Key Takeaways
Treat states and corporations as artificial agents with superhuman reach
Power without superior thought. Runciman’s central claim is that modern states and corporations are artificial agents: human-made organisations that act in the world and carry consequences beyond any individual’s reach. Their advantage is not greater intelligence. It is the ability to bind millions of people, persist as members change, and turn ordinary human decisions into actions at enormous scale. He calls this superagency. A government decides on war; the state mobilises soldiers and resources, including people who opposed the decision.
The first transformation. The First Singularity is the earlier transformation produced by modern states and corporations, which reshaped human life before digital AI existed. A possible Second Singularity would come from AI changing the human condition again. The point is not that these institutions are literally robots, but that their machine-like agency already makes us dependent on powers we built.
What's striking is how agency differs from intelligence. A corporation can produce immense consequences without having a mind, just as a legal system can enforce rules without understanding their human cost. This distinction is useful in debates about algorithmic decisions: asking whether a system is conscious may distract from asking who gave it authority and who bears the outcome. The analogy has limits, though. Institutions contain competing people and procedures, not a single mechanical will. Calling them agents clarifies their effects, but should not make their internal conflicts disappear.
Engineer group decisions; headcounts alone do not produce collective wisdom
Procedure shapes the result. A group’s conclusion can differ from what most members privately believe, depending on how their views are gathered. In the O. J. Simpson murder trial, jurors reportedly began with a vote that was 10-2 for acquittal. After the foreperson asked the majority to explain its doubts, the two jurors inclined to convict changed their minds. A different process, weighing each piece of evidence before asking for an overall verdict, might have produced another result.
Wisdom needs conditions. The Condorcet’s Jury Theorem suggests that a larger group can be more likely to reach the right answer when members judge independently and each has a better-than-even chance of being right. But independence is fragile. A crowd can also amplify a rumour or prejudice. A group becomes useful not just by gathering many people, but through rules that protect independent contributions and make the decision process clear.
The deeper lesson is that collective intelligence is an institutional achievement, not a natural property of crowds. This connects to experimental design: changing the order of questions can change the data a group appears to produce. The same is true in workplaces, courts, and public consultations. Yet a well-designed procedure cannot guarantee a just outcome. Independence may improve accuracy on questions with verifiable answers, but political choices often involve competing values that cannot be settled by counting correct guesses.
Ask whether a system discovers truth or makes a decision real
Answers and decisions differ. An answer describes what is true; a decision changes what happens. Averaging guesses about jellybeans in a jar aims to estimate a fixed number. A jury’s verdict, by contrast, changes the accused person’s legal situation, even if the verdict is mistaken. Confusion arises when systems do both, as markets do: prices reflect what buyers will pay, while those buying and selling also change the price.
Algorithms inherit choices. A search engine may present results as if they were neutral answers, but its output reflects users’ behaviour and the designers’ decisions about how to rank it. Likewise, when an insurer says a computer rejected a claim, the system may only have applied criteria chosen by people. The algorithm did not decide that those criteria were fair or should be binding. Treating a human choice as an impersonal answer can make prejudice look like fact and make accountability vanish.
This distinction resembles a basic problem in measurement: indicators can describe a world, but once institutions act on them, they can also reshape it. A credit score, for example, may summarize past information while influencing access to future credit. The book’s framework helps expose that feedback. Its caution also cuts both ways: not every algorithmic answer is disguised authority, and some decisions genuinely require technical evidence. The important question is whether the system is informing a choice or quietly making one.
Trace responsibility through institutions that outlive their human members
Durability shifts the burden. The US national debt does not belong to a particular president or generation of citizens. It belongs to the state, which can borrow and repay across changes in government. The same durability lets states make binding decisions about war. Their machinery can sustain obligations longer than any official or soldier involved, but the people who act for the state remain human.
Corporate personhood complicates blame. Limited liability, established in Britain through laws in 1844 and 1855, allowed investors to risk their contributions without putting all their personal assets at stake. That helped corporations attract capital and plan over time. It also created a moral hazard: members may take risks while the organisation absorbs the cost. If responsibility is assigned only to individuals, the group’s role can be obscured. If it is assigned only to the group, individuals may escape blame. There is no simple answer, but the tension should remain visible.
