The cycle that began with [YOU] on AI asks what it means to see the AI transformation clearly—without the narcotic of hype or the paralysis of fear. Ord supplies the most rigorous version of the answer at civilizational scale: see it as a hinge point, a narrow passage in which our destructive power has outrun our wisdom, and understand that the outcome is not predetermined. His concept of the Precipice—the distinctive and dangerous period humanity has entered since 1945, when for the first time we gained the power to destroy ourselves—frames AI not as a product or an industry but as the most consequential forcing event in the species' history.
Ord's framework amplifies the individual-scale question that [YOU] on AI poses to a reader. That book asks what a single life is worth, how one ought to live and work in the presence of these tools. Ord asks the same question of the species: what is the future worth, and what are we willing to do to protect it? His longtermist arithmetic—if almost all the people who will ever live are in the future, then almost all the value that humanity will ever realize is in the future too—is the civilizational version of the individual life-assessment that gives the orange pill its weight. The same logic that makes it worth taking the pill seriously as an individual makes it worth taking existential safety seriously as a civilization.
His contribution to the cycle is also methodological. Where other thinkers in the gallery offer qualitative diagnoses of AI's effects, Ord offers explicit probabilities defended in argument, inviting scrutiny rather than deference. His one-in-ten estimate for AI existential risk is not a prophecy but a considered judgment meant to make the danger legible—and the discipline of making such judgments explicit, of being willing to be wrong and say so, is itself a model for the kind of thinking the situation demands. The AI transformation is not inevitable in any specific form; the outcomes depend on choices being made now, and those choices will be made better by people who reason carefully about probabilities rather than by those who oscillate between techno-utopian confidence and existential dread.
His more recent analytical work on the alignment problem and on compute governance also represents an unusual discipline: a philosopher willing to learn the engineering well enough to see where the real levers are. His observation that inference-time compute is replacing training-time compute as the locus of capability—and that existing regulatory architectures built around training-compute thresholds may therefore be miscalibrated—is the kind of insight that only comes from someone unwilling to treat the technology as a black box to be reasoned about from the outside.
Ord began his academic career studying computer science before turning to philosophy, completing his doctorate at Oxford under John Broome and Derek Parfit. Parfit's influence is foundational: it was Parfit who argued that the difference between a catastrophe that kills ninety-nine percent of humanity and one that kills one hundred percent is far larger than the difference between killing none and killing ninety-nine percent, because only the latter forecloses the entire future. Ord took that asymmetry and built a career on it. If the future is long and rich and most of its people have not been born, then our generation holds a peculiar and weighty power over the whole span.
Before The Precipice, Ord was best known for founding Giving What We Can in 2009—pledging to cap his own income and give away everything above a modest threshold, and inviting others to commit at least a tenth of their earnings to the most cost-effective charities available. The organization grew into a movement; thousands of members have pledged hundreds of millions of dollars. This was not a sideline to his philosophy but its proving ground: Ord wanted to know how an individual could do the most good, and the same question, scaled to the species and stretched across millennia, eventually produced his account of existential risk. The intellectual path from Giving What We Can to The Precipice runs through a single question asked at different scales: given limited resources and vast uncertainty, what should we do?
The book that defined his public role, The Precipice: Existential Risk and the Future of Humanity (2020), appeared at an unusual moment: the year that COVID-19 made the concept of global catastrophic risk suddenly legible to a mass audience. It placed unaligned AI at the top of its list of existential threats, above nuclear war and engineered pandemics, on the grounds that AI uniquely combines unprecedented capability with the potential for irreversible foreclosure of the future. Since its publication Ord has increasingly engaged with the technical details of AI development—the scaling laws, the shift to inference-time compute, the governance architectures—producing essays that are among the most analytically rigorous contributions to the governance debate.
The Precipice and existential risk. An existential catastrophe, in Ord's precise definition, is the destruction of humanity's long-term potential—not necessarily extinction, but any irreversible foreclosure of the deep future. The Precipice is the stretch of time we are now in: powerful enough to ruin everything, not yet wise enough to be confident we won't. Ord estimates the per-century risk of an existential catastrophe at roughly one in six, with unaligned AI accounting for the largest single share at about one in ten. These are not precise measurements but considered judgments offered to make the danger legible and to invite scrutiny. Existential risk is the central concept of his work and the lens through which he evaluates every technological development.
The alignment problem as a moral problem. Most discussions treat AI alignment as an engineering challenge. Ord accepts that framing but insists it is incomplete: even if we could make AI do exactly what we tell it, we would still face the harder question of what to tell it. Alignment requires knowing whose values count, how to adjudicate among conflicting moral views, and how to avoid locking an impoverished conception of the good into a system powerful enough to entrench it forever. Ord's background in moral uncertainty—co-authored with William MacAskill and Krister Bykvist—informs his view that alignment is applied ethics with the highest imaginable stakes, not a problem that can be solved by engineering alone.
Existential security and the Long Reflection. Ord's goal is not merely to survive but to step back from the edge: existential security is the condition in which per-century risk has been driven low enough that humanity can expect to endure indefinitely. Beyond that lies his most speculative and most humanistic idea: the Long Reflection, a period in which humanity, freed from immediate existential danger, takes the time to deliberate carefully about what kind of future to build before locking in any particular vision. The Long Reflection is a plea for patience at civilizational scale—the recognition that some choices are so consequential and so permanent that they deserve far more deliberation than competitive dynamics permit.
Compute governance and inference scaling. Ord's recent analytical contribution concerns the technical architecture of AI capability. Existing regulations, including the EU's AI Act, were built on the assumption that capability tracks training-compute thresholds. Ord argues this assumption is becoming obsolete as inference-time compute replaces training as the locus of capability. A model trained below a regulatory threshold can be amplified through test-time reasoning to match frontier performance—meaning the regulatory net catches yesterday's risk while tomorrow's walks free. Compute governance, monitoring and potentially regulating access to the physical infrastructure of large-scale computation, is his proposed alternative: chips are physical, countable, and harder to hide than software.
Longtermism and the astronomical stakes. The moral framework underlying everything Ord does is longtermism: the view that positively influencing the long-term future is among the highest moral priorities of our time. The arithmetic is simple and vertiginous: if we can learn to value the lives of people in distant centuries as much as our own, and if there are far more future people than present people, then even a small reduction in existential risk outweighs enormous quantities of more immediate good. This reasoning, applied to AI, means the choices being made about how to build and deploy these systems are among the most morally significant decisions in human history. The stakes are not abstract to Ord; they are the future of everything humanity might ever become.