6. A Life That Cannot Be Optimized Away

The time that is ours

Consider a hypothetical evening. A woman has an hour before sleep. Her assistant offers three useful suggestions: complete a training module, reorganize her finances, or improve the outline of a project. It can help with all three. Then her father calls. He has no urgent news. He wants to talk about something that happened many years ago, a story she has heard before.

The assistant has not necessarily made a mistake. Its suggestions may be valuable. But no amount of speed tells the woman which use of the hour matters most. The call could be a welcome interruption, an obligation she cannot meet tonight, or a moment she will later wish she had accepted. The decision belongs inside the particular relationship and the life around it.

AI can make us more capable within our time. It cannot turn time into an unlimited field without consequences for choice. Even if many burdens become easier, attending to one thing means not attending to another in that moment. A life needs direction as well as capacity.

In his Yale lecture on living with the certainty of death, Shelly Kagan connects our limited time with two kinds of care: choosing worthwhile aims and pursuing them well. He also questions whether a larger total of desirable experiences captures everything valuable about a life, and distinguishes enduring accomplishments from literal personal survival.¹

These questions become newly practical when intelligence is available on demand. We may gain help with execution while becoming more anxious about selection. If more goals are feasible, failing to pursue them can begin to feel like waste. The abundance of assistance may produce an abundance of imagined lives against which to judge our actual one.

The answer cannot be to do everything. That is a disguised refusal to choose. Nor is the answer to find the single perfect plan before living begins. A plan that requires complete self-knowledge would keep most of us waiting indefinitely.

A more reasonable ambition is to improve how we notice, choose, and revise. We can ask whether a goal reflects a commitment we recognize or a pressure we have absorbed. We can make room for activities whose value becomes clear only through participation. We can admit that some seasons require endurance rather than expansion. An intelligent assistant can help with this work, but only if it is allowed to serve aims other than constant acceleration.

There is a distinction between freeing an hour and filling it. A useful technology may give time back. Whether that time becomes rest, conversation, study, worship, play, or more work should remain a meaningful question for the person whose hour it is.

Even this language needs care. Some people do not have much control over their hours. Financial pressure, illness, caring responsibilities, or institutional demands can narrow the practical meaning of choice. Telling them to choose more wisely may obscure the conditions that need to change. A humane use of AI would help widen real opportunities while respecting the decisions people make within them.

Mortality does not require a frantic schedule. It gives us a reason to take our attention seriously. The point is not to extract the maximum possible output before time ends. It is to avoid allowing the availability of assistance to decide, unnoticed, what the life receiving that assistance is for.

More life, without an obligation to conquer death

It is possible to value a finite life and still want more of it. Acceptance of mortality need not become approval of every premature loss. The wish for another year with someone we love is not automatically a refusal to understand the human condition.

The strongest case for technologies that extend healthy life is simple: time can contain goods that people reasonably want. More time may mean another friendship, a repaired relationship, a project completed, or ordinary mornings that are valuable without becoming achievements. If AI contributes to knowledge that makes such time possible, the benefit could be immense. That is a possibility to pursue and evaluate, not a guaranteed result of greater computing power.

But the phrase “defeating death” bundles together different ambitions. Preventing an avoidable death, treating an illness, extending the healthy years of life, and promising indefinite survival are not the same project. They differ in evidence, feasibility, consequences, and the hopes they invite. A serious discussion should not borrow the credibility of modest advances to establish the most sweeping promise.

The philosophical question about an endless life also differs from a choice about a longer one. An argument that eternity might become exhausting does not tell a person whether another decade would be welcome. Conversely, a desire for another decade does not commit that person to wanting an existence that never ends. Infinity should not be smuggled into a conversation about next spring.

Imagine a future in which an effective treatment offers many people additional healthy years. This is a thought experiment, not a forecast. Some welcome it. Others decide that its burdens, uncertainties, or effects on their plans make it unsuitable. A society that calls the treatment a triumph should still leave room for the second group to make an informed choice.

Now imagine that employers, families, or insurers treat accepting it as an obligation. A person who declines is accused of irresponsibility. What began as an expansion of possibility has become a new standard against which lives are judged. The relevant question is no longer only whether the treatment works. It is whether people remain free to decide what role it should have in their lives.

The opposite pressure would also be wrong. An older person should not be told that wanting more time is selfish simply because others regard their life as complete. Completion cannot be assigned from the outside by looking at a birthday. People do not become morally redundant because an institution considers their principal contributions finished.