The problem resembles the principal-agent gap: people make choices through organisations, but the consequences are distributed across members, customers, and the public. Corporate liability can make compensation possible, yet a financial penalty may be too small to deter decision-makers who do not personally bear it. Individual liability can improve incentives but may unfairly blame people with little control. Effective accountability therefore depends on tracing both authority and exposure, rather than treating either the institution or its members as the whole story.
Build repeatable institutions, not one-off empires, for lasting prosperity
Scale became reproducible. Ancient empires such as Rome could govern vast populations, but they were unique political giants whose subordinate institutions could not easily become independent models elsewhere. Modern states and corporations spread through more repeatable forms. Their durable rules and organisations helped support long-term investment, scientific work, and economic growth. The modern transformation was not simply capitalism: it depended on stable states and corporations able to make commitments beyond the lifespan of particular rulers or owners.
Balance matters. The Narrow Corridor describes the difficult space in which a capable state and an active society constrain each other without either gaining unchecked control. The divergent paths of North and South Korea, which began with many shared characteristics, show how institutions and political organisation can shape very different outcomes. Growth is not automatically fair: colonial extraction helped enrich some states while leaving others poorer. Replicable institutions expand prosperity’s reach, but do not guarantee its equitable distribution.
The argument is strongest when it treats institutions as infrastructure for cooperation rather than as a single recipe for success. Long-term commitments can lower uncertainty and make investment possible, while civic power can check public authority. This echoes institutional economics, which studies how rules shape incentives and trust. The caution is historical: the same institutional forms can produce very different results across cultures and power relations. Replication can spread useful capacities, but it can also export hierarchy, extraction, and rules designed to benefit outsiders.
Let public institutions carry early innovation risk, then inspect who profits
The state can fund failure. States can keep spending on uncertain projects because they can tax, borrow, and survive political changes. During the Second World War, US defence spending grew fortyfold between 1941 and 1945, funding many failed projects as well as advances such as radar and the atomic bomb. The book’s broader pattern is that states often support risky research, while corporations refine technologies and sell them at scale.
Public origins, private returns. DARPA, the US defence research agency, helped fund ARPANET, a precursor to the internet. Public research also contributed technologies later integrated into products such as the iPhone. This does not mean every subsidy is wise: the state can waste vast sums, and well-connected companies can capture the gains. But markets alone may not finance research whose rewards are uncertain, distant, or shared widely. The policy challenge is to preserve public capacity for experimentation while scrutinising who receives the benefits.
This account usefully challenges the story that innovation springs fully formed from lone entrepreneurs. Research systems often distribute risk across universities, government, and firms, while profits concentrate in the organisations best positioned to commercialise discoveries. That asymmetry makes public investment a question of governance, not just efficiency. A successful product does not prove that every state-funded project was justified, and public funding does not guarantee public benefit. Transparent conditions on access, returns, and accountability matter as much as the initial investment.
Judge automation by the jobs organizations create, not tasks machines perform
A task is not a job. A 2017 study gave sports officiating a 98 percent probability of automation within twenty years, because machines can call many plays more accurately than people. Yet technology has often added human roles as well as replacing tasks. Tennis uses automated line calls, while other sports employ replay officials and review centres. Some organisations retain human officials for the appearance and experience of a human-run game.
Work changes unevenly. When cars replaced horses, millions of new jobs arose around the automobile, but few new roles appeared for horses. The comparison warns that workers cannot assume every displaced skill will transfer. Humans, unlike horses, can organise and influence the institutions choosing how technology is used. The book also points toward more work in care and education as societies age, and considers universal basic income as one possible response to fragmented employment. Neither new jobs nor fair transitions are guaranteed.
The task-versus-job distinction improves on forecasts that treat employment as a checklist of automatable activities. Organisations hire people for trust, legitimacy, appearance, and relationships, not just output. That helps explain why a more accurate machine can coexist with more human work. The horse comparison is vivid but imperfect: workers can bargain, retrain, and change policy in ways animals could not. Those capacities matter only when institutions make them meaningful, especially for people with little time or financial security to absorb disruption.
