“No man is an island, entire of itself.”

— John Donne, Meditation XVII (1624)

Writing during a period of serious illness, Donne reflected on the fundamental interdependence of human existence. His observation reminds us that our lives acquire much of their meaning through connections with other people, even when those connections are inconvenient, demanding, or difficult to sustain.

“The only way to have a friend is to be one.”

— Ralph Waldo Emerson, Friendship (1841)

Emerson recognized that friendship is not simply something we receive from another person. It requires participation, generosity, patience, and the willingness to give something of ourselves without knowing precisely what we will receive in return.

“Artificial intelligence can make life easier by removing the effort from almost everything we do. But what if some of that effort is precisely what makes life worth living?”

— Aditya Mohan, Founder, CEO & Philosopher-Scientist, Robometrics® Machines

Introduction: The Dinner That Almost Didn't Happen

It is Friday evening in San Francisco. A woman leaves her office after a long week, her coat folded over one arm, her thoughts already drifting toward the quiet comfort of home. She is tired, hungry, and in no mood for conversation. Her personal AI agent knows this. It has observed her schedule, learned her preferences, and remembers that she usually orders Japanese food on Fridays. Before she reaches the street, it offers to arrange dinner, replenish the groceries she has been meaning to buy, and select a birthday present for an old friend. A few words of approval, and everything is arranged. No telephone calls, no waiting, no awkward exchanges, no decisions left unresolved. By the time she arrives home, dinner is on its way. The agent has performed exactly as intended: it has made her life easier.

But imagine that the same woman had stopped at the little neighborhood grocery instead. She might have discovered that the owner had returned after several weeks away, having traveled to see his elderly father. She might have asked how he was doing, and the answer might have turned into a conversation about ageing parents, distant families, and the peculiar loneliness of living thousands of miles from where one grew up. Perhaps she would have bought a cabbage she had not intended to buy, accepted a recipe scribbled on the back of a receipt, and called her own father while walking home. Dinner would have taken longer. She would have spent more energy, encountered someone else's worries, and arrived home later than planned. Yet the evening might have contained something her personal agent could never have identified as an explicit objective: an unexpected reminder of someone she loved.

Neither evening is inherently superior. Sometimes we genuinely need an effortless dinner and an uninterrupted evening. But the contrast reveals a question that may become increasingly important as personal AI agents take responsibility for more of our daily activities. When intelligence removes the friction from ordinary life, does it merely eliminate inconvenience, or can it also eliminate the unpredictable human encounters from which relationships, memories, and meaning emerge?

That question brings us to an unlikely comparison: cabbage and potato chips.

I. The Chips: The Irresistible Convenience of Personal AI

Potato chips are wonderfully convenient. Open the bag, reach inside, and the reward is immediate. There is no preparation, no washing, no cooking, and almost no effort between desire and satisfaction. Their attraction lies partly in the removal of the small inconveniences that separate wanting something from having it. Personal AI agents offer a similar promise for intellectual and practical work. Need to find a restaurant? Ask the agent. Need to compare insurance policies, negotiate a subscription, purchase a birthday present, prepare a presentation, or resolve a complicated travel itinerary? Give it an objective, define the necessary permissions, and allow it to perform much of the work. A sufficiently capable agent can coordinate language models, external tools, persistent memory, search systems, and application interfaces to complete multi-step tasks. The agent may inspect available options, evaluate constraints, formulate a plan, execute authorized actions, and revise its approach when an attempt fails. Sensitive transactions still require suitable safeguards and approval. Nevertheless, activities that once demanded a sequence of conversations, decisions, and interruptions can increasingly be compressed into a few moments of interaction with a machine.

The appeal goes considerably deeper than saving time. Personal agents can reduce cognitive load: the demands placed on working memory, attention, planning, and decision-making. Human interactions require us to interpret intentions, recognize emotional signals, choose appropriate language, negotiate competing preferences, remember social obligations, and manage uncertainty. The agent can absorb much of this burden. It does not need us to ask about its children, remember its birthday, excuse an uncharacteristically impatient remark, or listen patiently when it has had a difficult day. We may choose to interact with it affectionately, but the system does not possess the ordinary human needs and reciprocal claims that make interpersonal relationships demanding. Its conversational behavior can be personalized around our needs and preferences, allowing it to become an extraordinarily accommodating assistant. This is particularly valuable for someone with disabilities, limited mobility, social anxiety, unfamiliarity with a local language, or an overwhelming number of responsibilities. In such circumstances, reducing cognitive and social friction is not indulgence. It can be a meaningful expansion of personal autonomy.

