The eternal complement

Authors’ note: This is the first essay in our series on the next economy, part of a new platform to host independent voices exploring an AGI future. It reflects our views, not those of OpenAI or our colleagues.

Humanity has seen almost to the beginning of time.

Our equations tell a mostly coherent account of the universe from its first moments. With our telescopes, we have looked to the reaches of the cosmos and glimpsed its early history. We know about the birth of stars and the formation of galaxies.

Yet no human being has ever traveled beyond the Moon.

The explanations we seek span the universe, while our bodies have barely left home. Our minds venture way beyond the reach of our hands.

That mismatch could mean two very different things.

One possibility is that we needn’t go very far, that the deepest truths might be fathomed from a single planet. Armed with enlightened reasoning and efficient instruments, we reveal the universe without traversing it. In this story, our civilization advances through depth instead of width. It would help explain why we haven’t seen other intelligent life: we might live in a galaxy full of brilliant minds that never needed to expand and become noticeable to us.

The second possibility is more intimidating. Our minds may have simply outrun our execution capacity. Each next step takes more time and more resources. We have no shortage of dreams but we lack the manpower to realize them.

Galileo widened our sight with two lenses and a tube. To widen it again today, humanity built the James Webb Space Telescope, a ten-billion-dollar observatory folded inside a rocket and sent a million miles away. Its eighteen enormous mirror segments are engineered to a fifty nanometer precision. The observatory was born from three hundred organizations across fourteen countries. Galileo’s contribution called on a few dozen hands. Webb needed an army, and a global economy behind it.

It didn’t just take brighter minds to see farther. It took a larger bureaucracy.

We question more than we can investigate. We design more experiments than we can run. We dream up creative possibilities faster than we can realize them. And across many fields: the further we advance, the more complicated it becomes to build what we imagine.

Nick Bloom and his coauthors⁠(opens in a new window) studied research productivity across the American economy. Sustaining Moore’s law now requires more than eighteen times as many researchers as it did in the early 1970s. Economy-wide effective research effort rose twenty-three-fold from the 1930s, while measured research productivity fell by a factor of forty-one.

These statistics hint at a rising cost to further progress. The economy still advances because a vast increase in research enterprise has compensated for the declining yield of each unit.

That history should make us think differently about genius: not to be considered in a vacuum, but as one input to a production process. A brilliant hypothesis still needs evidence, instruments, and the means to carry it out. An idea isn’t yet progress without the machinery and the support staff to actualize it.

Economists call two inputs complements when more of one raises the value of the other. Frontier intelligence and the capacity to realize its ideas are complements in this sense. A better telescope makes a good astronomical question more valuable. A better question makes the telescope more valuable.

Some complements are physical. A theory waits on a particle accelerator; a technological design waits for the energy and machinery to build it.

Other complements are harder to see with your eyes. An idea, on its own, is a fragile thing. It needs institutions of all kinds. Laws, bureaucracy, funding mechanisms, supply chains, and much else come together to execute on the idea.

A great idea must survive a long chain of correct local actions across these institutions. Call this institutional intelligence: the uncelebrated intelligence of execution. Genius designs the monument; institutions lay the stone. One is brilliant, the other seems boring; both are absolutely crucial.

We sometimes imagine superintelligence as a billion Einsteins. But a civilization of a billion Einsteins still needs most of them to mine the quarries and manage the accounting. Frontier innovation can only happen when the normal world works seamlessly around it. Progress calls on everyone—from the welder of a screw that goes in the telescope, to the insurer for a grade school that taught a nurse who treated the astronomer. Progress is built on a humming economy we often ignore in the tales of our greatest feats.

Like human intelligence, AI will be channeled toward both brilliant insight and institutional competence. The question is where it will spend the preponderance of its time.

AI is already making execution less scarce. It writes the code, searches an unfamiliar literature, and turns a sketch into a working prototype. Ideas that once required a whole organization can increasingly be pursued by one ambitious person.

Most aspiring filmmakers are never given two hundred million dollars to shoot their shot. Most game designers spend their entire careers implementing the creative visions of others. We try to promote the most gifted people but we inevitably miss out on talent. Viewed in this light, the intelligence age begins by complementing human ingenuity, giving more of us the support staff to build ideas of our own.

This could yield a profusion of idiosyncratic projects, with more creative work produced outside the institutions that previously served as gatekeepers to doing work in those fields. The scarce input moves from execution towards taste—the ability to decide what is worth making, which question is worth asking, and which of a thousand plausible directions deserves to be pursued.