Restrain artificial growth before its appetite destabilizes the planet
Humanity escaped one trap. The Malthusian trap describes a cycle in which population grows faster than food production, leading to scarcity, disease, and death. Improvements in agriculture, health, and public organisation helped break that cycle. The Green Revolution, beginning in the 1950s, sharply increased crop yields and is credited with preventing widespread starvation. As living standards rose, many societies then entered a demographic transition: people lived longer and had fewer children.
Institutions have their own appetite. States and corporations now drive enormous consumption of energy and resources. The Anthropocene names the era in which human activity has become a dominant force in planetary systems; Runciman’s term Leviacene shifts attention to the artificial institutions that organise that activity. The current danger is a mismatch between their drive to expand and the planet’s limits. Technology might help reduce emissions, but it can also create new demands. Better efficiency alone does not guarantee restraint.
What's important here is the shift from individual behaviour to system-level incentives. People can make choices about consumption, but infrastructure, investment, and energy systems determine which choices are practical and profitable. Environmental economics also warns that efficiency can sometimes encourage additional consumption, offsetting some savings. That does not make technical improvements useless. It means they need to be paired with rules that limit total harm and make long-term costs count in present decisions.
Keep human judgment inside systems with power to end worlds
The danger is institutional. Nuclear conflict, climate change, biological disaster, and uncontrolled AI are often imagined as rogue-technology stories. Runciman argues that states and corporations are more central: they fund research, deploy powerful systems, and can scale harms beyond an individual’s reach. During the Cuban Missile Crisis, Soviet officer Vassili Arkhipov refused to authorise a nuclear launch. His intervention illustrates why speed and calculation cannot replace judgment about what is at stake.
AI enters a three-way contest. The question is not simply humans versus machines, but how states and corporations shape AI and which interests govern it. The book contrasts American competition between firms, China’s stronger state direction of technology companies, and the EU’s rules-based regulatory approach. Each model has risks. The essential safeguard is meaningful human authority over decisions involving war, surveillance, or human survival, rather than merely allowing a person to approve a machine’s recommendation.
A human veto is valuable only if the person exercising it has time, information, and real authority. Otherwise, oversight becomes ceremonial, while responsibility is still shifted onto the operator after the system has set the terms. This connects to safety engineering, where reliable control depends on independent checks and the ability to stop a process. The book’s emphasis on human judgment is persuasive, but it also raises a democratic question: whose judgment counts, and how can affected people contest decisions made in their name?
Analysis
The book’s most useful move is to shift attention from artificial intelligence alone to artificial agency. A language model may generate fluent answers, but states and corporations already turn human choices into durable, enforceable consequences. This makes the familiar question, “Will machines become more intelligent than people?” only part of the problem. A prior question is who gives machines authority, embeds them in institutions, and benefits when their outputs become decisions.
The distinction between agency and intelligence also explains why the book’s historical comparison matters. Modern states and corporations did not simply make life more efficient. They enabled scale, continuity, and replication, helping create prosperity while making organised violence and environmental damage possible on a comparable scale. The same durable machinery that sustains public debt can sustain war. The same organisations that spread useful technology can concentrate wealth and power. There is no clean division between liberation and dependence.
The framework is especially strong as a tool for locating responsibility. It resists both anthropomorphism, which treats an institution as if it had a conscience, and reductionism, which pretends that only individual choices matter. Its main limitation is that the machine analogy can flatten differences among organisations. A democratic state, a company, and an algorithm do not possess the same powers, constraints, or routes to accountability. Describing all of them as artificial agents is illuminating only when those differences remain explicit.
The resulting political challenge is not to choose the human over the artificial in the abstract. Modern life already relies on artificial institutions, and AI will increasingly operate through them. The sharper task is to design decision processes that preserve contestability: people must be able to see who set the rules, challenge the grounds of a decision, and identify who carries its costs. Better answers are valuable, but they cannot decide what society should value. That remains a political question, even when the machinery makes it tempting to outsource it.
