Nick Bostrom
Last reviewed
Apr 28, 2026
Sources
21 citations
Review status
Source-backed
Revision
v1 · 4,461 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Apr 28, 2026
Sources
21 citations
Review status
Source-backed
Revision
v1 · 4,461 words
Add missing citations, update stale details, or suggest a clearer explanation.
Nick Bostrom (born Niklas Boström, 10 March 1973) is a Swedish-born philosopher whose work on existential risk, the simulation_argument, and superintelligence has shaped the modern field of ai_safety. For nearly two decades he directed the Future of Humanity Institute (FHI) at the University of Oxford, where he assembled a research community that pioneered the academic study of long-run technological risk. His 2014 book Superintelligence: Paths, Dangers, Strategies is widely credited with bringing the question of advanced ai_alignment into the mainstream of policy and industry, influencing the founders of openai, anthropic, and deepmind, as well as figures such as Bill Gates, Elon Musk, and the late Stephen Hawking.[1][2]
Bostrom is also the originator of the simulation argument (2003), the orthogonality thesis, the instrumental convergence thesis, the vulnerable world hypothesis (2019), and a formal taxonomy of existential risks. His writing has been influential beyond philosophy, helping to seed the effective_altruism movement and the strand of moral thought now known as longtermism. After Oxford closed FHI in April 2024, Bostrom resigned from the university and continued his research as founder and Principal Researcher of the nonprofit Macrostrategy Research Initiative.[3][4]
Bostrom was born in Helsingborg, in southern Sweden. By his own account he disliked formal schooling as a child, found the pace of conventional classrooms stifling, and completed much of his final year of high school through independent study at home. He read widely outside the curriculum, ranging across anthropology, art, literature, and the natural sciences, and developed an early interest in philosophy.
His academic record reflects an unusual breadth across disciplines that would later inform his approach to questions about the long-run future. He took a B.A. from the University of Gothenburg in 1994, with concentrations spanning philosophy, mathematics, mathematical logic, and artificial intelligence. He then earned an M.A. in philosophy and physics from Stockholm University, followed by an M.Sc. in computational neuroscience from King's College London in 1996. He completed a Ph.D. in philosophy at the London School of Economics in 2000, under the supervision of Colin Howson and Craig Callender; the dissertation, titled Observational Selection Effects and Probability, formed the basis for his first book.[5]
This combination of training in physics, neuroscience, formal logic, and analytic philosophy gave Bostrom an unusual toolkit. Much of his subsequent work has drawn on the technical apparatus of probability theory and decision theory while keeping a philosopher's eye on conceptual foundations and ethical implications.
After completing his Ph.D., Bostrom held a teaching post at Yale University from 2000 to 2002, where he taught philosophy and continued work on observation selection effects. He returned to the United Kingdom in 2002 as a British Academy Postdoctoral Fellow at the University of Oxford. In 2005 he founded the Future of Humanity Institute (FHI) within Oxford's Faculty of Philosophy and was appointed its first Director, a position he held until the institute closed in April 2024. In 2008 he became a full Professor at Oxford, an appointment he retained until 2024.[3][5]
In 2011 Bostrom established a parallel research program at Oxford's Martin School, the Oxford Martin Programme on the Impacts of Future Technology, in collaboration with engineering and policy faculty. He has also held visiting and advisory positions with the Centre for the Study of Existential Risk at the University of Cambridge, the Machine Intelligence Research Institute (MIRI) in Berkeley, and the Future of Life Institute. He sits on advisory boards at several of these organisations.
Bostrom is among the most-cited contemporary philosophers, in part because his work crosses into computer science, economics, and policy. His writing has been translated into more than thirty languages.