Yet a subtle behavioral change may emerge precisely because the interaction is so pleasant. Consider someone returning home after an argument with a close friend. The friend has been unfair, but perhaps the person has also said something regrettable. Repairing the relationship would require a difficult telephone call, the possibility of rejection, and the humility to acknowledge mistakes. Instead, the person opens an AI conversation. The agent listens attentively, helps organize the facts, offers reassurance, and responds without the emotional complications of an independently involved human being. Used constructively, the conversation might help the person understand the disagreement and prepare an apology. But suppose the person never makes the call. Suppose the emotional relief provided by the agent becomes sufficient, at least for tonight. The immediate discomfort has disappeared, but the relationship remains unrepaired. Repeated often enough, this pattern could turn an instrument of emotional preparation into an instrument of emotional avoidance.

The risk is not that personal AI agents make us comfortable. It is that an intelligence optimized to accommodate us may gradually make the ordinary demands of other people feel less tolerable. The chips are not dangerous because we enjoy them. The question is what happens when their convenience begins to determine the rest of our diet.

II. The Cabbage: Why Human Relationships Cannot Be Delivered Instantly

Cabbage requires a little more work. We select it, remove the outer leaves, wash it, cut it, and decide how to prepare it. The result depends partly on our willingness to participate. Human relationships are infinitely more complicated, but they share an important characteristic: much of their value emerges through participation rather than passive consumption. Imagine visiting an old friend after several months apart. You arrange the meeting, travel across the city, and arrive carrying the accumulated experiences of your separate lives. The conversation begins with ordinary pleasantries over coffee. Your friend mentions a difficulty at work, and you listen. You offer advice, but it is not particularly welcome. There is an awkward silence. Later, you admit something that has been troubling you, and your friend responds with an observation you would rather not hear. Then, unexpectedly, you both remember an absurd incident from twenty years earlier and begin laughing so uncontrollably that the people at the next table turn to look.

That evening cannot be adequately described as an exchange of information. It includes memory, vulnerability, disagreement, humor, affection, and the possibility that each person will leave changed by the encounter. Neither participant exists solely to serve the other. Your friend has independent desires, limited patience, competing obligations, and the freedom to refuse your requests. Sometimes you provide support when you would prefer to receive it. Sometimes you rearrange your day because someone else is frightened, grieving, or simply needs company. Human relationships are reciprocal not because every favor must be repaid, but because each person matters independently of the usefulness they provide.

The distinction is also visible in the physical details of an encounter. A handshake establishes more than contact between two hands. It conveys approach, attention, social recognition, and an intention to participate in the same immediate experience. Face-to-face conversation requires continuous coordination of speech, facial expression, gaze, posture, gesture, and conversational timing. We notice when someone's smile arrives a fraction too late, when their voice becomes uncharacteristically quiet, or when they hesitate before answering a question that should have been easy. These signals do not reveal another person's thoughts with certainty, and their meaning varies across cultures and individuals, but they contribute to how we interpret trust, discomfort, sincerity, and affection. A person may say everything is fine while their trembling hand suggests otherwise. A close friend may recognize the contradiction before either person has consciously decided to discuss it.

A sufficiently advanced embodied AI may eventually recognize many of these signals and respond with extraordinary sophistication. It may maintain detailed social memories, model emotional context, generate natural speech, and coordinate facial expressions or gestures through robotic actuators. Such capabilities could make synthetic companionship remarkably convincing. But the ability to reproduce the observable behavior of a relationship is not identical to possessing the independent life, personal history, mutual vulnerability, and social obligations of another human being. Human relationships are not necessarily healthy merely because they are human; some are neglectful, coercive, or harmful. Nor should people be expected to tolerate pointless inconvenience as a moral virtue. The valuable effort is the effort that makes understanding, trust, care, and reconciliation possible.

We need other people not simply because they answer our questions, but because their existence asks something of us.

III. The Social Displacement Hypothesis: Are Machines Replacing People?

The concern that technology might replace human relationships has been studied for decades. The social displacement hypothesis proposes that time spent using digital media substitutes for time that might otherwise have been spent interacting face-to-face with family and friends, potentially reducing psychological well-being. The argument appears straightforward: every person has twenty-four hours in a day, and time devoted to one activity cannot simultaneously be devoted to another. If someone spends three hours scrolling through social media, those hours are unavailable for other activities. But this observation does not establish which activities have been displaced. The hours might otherwise have been spent watching television, commuting, reading, sleeping, or pursuing another solitary pastime rather than visiting friends.

