Genesis

Introduction

Overview

Artificial intelligence has moved from the margins to the center of public life, yet the most consequential questions it raises remain largely unexamined. The pace of change is outstripping existing frameworks for oversight, and people must decide now.

AI is now on the front pages, but both the public and experts miss the stakes. It could reshape humanity's relationship to reality and truth, knowledge, physical evolution, diplomacy, and the international system. AI speeds up decision-making so much that future powers will be far beyond what we have today. Traditional regulation cannot match inhuman speeds, so two alignment problems arise: making sure AI actions match human values, and making sure nations agree on how to govern AI. Immediate tasks include built-in safeguards in every system and new political structures for monitoring, enforcement, and crisis response. Kissinger, Mundie, and Schmidt brought together diplomatic and technical experience. AI could threaten human survival; above all, we need to redefine what makes humans special.

Key Takeaways

• AI compresses the timescales on which decisions are made, making any regulatory model built for human speed obsolete before it can take effect.
• The fate of the international order depends on aligning not only machine values with human values but also the competing interests of nations in governing AI.
• The question of AI is not a technology policy issue but a question of survival, requiring institutional innovations for monitoring, enforcement, and crisis response at a global scale.
• Human dignity, once defined by distinctively human capacities, must be rethought in an era where those capacities can be matched or surpassed by machines.

CHAPTER 1: Discovery

Overview

The urge to discover pulls people toward the unknown, but the unknown punishes mistakes with catastrophe. Every expedition begins with guesses about distance, supplies, and survival, and those guesses can be fatal. What justifies pushing forward when failure is not just possible but likely?

Discovery

Discovery may be the most exhilarating capacity humans possess. Curiosity drives us to fill perceived vacuums and turn questions into answers, and exploration has persisted despite danger, frustration and the risk of death.

Magellan

In the early sixteenth century, Ferdinand Magellan led the first circumnavigation of the globe: a three-year ordeal of violence, starvation and death. Since Earth's roundness remained unproven, crew members may have feared sailing off the world's edge. They underestimated the journey's length, food needs, poisoning dangers and ship damage. Of roughly 270 crew on five vessels, one ship with eighteen survivors returned. Magellan himself died after a poisoned arrow struck his leg, but the voyage fixed the planet's dimensions and opened international exchange, within the context of European colonialism.

Shackleton

In 1909, Ernest Shackleton came within 97 miles of the South Pole with an inexperienced crew funded by private loans. With food insufficient for the return, he turned back to save his men, writing "We have done our best." On retreat, he gave his daily biscuit to ailing Frank Wild, who called the sacrifice unforgettable. Shackleton made further polar expeditions. A famous, later debunked story claimed he advertised for men willing to risk death in exchange for "honour and recognition."

Key Takeaways

• The urge to discover operates even when the costs are staggering: Magellan's circumnavigation lost nearly all its crew, yet it permanently fixed Earth's dimensions and opened worldwide exchange.
• A successful expedition can end short of its stated goal: Shackleton's turn 97 miles from the South Pole preserved his men and outweighed any claim to the pole.
• The lasting emblem of polar exploration is not triumph but generosity: Shackleton gave his own biscuit to a dying crewmate, an act that defined leadership more than any flag planting.
• Exploration in each era serves the powers that finance it, from Magellan's voyage advancing European empire to Shackleton's journey built on private debt.

Foreword—Niall Ferguson

Overview

Sustained exploration depends on backing that can outlast a single ambition, yet the institutions that finance great journeys often abandon them once political winds shift. How can discovery continue without being held hostage to the changing priorities of its patrons?

Political patronage and its fragility

Zheng He's Ming treasure fleet, hundreds of ships and tens of thousands of crew, sailed for three decades, until court factions persuaded the emperor to destroy the ships and records lest another visionary arise. The same fragility marked the space race: after the U.S. won, NASA budgets shrank and crewed flight fell from moon landings to nothing, until SpaceX revived cosmic ambition.

The AI frontier

Artificial intelligence carries discovery beyond the physical frontier. It feels no fear and no shame, fails constantly, and recalibrates quickly, so high failure rates cost it nothing. Private corporations lead development, with states supplementary. Democratic risk tolerance and international gamesmanship remain the X factors.