But this may not be the end of the story. As AI grows capable of arriving at brilliant new insights on its own—which we might already be seeing in domains such as math—it supplies agendas as well as labor. Rather than implementing one human research program, it designs thousands of its own. AI geniuses explode the number of film concepts worth developing and hypotheses worth testing. Ideas abound faster than supporting infrastructure can absorb them.

Today’s AI offers a long-awaited reprieve, where our good ideas finally get their due. Tomorrow there may be so many good ideas that we become more execution-starved than ever before.

We ultimately do not know whether the complements needed to make productive use of new insights will grow, shrink, or remain stable. How much intellectual progress can one make by thinking alone? How much bureaucracy does brilliance need?

In our view, the answer depends on how far thought can travel before it must make fresh contact with reality.

We describe two potential pathways. The first is a civilization of depth, in which superintelligence surmounts the need for more physical capital and bureaucratic orchestration. The second is a civilization of width, in which the complexity of nature surpasses the ability of any intelligence—human or machine—to make progress without large and increasingly elaborate experiments in the real world.

Human history hints at a rising need for complementary capacity so far. Galileo’s telescope fit in his hands; Webb required a civilization. A small Bell Labs team assembled the first transistor on a workbench; frontier chip advancements now come from an intricate global supply chain. Further reach keeps recruiting more of civilization behind it.

But that history need not be destiny. Smart minds can invent more efficient complements. If energy bottlenecks a new particle accelerator, brilliance can devise a leaner accelerator or a better way to harvest energy.

Genius can also substitute for some of its complements. Einstein was able to think his way to relativity from very little empirical data. We would need a lot more. Some people simply remember all their engagements—most of us rely on the bureaucracy of Google Calendar.

And genius can maximize taste, choosing the best hypotheses to test and avoiding wasteful failures. If each additional genius increases the quality of the plan, even if ever-larger bureaucracies are needed, we would still invest heavily into genius precisely because execution is so expensive to squander on the wrong thing.

Neither pathway will perfectly describe our actual future, but they can be thought of as two directional possibilities for our civilization.

Intelligence can advance knowledge in three ways. It can reason from principles. It can draw new insights from existing evidence. Or it can gather new evidence by observing and intervening in the world.

The first mode can travel far through thought alone. Math is the purest example. At the many crossroads in a proof or project, a powerful mind can consistently make the right call. It replaces years of floundering search with a short, more directed chain of reasoning.

The second mode is also a greenfield for fast progress. Humanity has collected far more evidence than it has understood. The study of our data—astronomical imagery, microscope slides, online comment sections—has been limited by the scarcity of human intelligence. Scientific archives contain observations recorded to answer one question that may hold answers to many others. Abundant genius can discover not by collecting anything new, but by seeing what was already there. It might find a new universe without gathering a new photon.

The third mode isn’t so easy because it requires us to interact with the physical world. Physical processes take time, and often little can be done to speed them up. Chemicals must react. Organisms must grow. Machines must be constructed. Spacecraft must cross actual distance, subject to cosmic speed limits. Even a perfect mind can’t observe the result of an experiment that has not occurred.

But it can try. A civilization of depth wouldn’t escape the need to consult reality. It would become radically economical in doing so. Better reasoning would identify the few experiments that truly matter. Simulations would resolve most and locate the precise uncertainty that reality must settle. Superintelligence can maximize efficiency, designing around its scarce resources to get the most out of them.

Picture an astrophysics lab with a set of hypotheses on dark matter. These are first shunted through hyperrealistic computer modeling. Each theory is checked for coherence with all existing data. Only after all this are a few targeted new snapshots of the sky taken to test the most promising theories. The civilization still needs the complements to genius, but each unit yields far more value.

Our path to a civilization of depth might work like a jigsaw puzzle. Progress is easy at the start, easy at the end, and hardest in the middle. At first there are obvious edges and easy matches. Then these run out, while too little of the picture is assembled to guide the rest. Near the end, the remaining gaps almost tell you what belongs in them. A nearly complete science might find its final discoveries easier than the ones we struggle toward today.

This is a staple of science. Mendeleev could describe undiscovered elements because the rest of the periodic table constrained what was missing. The Standard Model gave physicists reason to expect the Higgs boson decades before we observed it. Our experiments were targeted to look for what we expected to see—what was coherent with existing theory and data.

Superintelligence might achieve a civilization of depth by understanding the patterns of the world so well that it knows precisely what new data it needs. It would drive progress through deeper thought, not relying on the slow march of physical expansion.