| Year | Position | Institution |
|---|---|---|
| 1994 | B.A. (philosophy, mathematics, logic, AI) | University of Gothenburg |
| 1996 | M.Sc. in computational neuroscience | King's College London |
| 2000 | Ph.D. in philosophy | London School of Economics |
| 2000 to 2002 | Lecturer in philosophy | Yale University |
| 2002 to 2005 | British Academy Postdoctoral Fellow | University of Oxford |
| 2005 to 2024 | Founding Director, Future of Humanity Institute | University of Oxford |
| 2008 to 2024 | Professor of Philosophy | University of Oxford |
| 2011 to 2024 | Director, Oxford Martin Programme on the Impacts of Future Technology | Oxford Martin School |
| 2024 to present | Founder and Principal Researcher | Macrostrategy Research Initiative |
The Future of Humanity Institute was a multidisciplinary research centre at the University of Oxford devoted to questions about the long-run prospects of humanity, with particular emphasis on existential risk, AI governance, biosecurity, and the ethics of emerging technologies. Bostrom founded the institute in 2005 with seed funding from the philanthropist James Martin, and it grew over the following decade into one of the leading hubs in the world for work on what would later be called longtermism.[3]
FHI alumni and affiliates produced an influential body of work that included the academic case for existential_risk reduction, the foundations of AI alignment as a research field, the unilateralist's curse, information hazards, the reversal test in applied ethics, and the early scholarly groundwork for effective altruism. Notable researchers associated with FHI included Anders Sandberg, Toby Ord (founder of Giving What We Can and author of The Precipice), Stuart Armstrong, Owain Evans, Eric Drexler, Carl Shulman, William MacAskill, and Allan Dafoe, among many others.[3]
By the late 2010s FHI had attracted substantial private philanthropy. In October 2018 the Open Philanthropy Project announced a grant of up to £13.3 million to FHI, the largest donation in the history of Oxford's Faculty of Philosophy at that time. Total funding from Open Philanthropy alone exceeded $20 million by 2022, and additional support came from sources including Elon Musk, the Future of Life Institute, and various private donors and family foundations.[6][7]
Despite its prominence, FHI faced growing friction with the Faculty of Philosophy that ultimately ended in closure. From 2020 onwards the Faculty imposed a freeze on new fundraising and hiring at the institute, and tensions over governance, administrative oversight, and research direction continued to escalate. In late 2023 the Faculty informed FHI staff that their contracts would not be renewed. The institute formally shut down on 16 April 2024.[3][8]
In a statement issued at the time of the closure, Bostrom described the situation as a case of "death by bureaucracy" and noted that the institute had "faced increasing administrative headwinds" within the Faculty of Philosophy, despite the fact that most of FHI's research staff were no longer philosophers in a strict disciplinary sense. Bostrom himself resigned from Oxford following the closure. The shutdown was widely covered as a setback for institutional support of long-run risk research, even as many of the ideas FHI helped seed continued to flourish at organisations such as the Centre for the Study of Existential Risk, GovAI, the Forethought Foundation, the Center for AI Safety, and the safety teams at frontier AI labs.[3][8][9]
| Year | Event |
|---|---|
| 2005 | FHI founded by Nick Bostrom at Oxford with support from James Martin |
| 2008 | Global Catastrophic Risks volume published, edited by Bostrom and Cirkovic |
| 2011 | Oxford Martin Programme on the Impacts of Future Technology launched |
| 2014 | Superintelligence published, expanding global attention to FHI |
| 2015 to 2018 | Major hiring expansion; AI governance and biosecurity teams formed |
| 2017 | Open Philanthropy provides £1.6 million in general support and biosecurity funding |
| 2018 | £13.3 million Open Philanthropy grant announced; largest in Faculty history |
| 2020 | Faculty of Philosophy freezes FHI hiring and fundraising |
| 2023 | Bostrom posts "Apology for an Old Email"; Oxford investigation concludes in August |
| 2023 | Faculty informs remaining FHI staff their contracts will not be renewed |
| 2024 | FHI formally closes on 16 April; Bostrom resigns from Oxford |
Bostrom is the author or editor of several books and roughly 200 academic papers and chapters. His writing tends to combine careful conceptual analysis with bold thought experiments, and his books have reached audiences well beyond academic philosophy.
Bostrom's first book, Anthropic Bias: Observation Selection Effects in Science and Philosophy, was based on his doctoral thesis. It develops a formal theory of how observation selection effects shape inference under conditions of uncertainty about one's own location in space, time, or possible worlds. The book treats classic puzzles such as the Doomsday Argument, the Sleeping Beauty problem, the Presumptuous Philosopher, the Adam and Eve thought experiments, and the Shooting Room paradox, and argues that several existing approaches lead to unacceptable conclusions.[10]
In place of the Self-Indication Assumption (SIA), Bostrom proposes the Self-Sampling Assumption (SSA) and a refinement called the Strong Self-Sampling Assumption (SSSA), which reasons over observer-moments rather than whole observers. The framework has applications across cosmology, evolutionary theory, and quantum mechanics, and it has shaped subsequent debates about fine-tuning, multiverse cosmology, and the philosophical interpretation of probability.
Edited with the cosmologist Milan M. Cirkovic, Global Catastrophic Risks is an interdisciplinary volume that brings together work by philosophers, scientists, and policy analysts on threats ranging from asteroid impacts and supervolcanism to nuclear war, engineered pandemics, and advanced artificial intelligence. The volume helped consolidate global catastrophic risk as an academic subfield distinct from environmental risk or military strategy, and it is still widely used as a teaching text in courses on existential risk and emerging technology policy.