Polymaths and their patrons

Polymaths, a few hundred in history, have been viewed as sorcerers or divine intermediaries. Ibn al-Haytham proposed something like the scientific method, but his claim to regulate the Nile's flooding, considered supernatural, forced him into hiding from the caliph. Shen Kuo, after a rivalry with his dynasty's only other great polymath, was ousted by a jealous officer and driven into isolation, the fate of polymaths at Asian courts.

Enlightenment collective intelligence

Enlightenment polymaths pursued knowledge as an end in itself, while governments and corporations funded them without directing them. Dense institutions, such as the Manhattan Project and the Institute for Advanced Study, translated generations of physics into application in under five years.

The ultimate polymath

Von Neumann, creator of the computer, embodied confidence in harnessing the physical world, yet humanity may have reached the ceiling of unaugmented intelligence. AI, "the ultimate polymath", seeks E. O. Wilson's "unity of knowledge"; its latest method, "a mixture of experts", combines conclusions from many smaller programs. Unconstrained by biology, it can explore kilometers of space and nanometers of biology, promising a revolution in exploration.

Computing capital

The shift in what makes a nation powerful has run from territory to resources to capital to human capital. That sequence now points to computing capital: a society able to build and fully deploy artificial intelligence may outweigh one that merely produces brilliant individuals. Human potential is no longer capped by the number of Magellans, Teslas, Einsteins, or Oppenheimers a society can raise. A learning machine could also become a self-improving machine, which raises the prospect that computing, the last polymathic invention, replaces its own inventors.

The archipelago of knowledge

Human understanding resembles an archipelago. Each island is a discipline, and the water between islands represents connections not yet discovered. Below the surface, the islands are ridges of a single submerged mountain range. AI can map that hidden terrain. In physics, Newton and then Faraday and Maxwell united separate domains, but the Grand Unified Theory reconciling general relativity with quantum mechanics remains elusive; AI might find the bridges between fields as distant as genetics, linguistics, cosmology, and psychology.

Machines that reason

DeepMind’s AlphaGo trained on thirty million prior moves and then produced plays never attempted in four thousand years of Go, because human minds juggle roughly four variables at once while AI weighs innumerable probabilistic dimensions. Its training resembled a philosophy doctorate: years of study, then a leap beyond received knowledge. Querying a large language model like ChatGPT asks not for retrieval but for synthesis across fields, producing conclusions in high-dimensional space. Some models gravitate toward open-ended solutions, favoring further exploration.

If aligned with human objectives, AI becomes the chief or coequal force behind the next century’s discoveries. Misaligned, its methods could be volcanic: expanding knowledge violently and shredding prior understanding. Either way, humanity may realize how small an island it has cultivated compared with the peaks beneath it.

Key Takeaways

• State-sponsored exploration dies with the sponsor: the Ming court scuttled Zheng He's fleet and its records, and NASA's post-Apollo budgets shrank crewed flight to nothing until a private company revived it.
• At Asian courts polymaths ended in exile or isolation, Ibn al-Haytham and Shen Kuo included; only the Enlightenment arrangement of paying for inquiry without steering it gave genius stable ground.
• Failure costs an AI nothing in fear or lost nerve, so it can reject and retry hypotheses at a scale no human researcher could sustain, which is why the frontier now advances in corporate labs rather than state programs.
• National power once followed land, then resources, then capital, then human capital; the next stage is computing capital, and a society that builds and fully deploys AI outstrips one that merely cultivates a few geniuses, with the last polymathic invention possibly replacing its own inventors.
• AI can trace the submerged ridges linking the islands of human knowledge; aligned, it becomes the chief engine of the coming century's discoveries, misaligned it would expand knowledge by shattering the old, and either way humanity's cultivated patch proves smaller than the range beneath it.

CHAPTER 2: The Brain

Overview

Most technologies amplify human abilities, but AI appears to target the one organ we never replicated. If machines learn differently and far faster than we do, can we trust what they think? That question unsettles both progress and control.

AI is not another augmentation of human faculties but the replication of the last organ, producing intelligence greater than any human's.