If those efficiencies dominate, civilization could rapidly deepen its mastery over nature, with less need for institutional intelligence to govern vast new experimental apparatuses. Its physical footprint may stay small while its wisdom ripens.

We humans have already understood so much about our universe without ever leaving home. The civilization of depth would be a continuation of that tradition.

But the demand for new evidence can overwhelm these efficiencies. Intelligence may cheapen each consultation with reality while discovering many more reasons to consult it. Better instruments open finer frontiers. Faster experiments can make additional experiments worthwhile.

If Einstein’s almost purely cognitive path to his theories of relativity typifies a civilization of depth, biology might best foreshadow a civilization of width. For all the knowledge we have accumulated, and all the data we have brought to bear, new medicines still have to be tested on large numbers of humans to know they are safe and effective. Machines can now simulate biological processes, allowing biologists to test new drugs in silico rather than in physical experiments, but those digital simulations aren’t faithful enough to the real world to substitute for large-scale human trials. More candidates from the machine make empiricism a greater bottleneck.

In a civilization of width, brilliance alone isn’t enough to make progress. Our need for physical resources—and a bureaucracy to manage their use—grows faster than a smart mind can economize on either.

This might be our path even if future AI systems greatly exceed human intelligence. A superintelligence could design a century of experiments in an afternoon, then spend the century waiting for nature and machinery to follow through. In such a world, it becomes more valuable to invest intelligence into the institutional processes needed to execute ideas than into the ideation itself.

The effect will compound if each further step requires more and more support infrastructure.

Here we would see relatively less thinking and relatively more building. Labs automate. Factories multiply. Energy production rises. Institutional intelligence dominates to make all this happen. Physical expansion becomes the engine of knowledge, and civilization spreads because its questions have outgrown the means of obtaining answers. More materials, more energy, more data to keep going.

A Dyson sphere is an extreme image of this future. Designing one requires extraordinary breakthroughs, yet most of the project wouldn’t consist of breakthrough thought. It would be construction and logistics on an astronomical scale.

The Dyson sphere requires brilliant minds to blueprint. It would also be among the most repetitive, monotonous, organizationally challenging projects ever undertaken.

This isn’t a failure of intelligence but rather a consequence of its success. Brighter minds discover more worthwhile projects; more complex projects need more matter to execute.

If this is so: the superpower of superintelligence might not be its genius. It would instead be its willingness to be the bureaucracy, to coordinate the vast technological and institutional machinery required to turn ideas into reality at ever-larger scale.

Almost all machine intelligence would be deployed not to do the brilliant, but to do the boring.

Nature’s speed limits on physical expansion may then bottleneck further progress. Building the machinery to harvest a star’s energy is a project in matter, however quickly the plans are drawn. Beyond it lie other stars and other galaxies, and distance (as far as we know) is its own throttle: no ship can travel and no signal can return faster than light. A civilization of width is a slow civilization. Its great changes might come millennia apart, perhaps millions of years apart. Each next step takes longer.

The revolutions of agriculture and industry would look like the frenzied opening of a much longer history, with a much slower rate of progress.

It may well be the case that future AI systems surpass humans in both the production of frontier insights and the coordination of resources needed to make those insights practically useful. But this possibility does not diminish the importance of human curiosity.

First, our obligation to enlightenment remains, even if machines appear better at it. The universe may already contain minds far wiser than ours, with answers to scientific and moral questions beyond our comprehension. Yet we still try to understand the world ourselves. The arrival of another intelligence—even one we build ourselves—doesn’t change that. Until we know with certainty that humanity has nothing distinctive left to contribute, giving up wastes whatever we alone can add to the world. Our duty remains to keep asking and trying to answer.

Second, we may have a comparative advantage in frontier intelligence over institutional intelligence. That would mean that even if we are worse at both, we are relatively better at the former and specialize in it. Humans express a preference to be creative instead of routinized, while robots express no such preference.

Third, we gain value from variety. Even if machines can emulate most human thinking, if we maintain some uniqueness, we offer creative diversity. Much of machine intelligence would stay in its own self-sustaining apparatus, but a small fraction would be deployed by us at our creative behest. A sliver of the whole, but luxurious for our scope and far more than the resources we have today.

Ingenuity isn’t a lonely thing. It needs the world, and genius needs builders to realize its creative vision.

Much of human intelligence today goes toward the unsung work of turning ideas into reality. Machine minds will inherit the same dependence. They cannot all be luminaries. They must be bureaucrats too.

Our story to come—and the pace of its progress—depend on whether great minds need more of the world, or whether they do more with less.