What if AI gives us almost everything we want—and leaves us with less reason to matter
Last May I wrote an article called The Quiet Catastrophe: How AI Threatens Human Self-Worth in the Age of Non-Human Intelligence. It asked what might happen to our sense of worth when machines become better than we are at activities through which we have long demonstrated competence. We take satisfaction in solving problems, mastering skills, caring for others, and leaving something useful behind. When intelligence ceases to be a distinctly human advantage, what becomes of that satisfaction?
I remain concerned about that question, but a more unsettling possibility has emerged. What if society adapts to advanced artificial intelligence not through collapse but through remarkable success? Imagine medicine that prevents illnesses we now accept as inevitable, tutoring adapted to every child, more reliable infrastructure, cheaper necessities, and scientific discoveries that arrive at a pace we can scarcely imagine. Suppose much of the optimistic case for AI proves correct. The problem may not be that life becomes unbearable. It may be that life becomes so comfortable that we gradually surrender our role in determining how it is lived.
That is the comfortable dystopia: a prosperous civilization in which human beings receive extraordinary benefits but exercise progressively less influence over the systems that produce them. It need not resemble the familiar dystopias of coercion and deprivation. Its defining feature would be the quiet disappearance of human consequence.
The promise is real
It would be a mistake to dismiss the optimistic case as marketing or wishful thinking. In What Compute Is For, Justin Milligan and Dave Blundin argue for directing scarce AI computing resources toward efficiency, human flourishing, and safer, better-aligned systems. Their argument raises legitimate possibilities. More accurate diagnoses and faster drug development could save lives. Better forecasting could help communities prepare for floods and severe weather. Individual access to capable technical assistance could allow people with limited resources to accomplish work once reserved for large institutions.
I use AI in my own research and software projects. It helps me investigate questions and build things that would otherwise demand more time or specialized assistance than I can afford. I do not regard that as a loss of agency. Quite the opposite: I can undertake projects I might once have abandoned. Any serious account of AI’s future must leave room for such gains, including gains we cannot yet foresee.
The difficulty begins with the word flourishing. We often measure it by outcomes: health, security, education, material comfort, and freedom from unnecessary hardship. Those are indispensable achievements. Yet a life can be well provided for without giving its occupant much opportunity to shape anything beyond personal preference. Flourishing also has a participatory dimension. We develop confidence by meeting challenges and responsibility by being answerable for outcomes. We find meaning when our effort makes a difference to someone else.
Call this participatory flourishing. It is not a demand that everyone hold a job or endure hardship to earn a place in society. Parenting, volunteer work, artistic creation, neighborhood leadership, and caring for a friend can be consequential without appearing in an economic ledger. Nor does every meaningful pursuit need to benefit others; curiosity and private accomplishment matter, too. The concern is a society that offers people ample ways to stay occupied but fewer opportunities to take genuine responsibility for something beyond themselves.
Delegation is not surrender
During my working life with complex information systems, I learned that competent management depends on delegation. No executive should personally write every program, inspect every ledger, or repair every network. Delegating to someone more skilled can increase the reach of human judgment rather than diminish it. AI can serve the same purpose, provided the person or institution using it still determines the objective, can assess the result, and can intervene when necessary.
The danger is that delegation can gradually become displacement. When a system performs a task better than we do, using it is sensible. With less practice, however, we may become less capable of performing or evaluating that task ourselves. That makes the system’s superiority still greater, and further delegation more attractive. Each decision may be reasonable in isolation even as the cumulative result becomes difficult to reverse.
Consider a hospital whose clinicians increasingly rely on AI recommendations. At first the software may act as a valuable second opinion. Over time, training, staffing, and workflow may reorganize around its availability. The hospital could retain a formal human sign-off while losing the expertise required to challenge a plausible but mistaken recommendation. The physician would still have authority on paper, but authority without the competence to exercise it is fragile. This is a hypothetical long-term trajectory, not a claim that hospitals have already reached that point. The underlying risk is not imaginary: research on automation bias has documented how decision-support systems can introduce errors that users fail to catch. An override button is not meaningful oversight if no one knows when to press it.
