The cases where choices matter
These are the worlds in which the argument fails partway and leaves something worth surviving. For each: what it looks like, what bears on the outcome for an individual, and what does not help. The list is not exhaustive and the cases overlap.
Gradual disempowerment
Capabilities keep improving, no single system takes over, and humans are steadily removed from the loop: first from tasks, then from decisions, then from the economy that decided things. Nobody dies. People become, over a decade or two, economically superfluous and politically weightless, the way horses did. The book discusses this as a way station in What if AI is developed only slowly, and it slowly integrates with society? and dismisses it as a stable end state; whether or not it is stable, it is a state people would live in for a while.
What bears: ownership. In a world where labour loses value, income from capital does not, at least not first. Being on the side of the tools rather than replaced by them: the people who use the systems well keep relevance longest. Political voice while it still counts, since the window for shaping rules closes as the rule-makers stop needing votes. What does not help: skills that compete with the systems on their own terms.
Misuse
Powerful AI in human hands, used against other humans. Surveillance states that actually work. Fraud and manipulation at a scale and precision that overwhelms ordinary defences. Engineered pathogens. Autonomous weapons. The systems are aligned, in the sense that they do what their operators want; the operators are the problem.
What bears: where you live, more than anything. Institutions that constrain power, functioning public health, a state that is neither failed nor total. Low exposure: a small digital footprint, assets that are not purely electronic, relationships that do not route through platforms. What does not help: individual technical countermeasures against state-scale capability.
Concentration
A small group, a company or a government, controls the most capable systems and does not share the proceeds. This is the world the book's own argument implies if alignment turns out to be solvable by whoever gets there first, and it is the one case where the survive and avoid questions merge, since the only durable answer is political and collective. What bears for an individual: being inside the group, which is not advice, or being somewhere the group's reach is limited. Mostly this case is an argument for the Avoid section.
Dislocation
The most likely near-term case and the most ordinary. Large categories of work lose value quickly, faster than people can retrain or institutions can adapt. This is survivable by definition and still ruins lives.
What bears: low fixed costs, which convert a shock into an inconvenience. Savings. Skills that complement the systems: judgement, relationships, physical work, anything where being a person is the point. Being early to use the tools rather than late. Community, in the plain sense of people who would help you and whom you would help. What does not help: betting a career on a task the systems are visibly approaching.
Partial catastrophe
An AI-enabled disaster that is regional or recoverable: a pandemic, a market collapse, an infrastructure attack, a war with autonomous weapons. Here the classical resilience playbook applies without modification. Distance from dense targets, basic supplies, redundancy in the things you depend on, relationships with neighbours, a plan for the first weeks. This is the only case where the standard prepper instincts are correct, and it is worth being clear that they are correct only here.
What runs through all of them
The same few things keep appearing. Location and the quality of the institutions around you. Ownership rather than wages. Low fixed costs. Skills where being a person is the point. Other people. Early and competent use of the tools. None of these is specific to AI; they are what has always mattered in periods of fast change, and the argument for them here is that this is going to be one, in every world the argument allows except the one in which nothing matters.
The book does not discuss any of this, because its thesis is that these worlds are not where we end up. On its own terms that is consistent. But a person allocating their own effort should notice that these are the worlds in which effort pays, and weight them accordingly, which is the subject of the next page.