Review Summary
The Handover explores the transfer of control from individuals to larger entities like governments, corporations, and AI. Readers find it thought-provoking, offering unique perspectives on states and corporations as artificial agents. Some praise Runciman's writing style and analysis, while others find it dense and repetitive. The book challenges readers to consider the balance between collective power and individual autonomy in our evolving world. Despite mixed opinions on its execution, many agree it raises important questions about our future relationship with AI and existing power structures.
People Also Read
Glossary
Artificial agency (AA)
Organised power to actThe capacity of a human-made organisation to act in the world and sustain the consequences of its actions. States and corporations depend on people for decisions and implementation, but their structures allow actions to bind others and persist beyond individual members. Artificial agency is distinct from artificial intelligence: an institution can act powerfully without thinking for itself.
Artificial general agency (AGA)
Broad scope for actionRunciman’s term for the state’s ability to act across a wide range of purposes, rather than being confined to one particular remit. A state can decide whether to pursue security, growth, war, or other goals, though its choices still depend on human intelligence. The term distinguishes the state’s broad agency from a corporation’s more specialised purpose and an AI’s assigned task.
superagency
Consequences beyond individualsThe unusual power of states and corporations to turn human decisions into consequences on a scale no individual could achieve alone. Their advantage is not superior reasoning. It comes from durable organisation, authority, resources, and the ability to coordinate many people. Superagency helps explain why an institution can be powerful, consequential, and still lack consciousness or a human conscience.
artificial person
Institutional identity over timeA human-made entity treated as a person for purposes of action, representation, or law. In Runciman’s account, states and corporations are not simply masks for their current leaders or members. Their separate identities let decisions, debts, rights, and obligations persist as people come and go. This durable identity is useful, but can also make responsibility harder to assign.
discursive dilemma
Group views can conflictA problem in collective decision-making where the group’s verdict can differ from the verdict implied by the majority’s views on separate criteria. The result depends partly on how questions are framed and aggregated. A group can reach a coherent overall decision even when its members’ reasoning, considered item by item, points elsewhere, or can produce conflicting conclusions from the same individual judgments.
Condorcet’s Jury Theorem
Larger groups can know moreA result in social choice theory stating that, under particular conditions, a larger group is more likely than a smaller one to select the correct answer. Members must decide independently, and each must have a better-than-even chance of being right. The theorem does not show that large groups are always wise; dependence, shared bias, or poorly framed questions can undermine its assumptions.
First Singularity
Transformation by institutionsRunciman’s name for the historical transformation driven by modern states and corporations. These human-made organisations amplified human capacity through durable, scalable forms, helping reshape economic life, health, and social possibilities. The term challenges the idea that a future AI transformation would be humanity’s first encounter with a profound shift in its condition.
Second Singularity
Possible transformation through AIA possible future transformation in which artificial intelligence changes the human condition as deeply as modern states and corporations did. The outcome is not predetermined: AI could extend human capabilities or contribute to human loss of control. Runciman argues that its effects depend heavily on how AI interacts with the existing artificial agents that build, regulate, and deploy it.
The Narrow Corridor
Balanced state and societyA model of political development in which a capable state and a strong society constrain one another. A state powerful enough to maintain order can also oppress; a society strong enough to resist oppression can also undermine public authority. Liberty and prosperity are more likely when neither side dominates, a balance that must be actively maintained.
Malthusian trap
Population outpaces foodA recurring condition in which population growth exceeds the growth of food production. Periods of plenty encourage population to rise, but scarcity then brings starvation, disease, violence, and higher death rates. Runciman uses the concept to describe the historical limits on population growth and to warn that systems built around relentless expansion can encounter different forms of constraint.
Leviacene
Planet shaped by artificial agentsRunciman’s alternative name for the present planetary era, shifting emphasis from human activity in general to the states and corporations that organise activity at immense scale. The term highlights how institutions mediate humanity’s impact on the natural world. It suggests that environmental damage is not simply the product of individual human nature, but also of artificial systems with powerful growth incentives.
Download PDF
Download EPUB
.epub digital book format is ideal for reading ebooks on phones, tablets, and e-readers.