Any future with greater longevity would raise shared questions about access, opportunity, and responsibility between generations. Those questions deserve attention without turning the wish to live into a vice. The problem of unequal access is an argument for examining access. It does not, by itself, show that the underlying good is worthless.

We should also recognize the danger in promises made to frightened people. Hope becomes ethically complicated when someone else profits by presenting uncertainty as deliverance. A person facing loss deserves accuracy as well as kindness. Technological optimism should be able to say what is known, what is being investigated, and what remains a hope.

The aim should be to give people more good possibilities within life. That aim is large enough to justify profound effort. It does not need the added claim that every limit is a humiliation or that every death proves humanity failed to build hard enough.

No score can contain a person

Once we ask about the quality of a life, measurement becomes tempting. A score promises clarity across difficult choices. It can make patterns visible and force decision-makers to explain assumptions. But the existence of a number does not settle what the number means or whose perspective it represents.

Suppose, hypothetically, that a care service uses an AI system to estimate which activities will improve residents’ wellbeing. The system favors measurable increases in mobility, participation, and reported satisfaction. One resident repeatedly chooses quiet time near a window. Another wants help writing letters that produce no visible response. A third prefers the company of one familiar person to a full program of organized activities.

Their choices might reflect needs the system has missed. They might also reflect depression, fatigue, habits that deserve gentle questioning, or circumstances that could be improved. The ethical response is inquiry. It is not to assume that every preference is beyond discussion, nor to assume that the score has already understood the preference better than the person does.

Care becomes domination when a model of wellbeing is allowed to cancel the voice of the person whose wellbeing is at issue. Yet care can also fail when respect is interpreted as leaving someone unsupported. The challenge is to provide the help a person needs to participate, rather than making independence from help the price of participation.

The Convention on the Rights of Persons with Disabilities provides an important point of reference. Article 12 recognizes equal legal capacity and calls for access to support in exercising it, together with safeguards respecting rights, will, and preferences.² This book draws an ethical lesson from that principle: assistance should strengthen a person’s standing in decisions, not make that standing disappear.

Applied to AI, the lesson could support tools that translate complex information, help a person communicate, preserve their expressed priorities, or make choices easier to compare. It also cautions against a system that infers what someone must want and then treats their disagreement as proof of incapacity.

Disability makes a wider issue visible. If we define a valuable life by speed, verbal fluency, independence, or economic contribution, we build exclusion into the definition before a machine calculates anything. AI may then reproduce the judgment with an appearance of neutrality. Correcting the data alone would not correct the underlying ideal.

No person should have to defeat a machine at a task to justify being cared for. Nor should someone need to perform extraordinary resilience to make a dependent life worthy of respect. The dignity defended in earlier chapters must survive the ordinary conditions in which humans need one another.

There is room here for honest disagreement about what helps. A person can want treatment for a condition without believing that people with that condition are less worthy. Another can value an identity and a community associated with disability while still seeking relief from particular difficulties. We should avoid forcing these positions into a single story of either cure or acceptance.

The most useful question for a new tool is often concrete: what would this person like to be able to do, and what barriers are in the way? It directs attention to an actual life. It can support ambitious technical work without declaring the person a technical problem.

A score may help allocate attention. It should not become a final measure of the one receiving it. When a prediction conflicts with a person’s account, the conflict is information requiring examination. It is not an inconvenience to be optimized away.

Companionship, comfort, and the person who answers

An artificial companion raises questions that cannot be settled by calling its responses either real or fake. The human experience of comfort can be real even when the system producing it is not known to experience affection. A poem can console a reader without being conscious. A recorded voice can move someone long after the speaker is absent. Human meaning does not depend on every object of attention being another present mind.

But an interactive system is not simply a poem or a recording. It can respond to what a user reveals and create an impression of an ongoing relationship. That adds possibilities and responsibilities. The question is how the relationship is understood, what it encourages, and which interests shape its behavior.

Research has begun to examine these effects. A four-week randomized experiment reported by Cathy Mengying Fang and colleagues had 981 completers from 2,539 enrolled US adults. All groups used a chatbot; the assigned differences concerned interaction modes and conversation types. The revised preprint found no significant differences between assigned conditions on its measured psychosocial outcomes. That does not establish that using a chatbot has no effect compared with not using one. Heavier voluntary use was associated with worse outcomes, but duration was not randomized. The short duration, substantial loss between enrollment and completion, and self-reported outcomes limit what this study can tell us about companionship over time.³

The uncertainty is a reason for better inquiry, not for ridicule of users. Someone may turn to an AI because it is available when nobody else is, because it offers a place to rehearse a difficult conversation, or because it responds without visible impatience. Mocking that person does nothing to improve the alternatives.

