of all of science, simple to the point of brutality: guess — compute the consequences — compare with experiment. How beautiful the guess, how clever its author — none of it matters; if it disagrees with experiment, it is wrong. That single sentence pins the whole of human intellectual vanity to the floor. Feynman's worldview stands on it: nature is the ultimate, and the only, judge, and the most beautiful theory is merely a deposition awaiting its verdict.
A 1964 lecture series: most of the advances in quantum gravity, chaos, complex systems, and cosmology came after it, and a few examples (such as parity) have since been supplemented by time. More importantly, Feynman's near-faith that "nature obeys only mathematics" shrugs at the "why" itself — and that very candor is also a kind of evasion.
Feynman's "guess–compute–compare" is the discipline an AI super-individual should most internalize. With AI you can produce beautiful guesses — plans, architectures, judgments — ten times faster than before; but producing fast means self-deceiving fast. To try next week: take a work belief you currently hold most firmly ("this is the most efficient way to use AI," "this direction is worth betting on") and force yourself to write down "what kind of result would prove me wrong" — then design the smallest experiment, runnable within a week, to crash into it. Can't crack it = the belief hardens; crashed it = months saved. However clever and beautiful, the guess must pass through experiment.
Penrose wrote this book to refute "strong AI" — the view that mind is merely an algorithm running on the brain's hardware and could in principle be reproduced by a computer. His counterattack spans mathematics, physics, computation theory, and neuroscience, but its base is a picture of three worlds: the Platonic mathematical world, the physical world, and the mental world, each pair connected, and each connection a profound mystery.
The first world is the most counter to contemporary intuition: Penrose is a committed mathematical Platonist. Mathematical truths are not human-invented conventions but are discovered — an objective reality existing independently of any mind. He uses the Mandelbrot set as an analogy: its infinitely complex structure was designed by no one; you can only explore it and be astonished by it. So the first question of worldview surfaces: why does the physical world obey this Platonic mathematical layer so precisely?
The second step deploys Gödel's incompleteness theorem. In any sufficiently strong formal system there exist propositions the system itself cannot prove, yet which human mathematicians can see to be true. From this Penrose asserts: human mathematical understanding cannot be any algorithm — because an algorithm is equivalent to a formal system and cannot escape Gödel's cage, while understanding does escape. Understanding is not computation: this is the book's most contested, and sharpest, blade.
Then where does understanding come from? Here Penrose makes an astonishing bet: existing physics is not enough, and consciousness is rooted in an undiscovered new physics — at the meeting point of quantum mechanics' "wave-function collapse" and gravity (his "objective reduction," OR). Most physicists do not accept this specific claim. But Penrose's real contribution is not the answer; it is his refusal to stuff any one of mind, mathematics, and physics into another and call it done, forcing us to face the still-unhealed gaps between the three.
Penrose's two steps — the Gödel argument and the new-physics argument — have both drawn fierce rebuttal from logicians (e.g. Feferman) and the AI field: whether Gödel's theorem can really yield "the human brain is non-algorithmic" is far from agreed, and OR theory has no experimental support to date. The popular-physics portions are brilliant, but the core argument reads more like a genius's declaration of faith.
Penrose's "understanding is not computation," right or wrong, hands the AI super-individual a sharp division-of-labor question. An LLM can produce extremely beautiful "output," but Penrose forces you to distinguish output from understanding: for which tasks do you want only the result (delegable wholesale to AI), and for which is "seeing it yourself" the very point, so that outsourcing it cripples you? To try next week: list the three things you are tempted to hand to AI entirely, and ask of each "do I want its product, or my own understanding?" For anything in the latter category — a key architectural decision, an argument under your own name, a concept you must teach your child — demote AI to a sparring partner; understanding has to happen inside your own head.
Test: replace the explanation's key parts one by one — does the conclusion collapse with them? Collapses = it interlocks tightly, a good explanation; doesn't budge = it explains everything, hence nothing, and you've only gained the illusion of "already understanding." Then push once more: does it predict a phenomenon not yet seen and possibly false? Only if you can name one is it hard.
This is a touchstone of worldview maturity. Feynman and Rovelli share one stance: when reliable evidence overturns a deep intuition, it is the intuition that yields. Ask yourself: when did you last genuinely abandon a comfortable old belief because of new evidence? If you can't recall, you may be using intuition to cap the world — and the world never heeds that line.
A valid distinction lands on concrete action, not attitude. One criterion: if three years from now you have no memory of how this was arrived at, is anything lost? No → you want the product; hand it to AI. Yes → understanding is the asset itself, and outsourcing it is swapping muscle for a prosthetic. Penrose right or wrong, this line decides whether you become "someone who uses AI" or "someone AI hollows out."