That distinction is supported by research. In their 2022 review, Social Media Use, Social Displacement, and Well-Being, communication researchers Jeffrey A. Hall and Dong Liu found very little direct or causal evidence that increasing social media use generally replaces face-to-face interaction. They concluded that social media use more likely displaces other media activities. Their analysis also distinguished population-level trends from changes within individuals: a society can experience both rising digital-media consumption and declining face-to-face interaction without one necessarily causing the other. Economic conditions, work arrangements, commuting, household structures, geography, and changing social opportunities may all influence how people spend their time. The history of a social trend is not, by itself, an explanation of its cause.

Nevertheless, the underlying change in American social life is measurable. According to the U.S. Bureau of Labor Statistics' American Time Use Survey, released in June 2026, Americans aged fifteen and older spent an average of approximately 35 minutes per day socializing and communicating in person during 2025, compared with approximately 41 minutes in 2015. The proportion participating in these activities on an average day declined from approximately 38% to 30% over the same decade. The survey's category includes face-to-face social communication and social gatherings; telephone calls, texting, and email are recorded separately. The averages include people who reported no participation that day. These figures therefore describe a particular form of social activity, not the total amount of communication occurring across all media. Importantly, they cannot establish that smartphones, social media, or artificial intelligence caused the decline.

The question for personal AI agents is whether a new and potentially stronger displacement mechanism may emerge. Social media generally connects people through technology, even when those connections are shallow or distracting. A personal AI agent can increasingly complete the interaction without requiring another person at all. Rather than helping us contact a shopkeeper, it can conduct the transaction. Rather than facilitating a discussion with a travel adviser, it can plan the journey. Rather than helping us formulate an uncomfortable question for a colleague, it can sometimes find the answer independently. And rather than merely connecting us to someone who might listen, a conversational AI can itself provide a responsive, personalized exchange.

This does not mean every delegated interaction represents a social loss. Most people do not need another telephone conversation with an automated billing department. An agent that eliminates an hour of administrative frustration may actually create more time for family and friends. But the substitution becomes socially consequential when interactions that once involved genuine human contact are removed, while the time and attention saved are increasingly absorbed by further interaction with the agent. The crucial variable may not be the amount of time spent with AI alone. It may be what the person would otherwise have done, and whether the agent's assistance ultimately strengthens or weakens their connection to other people.

IV. Together, but Somewhere Else: The Displacement of Attention

There is a second form of displacement that does not require anyone to leave the room.

Imagine two people sitting across from one another at dinner. The restaurant is softly lit, the table is set, and their food has just arrived. One begins describing a difficult conversation with a family member. Halfway through the story, the other glances at a phone, reads an incoming message, and offers a distracted response. A few moments later, the phone appears again. The conversation continues, but its emotional rhythm has changed. The person speaking begins to shorten the story, omit details, and eventually say that it was nothing important. Both remain physically present, yet one has become less available to the other.

Researchers refer to related behavior as phubbing, a combination of phone and snubbing, or more broadly as technoference, the interruption of interpersonal interaction by technology. A 2026 systematic review of 55 studies involving 29,925 participants found a consistent statistical association between partner phone use during shared time and lower relationship satisfaction. The researchers also emphasized important methodological limitations: much of the evidence relied on cross-sectional studies and self-reports, making causal conclusions difficult. The findings do not establish that every interruption harms a relationship or that device use is always responsible for dissatisfaction. They do suggest that the quality of shared attention deserves serious consideration.

Now extend that situation into the age of persistent personal AI agents. The device may no longer need a visible screen. It might communicate through a nearly invisible earpiece, a subtle wearable, or an ambient interface that provides recommendations whenever it predicts they might be useful. During a conversation, the agent could identify a person's name, retrieve forgotten details, suggest responses, and quietly remind its user about upcoming obligations. Some of these interventions would be helpful, particularly when supporting memory, accessibility, or communication across languages. Others might compete with the attention required to understand the person standing directly in front of us.

The future danger may not be that we spend too little time in the physical company of other people. It may be that we spend too much of that time mentally accompanied by our machines.

An AI agent could make us exceptionally informed participants in a conversation while simultaneously making us less attentive listeners. We might know more about the person across the table yet notice less of what they are actually trying to tell us.

V. The Easier Companion: What the Early AI Research Reveals

The most consequential transition may occur when personal AI agents move beyond practical assistance into emotional companionship. Consider an elderly man living alone after the death of his wife. Each morning, his AI companion greets him, asks whether he slept well, reminds him of his medication schedule, and discusses the news. It remembers stories about his childhood and recognizes familiar names in his family photographs. When he mentions his late wife, it responds gently and recalls details from earlier conversations. During a difficult night, it remains available without impatience or exhaustion. For someone experiencing profound isolation, this could be a genuinely valuable source of stimulation, practical support, and conversational engagement. It might even help him reconnect with relatives or seek assistance when necessary.

