Hard to Vary

You were probably taught that science works by evidence: gather observations, and the true theory emerges from them. It doesn't work, and Deutsch's opening move is to show you why with the seasons.

Both explanations above fit every observation available in ancient Greece. Every one. The myth predicts cold winters, warm summers, an annual cycle, and it does so with total accuracy for the entire Mediterranean. Evidence could not separate them — because a bad explanation can be adjusted to fit any evidence at all. That is what makes it bad.

The thing that separates them is structural, and you can check it from your armchair before a single measurement arrives. A good explanation is hard to vary while still accounting for what it accounts for. Every component is load-bearing. Pull one out and the whole thing stops working.

Now run that test over the things you actually believe — about markets, about people, about why last year went the way it did. Most of them are Persephone. They fit, they feel explanatory, and every part of them could be swapped for something else without losing a thing.

The story where you watch the world carefully and the truth accumulates is called induction, and Deutsch's position — Karl Popper's before him — is that it has never once happened. Not in any laboratory, not in any human head. It is not a slow method. It is not a method.

You cannot start from observations because there is no such thing as an observation without a theory already in it. Something has to tell you which of the infinite facts in front of you is worth writing down. That something is a conjecture you already had.

What actually happens is this:

There is no arrow here from "observations." Experiment enters only at step 03, and only as a way of killing things. It never generates the conjecture. This is why a room full of data produces nothing without someone in it willing to guess.

Two consequences follow, and both are uncomfortable.

Every one of your beliefs is a conjecture. There is no foundation underneath, no source of certainty — not the senses, not authority, not the data. This is fallibilism, and it is not the same as saying all views are equal. Some conjectures have survived enormous criticism and some have survived none, and telling the difference is the whole job.

Errors are not the failure mode. Errors are the medium. A system that expects to be right is fragile. A system built to find and correct its own errors compounds. The question worth asking about any process you run — a method, a business, a book of trades — is never is it correct. It is how fast does it find out when it isn't.

Here is the strangest fact in the book. When you find a genuinely good explanation of a small thing, it turns out to explain enormous numbers of other things you were not thinking about and could not have anticipated.

This is reach, and it is unearned. Nobody put it in. It arrives because a good explanation is about what is really out there, and what is really out there does not stop at the edges of your problem.

Reach is the strongest signal available to you that you have touched something real rather than curve-fitted something local. A model that only ever explains the cases it was tuned on has told you about your tuning. A model that starts answering questions you never posed to it has told you about the world.

That wall is the shape of the whole pattern. Systems built for a small, parochial purpose keep hitting limits — and then one of them, without anyone intending it, crosses a threshold and becomes unlimited.

It has happened at least four times. An alphabet: invented to write one language, and it turns out to write every sound a human can make, including in languages that did not exist yet. Movable type. Babbage's engine, designed to grind out navigation tables, whose design was universal for computation and its author had no way to know. And, four billion years ago, a molecular code that started copying and could in principle encode any organism whatsoever.

The jump is never gradual, and it is never anticipated by the people making it. They were solving something small. They over-built by one component, and the system fell through into infinity.

And there is one more universal system in the list, which is the reason any of this matters to you. People. Not a species that is good at a wide range of things — a species that is a universal explainer, capable in principle of understanding anything that can be understood and building anything the laws of physics permit. There is no category of phenomenon that is closed to us by our design. That is a claim about physics, not a compliment.

Deutsch's most disorienting chapter takes the phrase "Spaceship Earth" and inverts it completely.

The image is of a planet that supports us, a delicate biosphere providing what we need, which we are damaging. Every part of that is backwards. Take a modern human and put them in the Great Rift Valley — the place our species is best adapted to, where we evolved, the ancestral home — and they die. Quickly. That environment supported a few tens of thousands of hunter-gatherers living at the edge of extinction, and it did nothing to support them; they survived it using knowledge, and barely.

Deutsch's own Oxfordshire would kill him unaided within days. It sustains him because of clothing, agriculture, buildings, refrigeration, antibiotics, sewerage — knowledge, all of it, none of it native. The biosphere does not support human life. Human life supports itself, on hostile ground, using explanations.

Which makes "sustainability" a category error when it is offered as the goal. No way of life has ever been sustainable, because problems keep arriving — an ice age, a pathogen, a rock from orbit — and a static way of life has, by construction, no means of generating the solution. Every static society in history eventually met a problem it could not solve and ended. The only thing that has ever actually sustained anyone is the continued rapid creation of new knowledge, which is inherently unpredictable and cannot be planned.

Stephen Hawking once described humans as "chemical scum on a moderate-sized planet, orbiting a very average star." Deutsch's answer is the book's spine: exactly backwards. In a universe that is almost entirely inert, the rarest and most consequential thing that exists is a system that generates explanations — because there is no bound on what it can eventually know or do. Astrophysically we are a rounding error. In the only ranking that tracks causal significance, we are the interesting part.

If knowledge grows by criticism, the obvious question is why it almost never did. For most of human history, nothing changed for centuries at a stretch. That stasis was not an absence of clever people. It was an achievement — actively maintained, and very hard to build.

Deutsch's account is about which ideas survive. An idea can be transmitted down generations in two entirely different ways. It can be useful and true, so people keep it because it works. Or it can be structured to disable the recipient's ability to evaluate it — and that second kind spreads without needing to be true at all.

The Enlightenment, on this reading, was not the discovery of a set of facts. It was a rebellion against the authority of received knowledge, and its content was almost nothing: we do not know, but we can find out. It is roughly four centuries old, it has failed and been extinguished before, and it is not guaranteed to continue.

The test for an anti-rational idea is worth carrying with you, because they are not rare and they do not announce themselves. Ask: does this idea make me better at judging ideas, or worse? Anything that arrives with a reason you shouldn't examine it — that questioning it is disloyal, naive, dangerous, or beneath you — has told you its category.

Deutsch means something technical by optimism and it is worth getting precise, because the word has been ruined by people using it to mean cheerfulness.

The Principle of Optimism: all evils are caused by insufficient knowledge.

That is a claim about causation. Not that things will go well — they frequently won't. Not that we will solve any particular problem in time — we may not, and failure is real and permanent for the people it lands on. The claim is narrower and stranger: every failure, examined closely enough, turns out to be a failure to have known something in time. There is no category of misfortune that is simply built into reality and immune to being understood.

Which is why optimism in this sense forbids prophecy in both directions. You cannot predict the future growth of knowledge, because to predict the content of a future discovery is to have made it. So confident forecasts of doom and confident forecasts of triumph are the same error wearing different clothes: both claim to know what will be known.

What survives is a rule for the present. When something fails, the question is not who was at fault or was this fated. It is: what would we have needed to know? Every other framing ends the inquiry. That one continues it.

Terms of art