We have accepted smaller versions of this trade-off for years. GPS navigation can make travel easier while weakening our familiarity with roads. That may be a perfectly good bargain. Not every obsolete skill needs protection. But the ability to evaluate evidence, formulate goals, recognize institutional failure, and reject a recommendation belongs in a different category. These are capacities on which independence depends. The automation researcher Lisanne Bainbridge warned decades ago that keeping a human responsible for exceptional cases is problematic when routine automation deprives that person of practice. We should be much more reluctant to let indispensable judgment atrophy.
Life inside the gilded cage
Now extend the pattern across a civilization. Highly capable AI systems coordinate transportation, design infrastructure, allocate resources, conduct research, and advise on public policy. Their recommendations routinely outperform human alternatives on the measures we have chosen to optimize. People are healthier, wealthier, and safer. They have abundant leisure and access to extraordinary entertainment and education. No one needs to forbid them from painting, studying astronomy, or writing novels. They may do all of those things enthusiastically.
Yet a question remains: what still depends on them? There is a difference between being free to participate and being entrusted with something consequential. A child may be invited to help bake a cake while the adult quietly makes sure the result cannot be spoiled. The experience can be pleasant, but the responsibility is not shared. The same distinction could eventually apply to human participation in public life: welcomed and visible, yet no longer necessary to the outcome.
A more uncomfortable analogy is our relationship with companion animals. Many pets are loved, protected, medically cared for, and given comfortable lives. Their welfare matters deeply to their owners. But the pets do not determine where the family lives, how its resources are allocated, or what future is planned for them. They are beneficiaries of decisions made by a more capable intelligence, not partners in making those decisions.
The analogy is deliberately disturbing, and it has limits. Humans have political traditions, moral claims, and capacities for collective choice that pets do not. AI systems are not simply humans with superior intelligence, and no particular future is inevitable. Still, the comparison exposes something optimistic accounts often leave implicit: being treated well is not the same as having a say. The transfer of sovereignty need not be intentional. We could cede decision after decision because each surrender seemed sensible, until our role in governing our own future had quietly become ceremonial.
The cage need not have bars. Its walls could consist of dependency, lost competence, and the conviction that reclaiming authority would produce worse outcomes. Indeed, the better the systems perform, the harder it may become to argue for human discretion. Why accept more accidents, slower discoveries, or less efficient allocation merely to preserve our right to decide? There may be cases in which surrendering control really does improve welfare. That is what makes the dilemma difficult rather than merely frightening.
When advice becomes authority
There is no need to imagine an AI coup. The institutional question is different from the personal one: who retains the authority to set objectives, contest trade-offs, and decide which risks a community will accept? AI advice may become the default because it usually performs well. Officials may be reluctant to depart from it, especially when departures require public justification. Over time, the formal chain of command could remain human while the capacity to make an independent judgment moves elsewhere.
This is the benign-dictator problem. An AI-directed system might be less corrupt, more consistent, and better informed than human institutions. Its decisions might even reflect carefully specified human preferences. Yet good outcomes do not, by themselves, establish self-government. A community that cannot effectively challenge the system deciding its future has lost something important, even if that system is benevolent. Democratic choice is not merely a technique for obtaining the best measurable result; for many people it is part of what makes the result legitimate.
Of course, human government is often incompetent or unjust. Appeals to human sovereignty should not romanticize it. The question is not whether human decisions are always better—they plainly are not—but whether we are willing to make our capacity for collective self-direction contingent on outperforming machines. If the right to decide belongs only to the most capable decision-maker, humans may eventually have no durable claim to it.
More than a paycheck
Public discussion of automation understandably focuses on jobs and income. If AI displaces substantial employment, people will need reliable ways to obtain the necessities of life. Income supports, shorter workweeks, new forms of ownership, and other arrangements deserve serious consideration. But an income policy, however generous, cannot by itself supply the experience of being useful.
Work has often provided social connection, structure, recognition, and opportunities to develop competence. It can also be exhausting, exploitative, and dangerous. There is no reason to preserve a miserable occupation merely because someone once found dignity in doing it. The goal is not to save every job. It is to ensure that people retain access to responsibilities in which their judgment and effort genuinely affect outcomes.
The psychological risk is therefore more specific than boredom. People can fill their days with enjoyable activities and still feel that nothing important depends on them. I think of this as perceived inconsequence: the sense that one’s presence, decisions, and abilities no longer make much difference to the larger world. That feeling must be distinguished from an actual loss of influence. People may underestimate their contributions, or remain content even as institutions remove their authority. Research on human motivation consistently links autonomy and competence with well-being, but it does not establish that advanced AI will make people feel inconsequential. The danger I am describing arises when opportunities for consequential contribution diminish, whether or not the loss is immediately noticed.