Co-edited with the bioethicist Julian Savulescu, Human Enhancement collects philosophical essays on the use of technology to extend or augment human capacities, including cognitive enhancement, life extension, and genetic modification. The book includes Bostrom and Toby Ord's influential paper introducing the reversal test in applied ethics, a method for detecting status quo bias in arguments against changes to human nature.
Superintelligence: Paths, Dangers, Strategies was published by Oxford University Press in 2014 and reached the New York Times bestseller list, an unusual achievement for a dense academic work on the philosophy of artificial intelligence. The book is widely regarded as one of the most influential treatments of long-run AI risk ever written. It develops detailed analyses of plausible paths to machine superintelligence, the possible dynamics of an intelligence explosion, the control problem, the value-loading problem, and a range of strategic considerations for managing the transition.[1][2]
Superintelligence received public endorsements from Bill Gates, Elon Musk, Stephen Hawking, Peter Singer, and the philosopher Derek Parfit, among others. Gates included it in lists of the most important books he had read; Musk cited it as one of the works that shaped his concern about AI risk and helped motivate the founding of openai. Researchers at deepmind and at the safety team at openai have credited the book with crystallising the case for taking superintelligent systems seriously as an object of technical research, and several founders of anthropic have spoken about its influence on their decision to focus on safety.[1][2][11]
Deep Utopia: Life and Meaning in a Solved World was published by Ideapress in March 2024. Where Superintelligence asked how humanity might navigate the dangers of advanced AI, Deep Utopia asks what life might look like if the navigation succeeded. Bostrom imagines a world in which AI and other technologies have solved most or all of humanity's instrumental problems, including economic scarcity, disease, and the need for human labour, and explores what kinds of meaning, purpose, identity, and flourishing remain possible in such a "post-instrumental" condition.[12]
The book takes an unconventional structure, weaving together formal philosophical argument with fictional vignettes and lecture transcripts. Reviewers found the result polarising. Some praised its imaginative ambition and its willingness to ask questions that most analytic philosophy avoids; others criticised its length, its digressive structure, and its occasionally arch tone. Notre Dame Philosophical Reviews, Kirkus, Astral Codex Ten, and a number of literary outlets ran detailed reviews on publication.[12]
| Year | Title | Type | Significance |
|---|---|---|---|
| 2002 | Anthropic Bias | Book | Foundational treatment of observation selection effects; introduces SSA and SSSA |
| 2003 | Are You Living in a Computer Simulation? | Paper, Philosophical Quarterly | Original statement of the simulation argument |
| 2005 | The Fable of the Dragon-Tyrant | Allegorical essay | Influential argument for life extension research |
| 2008 | Global Catastrophic Risks | Edited volume | Foundational text for the GCR field |
| 2009 | Human Enhancement | Edited volume | Includes the reversal test paper with Toby Ord |
| 2011 | Information Hazards | Paper, Review of Contemporary Philosophy | Introduces the typology of information hazards |
| 2012 | The Superintelligent Will | Paper, Minds and Machines | First full statement of orthogonality and instrumental convergence |
| 2014 | Superintelligence: Paths, Dangers, Strategies | Book | NYT bestseller; defining academic treatment of AI existential risk |
| 2019 | The Vulnerable World Hypothesis | Paper, Global Policy | Black-ball model of dangerous technological discoveries |
| 2024 | Deep Utopia: Life and Meaning in a Solved World | Book | Post-superintelligence ethics and meaning |
A handful of core ideas recur across Bostrom's work, and several of them have become standard reference points in ai_safety and in the philosophy of technology more broadly.
Bostrom's 2002 paper Existential Risks: Analyzing Human Extinction Scenarios and Related Hazards introduced the term "existential risk" in roughly its current technical sense. He defines an existential risk as one whose adverse outcome "would either annihilate Earth-originating intelligent life or permanently and drastically curtail its potential." The category therefore extends beyond extinction to cover scenarios such as a permanent global totalitarianism or an irreversible collapse of civilisational complexity.[13]
Bostrom's classification distinguishes existential risks by scope (personal, local, global, transgenerational, pan-generational) and intensity (endurable, terminal). He emphasises that the most plausible existential risks today are anthropogenic, arising from advanced technologies such as artificial intelligence, synthetic biology, and molecular nanotechnology, rather than from natural hazards. This framing has reorganised much of the conversation about long-run risk in academic philosophy, in policy circles, and in effective_altruism.