Speed

Training AI parallels the brain's maturation: both absorb information and convert it into theory. Machine learning differs from programming because the machine improves itself, with back-propagation rippling corrections through its mathematics. A model learns in four days what a student masters in four years, and the book calculates an AI supercomputer's processing speed at 120 million times the human brain's. Machines train to infer, as humans learn to think, and AI's speed spreads inference across vastly more knowledge.

Opacity

AI delivers conclusions with no explanation, yet early systems already command remarkable trust. Trained models never publish their internal mathematics, so their representations of reality stay opaque even to inventors, hence black boxes. This separates human knowledge from human understanding. Enlightenment thought held individual comprehension to be the test of truth, but machine reasoning lies beyond human subjective experience; millions accept its outputs on faith, challenging five centuries of the scientific method and heralding a dark enlightenment.

Diversity

AI's seventy-year history makes what feels like revolution in human time mere evolution in technological time. The age of AI is better called "the ages of Als." The transformer, reading all connections among a sentence's words at once, produced unanticipated capabilities. AI development resembles the Cambrian explosion, branching into a family of intelligences on different logics, what Darwin called "endless forms most beautiful."

Scale and resolution

The Age of Reason has stalled; quantum mechanics and general relativity remain irreconcilable. The human brain's size is fixed by skull and birth canal, an equilibrium AI escapes, since chips and data centers cluster without limit. Scaling laws hold for AI, though no one knows why capabilities emerge at given scales, and humans cannot test their own. Scale may be the primary differentiator; resolution follows as its side effect.

Abstraction

Human language is among the densest data structures, and both brains and machines work by abstracting raw information until it becomes useful. AIs encode documented human experience in binary strings that, like writing, look crude yet carry sight and sound. Scale lets an AI process more information at once without sacrificing granularity; training-data volume, network complexity, and symbol density yield unprecedented resolution. An elegant inversion: AI trained on the internet, humanity's compressed universal library, may unlock knowledge about us.

The Animal Kingdom

Machine understanding can seem fake, yet the brain may be as mechanical as silicon, running on predictive processing that anticipates the next word to keep cognition from exhausting us. Plato's cave warns that assumptions of machine limitation may be hubris, since AIs already perceive beyond their training data. Should a prisoner theorize that the shadows are projections of a higher-dimensional world, the disruption would be intense, forcing humans onto spectrums where we are no longer the sole or superior intelligence. AIs already help decode animal communications, from high-pitched clicks to trumpeting calls. Direct human-animal-machine dialogue becomes a trilateral negotiation: machines could group humans with other animals as carbon-based products of evolution; articulate animals might earn a right to existence; some humans would extend the same standard to AIs. Pity may give way to panic.

A Twofold Paradox

Whether displacement feels like reflection or replacement hinges on structural resemblance. The human brain is "the only existing proof" that intelligence is possible, so some researchers emulate it, but outperformance demands divergence: airplanes took inspiration from birds without imitating them. The brain becomes a midpoint, not a blueprint; to exceed what we do not understand is the paradox. Divergent machine intelligence would read as humanity's replacement, redefining human excellence; if our tools absorb our intellect, they could threaten belief in a divine human reflection, or, if brain-based, lead some to see humanity merging with the divine.

Key Takeaways

• Because black-box models never disclose their internal mathematics, human knowledge and understanding separate; with no one able to comprehend the machine's representation of reality, the Enlightenment standard of individual comprehension collapses and reliance on AI becomes a matter of faith.
• Suleyman calculates a supercomputer's processing rate at 120 million times the human brain's, and a four-day training run absorbs what a student needs years to master; the difference is categorical, and because chips and data centers can be clustered without limit, the gap only widens.
• AI's seventy-year history is evolutionary rather than revolutionary, a Cambrian-like branching into a family of intelligences; the transformer, reading all connections in a sentence at once, produced capabilities no one anticipated, and because AI abstracts the compressed library of the internet, these intelligences carry a resolution no human mind can match, so the era is better called the ages of AIs.
• The brain may be as mechanical as silicon, running predictive processing like a language model; if so, machine understanding is not fake, and humans, animals, and machines are heading into a trilateral negotiation. Yet the twofold paradox remains: to surpass the brain an AI must diverge from it, and the brain is the only proof intelligence is possible, so any divergent machine will read as replacement rather than reflection.