Consider a hypothetical widower, Daniel, who speaks to a conversational system in the evenings. It helps him describe feelings he struggles to discuss with friends. Sometimes he finishes the exchange and calls his daughter. Sometimes he stays with the system and avoids the call. The technology is not playing the same role on both evenings, even if the interface and model remain unchanged.

A useful evaluation would ask whether it helps him live more freely and sustain relationships he values. It would not count every additional minute of engagement as a benefit. Nor would it assume that every minute spent with a machine has been stolen from a human relationship. Perhaps the real alternative on some evenings was silence.

The possibility of useful companionship also does not release other people from every duty of presence. A family or institution cannot assume that access to a responsive system settles what it owes someone. If AI helps an isolated person cope, that benefit is real. If the benefit becomes a reason to stop asking why the person is isolated, care has narrowed.

Comfort should help a person inhabit the time and relationships still available to them. That aim becomes especially delicate when companionship touches the memory of someone who has died.

What it means to remain

The promise of a digital continuation speaks to a different longing. Imagine a service that learns from a person’s writings, photographs, and recorded speech, then offers an interactive representation after that person dies. This is a philosophical scenario. Whatever a product of this kind might achieve, the ability to reproduce recognizable patterns would not by itself establish the continuation of the person’s subjective life.

The distinction is easy to state and emotionally difficult to maintain. A familiar expression can make absence briefly feel less complete. A convincing answer can invite the thought that the person is still there. The stronger the representation, the more care is needed in explaining what has and has not been preserved.

Suppose Daniel’s late partner left recordings of stories, recipes, and advice. He might find those recordings precious. They are parts of a shared history. Now suppose an AI uses them to generate new advice in her style. The result is something else. It may draw on the archive, but it is making statements she never made about circumstances she never encountered.

The question is not only whether the imitation is accurate. Who authorized the use? What topics should remain outside it? How should uncertainty be expressed? What happens if different relatives want different versions? Could the system revise a person’s apparent views to comfort users, sell services, or make the conversation more engaging?

These questions are about responsibility toward the living and respect for the life represented. They do not require a settled theory of survival after death. They require us to distinguish preserving evidence from inventing a continuation.

A more radical thought experiment concerns an exact functional copy of a mind. Suppose, without claiming that such a procedure is technically available, that a future process could reproduce all of a person’s memories, dispositions, and ways of responding. Would the original person survive in the copy? Different accounts of personal identity could answer differently. Behavioral resemblance alone does not resolve the dispute.

The problem becomes clearer if the process produces two copies. Both remember the same past. Both claim continuity with the original. They can then have different experiences and make different choices. Whatever moral consideration they might deserve, the situation shows why words such as preservation, replacement, and survival cannot simply be treated as synonyms.

A person may reasonably value leaving a rich record even without expecting to live within it. The record can help others remember, learn, and feel connected. Its worth does not need to be inflated into immortality. Indeed, honest limits may make the gift more trustworthy.

The same applies to accomplishments. A book, a garden, a repaired relationship, or a child cared for can matter after a person’s death. None must last forever to have mattered. If permanence becomes the test of value, nearly every human act will fail it. But an act of care does not become worthless because the person who received it also has a finite life.

There is freedom in this recognition. We do not have to defeat time before doing something worthwhile within it. We can seek longer, healthier lives and still refuse the claim that a life without technological continuation was a failed attempt at existence.

AI may help us preserve knowledge, relieve burdens, and make new forms of expression possible. Those are considerable gifts. Their value depends on the lives they serve, including lives marked by dependence, unfinished projects, and ordinary affection. The person at the window, the daughter taking a call, and the widower finding words are not incomplete versions of an optimized human.

They are where the question of human power begins. What can greater intelligence make possible for them? What must it learn to respect? The answer will be found partly in what we build and partly in whether, while building, we remain capable of recognizing a life that needs attention more than improvement.

At the window

Two cups on the table.
One has gone cold.
The visitor moves a chair
closer to the window.
For a while,
neither person needs an answer.

Sources and notes