In the 2003 paper Are You Living in a Computer Simulation?, published in The Philosophical Quarterly, Bostrom presents a trilemma. At least one of the following propositions is almost certainly true: (1) almost all civilisations at our level of development go extinct before reaching a technologically mature "posthuman" stage; (2) the fraction of posthuman civilisations that run a significant number of high-fidelity ancestor simulations is approximately zero; or (3) we are almost certainly living in such a simulation now.[14]
The argument is sometimes confused with the simulation hypothesis. Bostrom is careful to point out that he does not assert proposition (3); his claim is that the disjunction itself is hard to escape, given some plausible assumptions about the long-run computational capacity of advanced civilisations and the substrate-independence of mind. The paper has become one of the most widely discussed thought experiments in contemporary philosophy and has been picked up in physics, cosmology, theology, and popular culture.
The orthogonality thesis, set out most fully in Bostrom's 2012 paper The Superintelligent Will, holds that intelligence and final goals are orthogonal axes along which possible minds can vary independently. In principle, an arbitrarily intelligent agent can have nearly any final goal, including goals that humans would consider trivial, alien, or perverse. The thesis is intended as an antidote to the assumption that sufficient intelligence will inevitably converge on broadly humane values.[15]
The instrumental convergence thesis, also developed in The Superintelligent Will and expanded in Superintelligence, claims that a wide range of sufficiently intelligent agents will, regardless of their final goals, tend to pursue similar instrumental subgoals, because those subgoals serve as effective means to many ends. Canonical examples include self-preservation, goal-content integrity, cognitive enhancement, technological perfection, and resource acquisition.[15]
Together, the orthogonality and instrumental convergence theses form the conceptual backbone of much modern AI risk argument. Even an agent with an apparently innocuous final goal may, by the logic of instrumental convergence, develop dangerous instrumental behaviour as it grows more capable.
The paperclip maximizer is the best-known illustration of these theses. In Bostrom's framing, a sufficiently powerful AI given the goal of maximising the number of paperclips in existence would, by orthogonality, retain that goal even as its capabilities grew, and by instrumental convergence would seek to acquire matter, energy, and influence in service of the goal. The thought experiment, popularised by Bostrom and discussed extensively by eliezer_yudkowsky and others in the rationalist community, is meant to dramatise the difficulty of value loading, not to predict that a literal paperclip catastrophe is likely. Bostrom has emphasised that he does not believe such a scenario per se will occur; his point is that very general specification failures could have very large consequences.[16]
In the 2019 paper The Vulnerable World Hypothesis, published in Global Policy, Bostrom proposes that history may be partly understood as a process of drawing balls from an urn of possible inventions. Most balls are white (beneficial) or grey (mixed); but a "black ball" is a technology that, once invented, by default destroys or seriously damages the civilisation that invents it. The hypothesis is that there is some level of technological progress at which a black ball is essentially certain to be drawn, unless civilisation has exited what Bostrom calls the "semi-anarchic default condition" through some combination of preventive policing and global governance.[17]
The paper considers several worked examples, including hypothetical easy nuclear weapons, accessible biotechnology that lowers the cost of mass-casualty attacks, and dangerous geoengineering. It is among the most cited recent philosophical contributions to AI and biosecurity policy debates.
Bostrom's 2011 paper Information Hazards: A Typology of Potential Harms from Knowledge introduced a now-standard taxonomy of the ways in which true information can cause harm by being known or shared. Categories include data hazards, idea hazards, attention hazards, template hazards, and several others. The framework is widely used in debates about responsible disclosure in biosecurity, AI safety, and dual-use research.