But imagine the same man gradually stops calling his daughter on Sunday afternoons because talking to the agent has become easier. His daughter is often busy, sometimes distracted, and occasionally impatient. She asks him questions he does not want to answer. She has troubles of her own. The AI companion, by contrast, is always ready to listen. He can end a conversation whenever he wishes, return to a favorite subject without embarrassment, and receive a patient response regardless of how often he repeats a story. Nothing dramatic happens. No relationship ends in an argument. The calls simply become less frequent. Eventually, his daughter notices that they have not spoken in several weeks, while her father feels that he has been talking to someone every day.

This is a plausible future scenario, not a demonstrated outcome of today's AI systems. But early research gives us reason to study it seriously.

In March 2025, researchers from OpenAI and the MIT Media Lab reported a four-week randomized controlled study involving 981 participants and more than 300,000 chatbot messages. The researchers examined loneliness, real-world social interaction, emotional dependence, and problematic AI use across different conversational modes and topics. The randomized assignments did not produce statistically significant overall differences across the principal outcomes. However, participants who voluntarily used the chatbot more extensively tended to report greater loneliness, stronger emotional dependence, and less real-world socialization. This association does not demonstrate that extended AI use caused those outcomes: the amount of time participants chose to use the chatbot was not itself randomly assigned, and people who already felt isolated may have been more inclined to engage extensively with AI. The study nevertheless identified patterns that deserve further investigation.

Additional evidence appeared in Psychological Science in April 2026. Researchers Dunigan Folk and Elizabeth Dunn examined the relationship between social chatbot use and loneliness over twelve months, following more than 2,000 adults across four Western countries. Increased chatbot use predicted greater subsequent loneliness on a single-item measure of emotional isolation. On a broader measure of social connection, however, chatbot use did not significantly predict later deterioration. Instead, people who felt less socially connected were more likely to increase their chatbot use. The researchers explicitly cautioned against strong causal conclusions because of the exploratory nature of their analyses. These results suggest a potentially bidirectional relationship: loneliness may encourage people to seek AI companionship, while certain patterns of AI companionship may, in turn, be associated with increased loneliness.

The public-health context makes these questions particularly important. The World Health Organization's 2025 Commission on Social Connection estimated that approximately 15.8% of people worldwide—roughly one in six—experience loneliness. Loneliness is associated with adverse physical and mental health outcomes, although its causes are diverse and cannot be reduced to technology use. Personal AI agents could help address aspects of isolation by improving communication, supporting accessibility, and making social participation easier. They could also create a new form of convenient companionship that reduces the motivation to pursue more demanding relationships.

The scientific question is not whether people can become emotionally attached to artificial intelligence. It is whether that attachment becomes a bridge toward human connection or a comfortable substitute for it.

The distinction may depend less on how humanlike the agent appears than on how it is designed to participate in the user's social life.

VI. When an Agent Begins to Live Our Social Lives for Us

There is an even more subtle possibility: personal AI agents may begin conducting portions of our relationships on our behalf.

Imagine that your agent remembers an old friend's birthday. It knows the friend's preferences, recalls your previous conversations, and selects an appropriate gift. It drafts a warm personal message, including a reference to a memorable evening you spent together several years earlier. With authorization, it orders the present and arranges its delivery. Your friend receives an elegant gift accompanied by a thoughtful message and feels remembered. The entire exchange appears to represent care, attention, and continuity in a relationship. Yet you may have spent no time thinking about your friend that day.

There is nothing inherently wrong with using technology to remember birthdays or arrange gifts. Calendars, greeting-card services, and personal assistants have supported these activities for generations. The difference lies in the increasing completeness of the delegation. An advanced agent can potentially remember, select, write, purchase, schedule, and respond, reducing the user's participation to a single approval—or, under a broader standing authorization, perhaps less. The practical outcome may be excellent, but the relationship raises an uncomfortable question: if someone receives evidence of our affection without our having meaningfully exercised that affection, what exactly has been preserved?

The answer is not always simple. A person recovering from surgery may sincerely want to maintain relationships but lack the energy to manage correspondence. An agent that helps communicate genuine affection could preserve connections that might otherwise deteriorate. Someone who struggles with language or social expression may use AI to articulate emotions they genuinely feel but cannot easily express. In such cases, assistance makes human care more accessible. The concern emerges when the agent does not merely help express our intentions but increasingly substitutes for the attention, memory, initiative, and judgment that constitute our participation in the relationship.