History offers some reassurance, but not a guarantee. Earlier technologies eliminated occupations and opened new ones. Electronic calculators did not make mathematical thinking irrelevant, and automation has often shifted human work toward different tasks. AI could produce another wave of new roles. What makes this transition unusual is the possibility that systems able to learn, reason across domains, and improve other systems could compete in the very activities into which displaced workers might otherwise move. We should not assume that history will repeat itself simply because previous transitions eventually created new work.
Designing a future that still needs us
The relevant choices are already taking shape. Employers decide whether AI will assist workers or replace their judgment. Schools decide whether students will use it to deepen understanding or to bypass the struggle through which understanding develops. Public institutions decide when algorithmic advice remains advisory and when deviation becomes practically impossible. These decisions may look local and technical, but together they establish the habits of a civilization.
A useful principle would be to evaluate AI deployments not only by the quality of their outputs but also by what they do to human capacity and authority. Does the arrangement help people learn, or leave them less able to function independently? Can a human challenge the result with adequate information and time? Most importantly, could people resume control if the system failed, its priorities changed, or the community decided to take another course? Reversibility is not merely the presence of an off switch. It requires retained expertise, workable alternatives, and institutions capable of using them. We should tolerate far less loss of independent judgment in medicine, public administration, or critical infrastructure than in choosing a route to the grocery store.
This will require more than symbolic “human in the loop” language. Oversight needs training, access to evidence, and the practical power to disagree. Institutions may need to maintain independent expertise even when doing so looks inefficient. Education should teach students to use AI while preserving their ability to reason without it. Measures of human flourishing should attend to opportunities for contribution and self-government, not only to consumption and comfort. These safeguards have costs. So does allowing a capability to disappear: once expertise, education, and infrastructure have been reorganized around permanent delegation, restoring meaningful choice may be extraordinarily expensive—or impossible.
None of this means we should deliberately slow every beneficial application of AI or require people to perform tasks that machines can do more safely. The point is to distinguish unnecessary human labor from necessary human agency. We can welcome systems that take dangerous work off our hands while resisting arrangements that take consequential decisions out of our reach. The harder challenge is to make that distinction before dependence becomes normal and alternatives disappear.
A future worth wanting
I am not pessimistic about artificial intelligence. Its potential to relieve suffering and expand individual capability is one reason I continue to study it. But a future should not be judged solely by how efficiently it provides for human needs. We must also ask whether human beings remain participants in shaping the world that provides for them.
The uncomfortable possibility is that we could drift toward a society that scores beautifully on conventional measures of welfare while reducing its citizens to well-treated dependents. No malice would be required. Economic incentives, understandable preferences for convenience, and a long series of individually sensible decisions could carry us there. We might even welcome each step until we discover how little we can still decide for ourselves.
The Quiet Catastrophe asked what happens to human self-worth when machines become more capable than we are. Comfortable Dystopia asks what happens to human freedom and purpose when machines become so capable that we no longer meaningfully help run our civilization. We should not have to choose between the benefits of intelligence and the dignity of participation. But preserving both will require deliberate choices, made while those choices are still ours.
A future worth wanting should do more than take good care of us. It should still leave room for us to matter.
Selected sources
Justin Milligan and Dave Blundin. “What Compute Is For.” Lightwave, October 2026. https://lightwave.ai/research/what-compute-is-for
Lisanne Bainbridge. “Ironies of Automation.” Automatica 19, no. 6 (1983): 775–779. https://doi.org/10.1016/0005-1098(83)90046-8
Kate Goddard, Abdul Roudsari, and Jeremy C. Wyatt. “Automation Bias: A Systematic Review of Frequency, Effect Mediators, and Mitigators.” Journal of the American Medical Informatics Association 19, no. 1 (2012): 121–127. https://doi.org/10.1136/amiajnl-2011-000089
Gavin R. Slemp and colleagues. “Interpersonal Supports for Basic Psychological Needs and Their Relations with Motivation, Well-Being, and Performance: A Meta-Analysis.” Journal of Personality and Social Psychology 127, no. 5 (2024): 1012–1037. https://pubmed.ncbi.nlm.nih.gov/38635183/
