Steering Is Harder Than It Looks

Napoleon thought he was steering events. Events were steering him.

That’s how Tolstoy writes Napoleon in War and Peace. His theory of history: history is the sum of infinitesimal forces, millions of individual decisions, accidents, weather, morale, disease. The higher you go in the command hierarchy, the less effect you have on actual events.

I’m part of a reading group exploring what Venkatesh Rao calls World Machines — a framework for thinking about planetary-scale dynamics, the overlapping forces that shape civilization across decades and centuries — with the ambition of building something like a psychohistory for our era. The question that keeps pulling us in: if nobody steers, what does it mean to watch history? To engage with it? To intervene? How could psychohistory actually work?

For those who haven’t read Asimov: In Foundation, Hari Seldon uses “psychohistory” to predict civilizational collapse centuries ahead; a hidden Second Foundation watches and intervenes via the “Prime Radiant” when reality diverges from plan.

Making Maps

In 1884, delegates from twenty-five nations met in Washington to solve a coordination problem: where is zero? Every country had its own prime meridian. The French used Paris. The Spanish used Cadiz. The Americans had different meridians for different purposes.

Greenwich won. Not because it was natural (there is no natural zero) but because Britain had the ships. Two-thirds of the world’s commerce already navigated by British charts. The arbitrary became universal.

France held out for a generation. French maps kept Paris at zero until 1911. But holding out costs energy, and eventually you fold. The prime meridian isn’t where physics says it should be. It’s where path dependency and accumulated infrastructure say it is.

This is the simple version of how reference frames work: whoever draws the map determines how everyone navigates. If you want to shape how people think about the future, build the oracle. Be Seldon.

Many maps, no victor

But history isn’t mapmaking.

The Prime Radiant, Seldon’s device for displaying and editing psychohistorical equations, assumes a single authoritative model. One set of equations, maintained by one elite (the Second Foundation), broadcast as prophecy to everyone else. The Time Vault pronounces; the masses receive.

In reality, there are always multiple Second Foundations. Competing oracles, each trying to establish its own prime meridian. Think about who’s trying to steer right now: central banks using forward guidance to shape expectations, AI labs racing to establish the default infrastructure, platform algorithms deciding what billions see, elite scenario planners preparing executives to recognize regime changes, major philanthropies funding according to explicit theories of change. Latour calls these inscription devices: technologies that create facts by recording them. Every inscription device has blind spots built into how it works. What you can model shapes what you can see.

I’ve started calling the blind spots negative inscription: what the map doesn’t show becomes terra nullius. Futures the oracle can’t see become futures that can’t happen. A prediction market can only see what can be written down as a resolvable bet — everything that resists clean operationalization becomes invisible. A media ecosystem optimized for engagement makes slow-moving structural change invisible. An LLM trained on internet text converges toward consensus and away from strange perspectives.

The contestation never fully resolves. There’s no 1884 conference where one oracle wins and the others fold. Instead, we get overlapping, competing inscription devices, each collapsing different possibility-spaces.

But suppose one oracle did win. Could it then steer where everyone goes?

No one steers history

Here’s the harder problem: even if you draw the map, you can’t steer the territory.

Tolstoy saw this clearly. Pierre, an aristocrat who wanders onto the battlefield at Borodino hoping to witness history, sees only “ordinary confusion.” No grand strategy, no decisive turning points. Just smoke, noise, and frightened men. Pierre’s confusion is closer to the truth than any historian’s panoramic view. Kutuzov, the Russian general who defeated Napoleon, wasn’t wiser because he had better predictions than the German military theorists. His wisdom was better perception of his own ignorance.

Tolstoy shows why steering is hard: Napoleon commanded armies, but he didn’t command weather, morale, disease, the million small decisions of soldiers and civilians. Causes are too distributed. Sewell adds a second problem: events don’t just resolve tensions; they transform structures. The French Revolution didn’t just change which faction held power. It created “citizen,” “nation,” “popular sovereignty” as new ontological categories.

This is the Mule problem. In the Foundation series, the Mule is a mutant with unforeseeable powers who derails Seldon’s thousand-year Plan. But the Mule doesn’t just defeat the Plan — he reveals that the Plan was wrong about what kind of system human civilization is. Seldon’s math assumed individual deviations average out. The Mule is proof that the averaging assumption was wrong.

Every oracle carries an implicit claim: “the system I model is one where regularities persist.” But Sewell says: sometimes the rules themselves change. And Tolstoy says: even when the rules don’t change, the causes are so distributed that no one steers.

Where Oracles Work

Horizontal: how stable are your categories? Vertical: do tensions resolve or persist?

                         RESOLVES
                            │
     PREDICTION             │         STANDARDS WARS
     (Seldon, Tetlock)      │         (VHS vs Betamax)
                            │
  STABLE ───────────────────┼─────────────────── CONTESTED
                            │
     SCENARIO PLANNING      │         EVENTFUL TEMPORALITY
     (Shell)                │         (Sewell, Tolstoy)
                            │
                         PERSISTS

Seldon lives in the upper-left: stable categories, tensions that resolve. That’s where psychohistory works — if it works anywhere. Sewell and Tolstoy live in the lower-right: contested categories, unbounded horizons, no clear resolution. That’s where world machine theory lives.

The interesting questions live in the contested half: what do you do when your categories are in flux and tensions persist?

What Second Foundations actually do

So what’s left? If you can’t be Seldon, if the map doesn’t control the territory and the territory keeps reorganizing itself, what are Second Foundations actually doing?

I think they’re not predicting. They’re acting.

The oracle doesn’t forecast the future; it tries to initiate one. The declaration is the intervention. When Seldon’s Time Vault pronounces, it’s not describing what will happen — it’s performing a speech act that reshapes what the listeners believe and therefore do. Reality is socially constructed; the prophecy is a move in the construction.

But action is fragile in ways prediction isn’t. You can’t control what you start. The Second Foundation keeps having to intervene because reality keeps diverging from the Plan — because actions produce consequences no one can control.

There’s a version of this that’s just manipulation: hidden elites steering the masses through carefully timed pronouncements. That’s what the Second Foundation actually is in the books. But there’s another version that’s more interesting.

What if the oracle isn’t a broadcast device but a coordination device?

How Oracles Work

Horizontal: who authors the oracle? Vertical: how surprising is its output?

                        SURPRISING
                            │
     SECOND FOUNDATION /    │         MARKETS
     DELPHI                 │
                            │
  ELITE ────────────────────┼────────────────── DISTRIBUTED
                            │
     AUDIT                  │         FOLKLORE
                            │
                        FAMILIAR

Most oracles live on the left side of this diagram. Elite-authored, whether they pretend to be representational (Delphi, punditry) or openly performative (Second Foundation, scenario planning). The interesting quadrant is upper-right: distributed and performative. Prediction markets, sure — but also social movements, open-source projects, any collective process that constructs reality through participation rather than pronouncement.

Leverage points

Bruce Sterling called our time “dark euphoria” — the vertiginous feeling when the old maps stop working but you can’t believe the possibilities. The old broadcast model is collapsing. Social media fragments into competing inscription devices. No shared map (and no real hope for one). The contestation is live.

Seldon’s fantasy is total steering — one oracle, one plan. The reality is messier: distributed causes, competing inscription devices, events that transform the rules. World machine theory is one attempt to build a psychohistory that starts from that mess — not a single Prime Radiant, but a framework for thinking about civilizational trajectories when no shared map holds.