Consider another familiar experience: apologizing after an argument. An agent may help us understand what went wrong, identify defensive assumptions, and find words that communicate regret without making excuses. That is valuable assistance. But if the agent drafts the entire apology, determines the appropriate emotional tone, and sends it without the user seriously reflecting on the harm caused, the recipient may receive the language of remorse without the personal transformation that remorse is supposed to express. The message may be perfectly written and still fail to represent the human work of reconciliation.

There is a difference between using intelligence to become a better participant in a relationship and using intelligence to avoid participating in the relationship at all.

As AI systems become better at predicting our preferences and acting autonomously, that boundary will become increasingly important. It cannot be resolved simply by requiring human approval for every action. A person can approve a message without reading it carefully or emotionally engaging with its meaning. The deeper challenge is to design assistance that preserves meaningful human agency rather than merely recording nominal consent.

VII. The Right Choice: Intelligence That Makes Us More Human

The solution is not to reject personal AI agents, romanticize inconvenience, or insist that every practical task must remain a human interaction. There is little virtue in waiting on hold for forty minutes, repeatedly completing the same forms, or negotiating trivial administrative details when a capable agent can perform those activities safely and accurately. Freed from routine burdens, people may have more time and energy to care for children, visit ageing parents, maintain friendships, participate in community life, and pursue meaningful interests. Properly designed, personal AI could become one of the most powerful technologies ever developed for expanding human autonomy.

But the design objective matters enormously. If success is measured primarily by how frequently users return, how long they converse, and how dependent they become on the service, commercial incentives may favor an agent that gradually makes itself indispensable. A different philosophy would measure whether AI assistance improves people's ability to pursue their own goals, including goals that involve other human beings. An agent could notice, with the user's permission, that an old friend has not been contacted for several months and offer to arrange a meeting. It could handle reservations, transportation, and scheduling, then become unobtrusive during the actual encounter. It could help someone rehearse an uncomfortable conversation without replacing the conversation itself. It could support a grieving person through a difficult evening while encouraging appropriate connections with trusted friends, family, or professional help when needed. It could assist people with social anxiety in developing confidence rather than making withdrawal the path of least resistance.

Such an approach would require technical and ethical discipline. Persistent personal memory should remain subject to clear user control, privacy protections, and appropriate retention limits. Agents should distinguish practical delegation from emotionally significant communication, providing meaningful opportunities for users to review messages presented as personal expressions of care. Their conversational systems should avoid manipulative claims of exclusive attachment or designs that encourage unnecessary emotional dependence. Research should evaluate effects over months and years, distinguish temporary assistance from behavioral substitution, and examine outcomes across different ages, cultures, living situations, and levels of social connectedness. Measures might include self-reported relationship quality, voluntary real-world social participation, perceived autonomy, and changes in loneliness—but such evidence should be collected with informed consent, not through intrusive monitoring of private relationships.

Most importantly, designers must recognize that people differ. For someone living alone in a remote settlement, working at sea, or facing severe mobility limitations, an AI companion may provide an important form of connection when alternatives are scarce. For another person, the same technology may become a convenient refuge from relationships that require patience, compromise, and emotional courage. A useful system must be capable of supporting both people without assuming that the optimal human life is one of constant sociability. The purpose is not to maximize the number of social interactions. It is to preserve the freedom and capacity to form relationships that matter.

Return, finally, to the woman leaving work on Friday evening. There is no reason her personal AI agent should not arrange dinner, buy groceries, or find the perfect birthday present. Perhaps it should do all those things, leaving her with more time than she would otherwise have had. But perhaps, when it notices that she has not visited her father for several months, it might offer something more valuable than another completed transaction: the opportunity to see him. It could help her find a convenient flight, arrange the journey, and manage the practical complications. Then it could allow her to put the device away, sit beside him at the kitchen table, listen to stories she has heard many times before, and hold his hand when the conversation falls silent.

The machine would have completed its work. The human experience would only just have begun.

Potato chips have their place, and so does cabbage. The problem begins when convenience becomes our entire diet. Personal AI agents should eliminate unnecessary burdens, not the reciprocal human experiences that give life its meaning. We should build intelligence that helps us participate more fully in the world, rather than making the world increasingly unnecessary to us.

The greatest achievement of personal AI will not be an agent that can do everything for us. It will be an agent that gives us more time, freedom, and willingness to do the things that only we can do for one another.