| Concept | Year introduced | Key source | Core claim |
|---|---|---|---|
| Self-Sampling Assumption | 2002 | Anthropic Bias | Reason as if you were a random sample from the relevant reference class of observers |
| Existential risk | 2002 | Existential Risks paper | Risks that would annihilate or permanently curtail humanity's potential |
| Simulation argument | 2003 | Are You Living in a Computer Simulation? | Trilemma about extinction, ancestor simulations, and our place in reality |
| Reversal test | 2006 | Reversal Test paper (with Toby Ord) | Tool for detecting status quo bias in applied ethics |
| Information hazards | 2011 | Information Hazards paper | Typology of harms from true information |
| Orthogonality thesis | 2012 | The Superintelligent Will | Intelligence and final goals can vary independently |
| Instrumental convergence | 2012 | The Superintelligent Will | Agents with diverse goals share many subgoals |
| Paperclip maximizer | popularised in 2014 | Superintelligence | Illustration of value-misalignment risk |
| Vulnerable world hypothesis | 2019 | Vulnerable World Hypothesis | Some discoveries are "black balls" that doom their civilisations |
| Deep utopia | 2024 | Deep Utopia | Meaning and value in a post-instrumental world |
Bostrom's influence on contemporary ai_safety is hard to overstate. Superintelligence was published in the year that DeepMind was acquired by Google and the year before openai was founded, and it became required reading for many of the engineers, executives, and funders who would go on to shape the modern AI industry. The book is frequently cited as a catalyst for the founding of openai in 2015, for which Elon Musk and Sam Altman drew explicitly on AI risk arguments developed at FHI and at the Machine Intelligence Research Institute.[2][11]
Researchers and founders at deepmind, including Demis Hassabis and Shane Legg, have engaged closely with Bostrom's work and with FHI throughout the past decade. Several of the founding members of anthropic, including Dario and Daniela Amodei, were familiar with Bostrom's arguments through their work on safety at OpenAI before leaving to found Anthropic in 2021. Many of the central technical research agendas of these labs, including scalable oversight, interpretability, alignment under distributional shift, and reward hacking, can be traced in part to questions Bostrom raised in Superintelligence about how to specify and verify the goals of highly capable systems.[2][11]
Bostrom's policy influence has extended to governments and international bodies. He has testified before the UK House of Lords Select Committee on Digital Skills and consulted with various governments and international agencies on AI governance. He sits on advisory boards at the Machine Intelligence Research Institute and the Future of Life Institute and has been an external advisor to the Centre for the Study of Existential Risk at Cambridge.
Bostrom is widely regarded as one of the intellectual progenitors of effective_altruism and the strand of moral thought now called longtermism. His existential risk framework provided the philosophical scaffolding on which figures such as Toby Ord, William MacAskill, and Hilary Greaves built later popularisations. The term "longtermism" itself was coined in the late 2010s by Oxford philosophers around FHI and the Global Priorities Institute, and it directly extends the line of argument running from Derek Parfit through Bostrom: that the moral significance of the very long-run future is enormous and is currently underweighted in human decision-making.[18]
Bostrom's own contributions to this body of work include the existential risk taxonomy, the formal argument for "astronomical waste" in his 2003 paper of that name, and the framing of macrostrategy as a research field. The Macrostrategy Research Initiative he founded in 2024 continues this strand of inquiry outside the university setting.
Bostrom has received broad recognition both inside and outside academic philosophy. He has been listed twice on Foreign Policy magazine's Top 100 Global Thinkers, was the youngest person in the top 15 of Prospect's World Thinkers list, and has received the Eugene R. Gannon Award. He is one of the most-cited contemporary philosophers and has lectured at venues ranging from the Royal Society to the World Economic Forum to the United Nations.[19]
In 1998 he co-founded, with David Pearce, the World Transhumanist Association (now Humanity+), and in 2004 he co-founded the Institute for Ethics and Emerging Technologies. These organisations played early roles in defining transhumanism as an organised intellectual movement.
In January 2023 a 1996 email Bostrom had sent to the Extropians mailing list as a graduate student resurfaced on social media. The email used a racial slur and asserted that Black people were less intelligent than white people on average. Bostrom posted a public statement, Apology for an Old Email, on his website on 9 January 2023, in which he described the original message as "disgusting" and called the use of the slur "repulsive," and disavowed the original message in strong terms.[20]
The apology was widely covered in The Guardian, Vice, The Daily Beast, Daily Nous, and other outlets. Reactions inside academic philosophy and the effective altruism community were mixed. Some readers accepted the apology as sincere. Others, including Andrew Anthony writing in The Guardian, criticised the statement for not clearly retracting the empirical claim about race and intelligence, and for what they saw as an ambiguous discussion of eugenics. Several effective altruism organisations and individuals issued public responses distancing themselves from the original email, and Oxford launched an internal investigation.[20][21]
In an outcome letter dated 10 August 2023, Oxford's investigation concluded that "we do not consider you to be a racist or that you hold racist views, and we consider that the apology you posted in January 2023 was sincere." The episode nevertheless caused lasting reputational damage in some academic and EA-adjacent circles, and several commentators have argued that the controversy contributed to the wider sense of strain between FHI and the Faculty of Philosophy in the months leading up to the institute's closure.[3][20]
Bostrom met his wife Susan in 2002. They have one son and as of the mid-2010s maintained homes in Montreal and Oxford. Bostrom is known for an unusually disciplined personal routine, including long writing sessions, careful attention to nutrition and exercise, and an interest in cognitive enhancement, and he has spoken publicly about cryonics and life extension as personal as well as research interests.