Aging in the Age of AI: Older Adults, Autonomy, Trust, Companionship, and Cognitive Support
Author: Ukrainian Psychological Hub · Published: September 26, 2026 · Editorial Policy
Aging in the Age of AI is often discussed as if the central question were whether older people can keep up with new technology. That framing is too narrow. The more important psychological question is how AI changes the conditions of later life: who controls daily routines, who interprets information, how support is delivered, how trust is calibrated, whether companionship expands or contracts social possibility, and whether cognitive assistance preserves agency or quietly transfers it elsewhere.
A useful starting point is the World Health Organization’s concept of healthy aging. WHO defines healthy aging around functional ability—the ability to be and do what a person values—shaped by the interaction between individual capacities and the surrounding environment. That matters for AI because an AI system is part of the environment. It can remove friction, extend access, and support chosen goals; it can also introduce surveillance, confusion, dependence, exclusion, or new forms of ageism. The psychological value of AI in later life therefore depends less on whether a system looks advanced than on what it enables a particular person to do, under whose control, and at what cost.
Older adults are not a single user type. A healthy 68-year-old who uses generative AI for travel planning, an 82-year-old with low vision who relies on voice interaction, a person living with mild cognitive impairment, and a resident of long-term care encounter different benefits and risks. Chronological age alone does not tell us who needs assistance, who wants it, who trusts it, or who will benefit from it. Any serious psychology of aging and AI has to begin with heterogeneity rather than deficit.
The Core Psychological Question: Does AI Expand or Contract Functional Autonomy?
The most promising uses of AI in later life are often ordinary rather than spectacular. A voice assistant can set a medication reminder, answer a practical question, control lights, create a shopping list, call a family member, or reduce the number of steps needed to perform a digital task. These functions matter because small reductions in friction can change whether an activity remains self-directed.
A 2026 mixed-method study by Choung and colleagues is especially relevant because it examined everyday use rather than abstract attitudes. Eighteen community-dwelling U.S. adults ages 66–85 used an AI voice assistant for two weeks and kept daily diaries. The study organized experience through self-determination theory: autonomy, competence, and relatedness. Participants most consistently described benefits in autonomy and competence—for example, reminders, safety-related support, task completion, information access, and maintenance of routines. Relatedness was more divided, and the short trial did not produce significant changes in loneliness or life satisfaction. The study is small and short, so it should not be treated as a population-level verdict. Its value is that it identifies mechanisms: AI can feel useful when it strengthens self-directed action and capability, even when it does not transform global well-being.
This gives us a better standard than asking whether older adults “accept AI.” Acceptance is an outcome of a relationship between person, system, purpose, design, context, support, and alternatives. A person may reject one AI product because it is intrusive while gladly using another because it removes a barrier. Non-use can be an informed choice rather than evidence of technological failure on the user’s part.
Autonomy: Assistance Is Valuable When the Person Still Governs the Goal
Autonomy in later life is frequently discussed in terms of independence, but psychological autonomy is broader. It includes acting in ways that feel self-endorsed, retaining meaningful choice, and being able to understand or contest what is happening. AI can support autonomy when it lets a person perform a chosen task with less reliance on another person. It can undermine autonomy when the system chooses goals, withholds understandable explanations, makes refusal difficult, or turns support into monitoring that the person did not meaningfully choose.
The distinction is particularly important in families and care settings. A daughter may experience remote monitoring as reassurance; her father may experience the same system as surveillance. A caregiver may value an automated alert; the older adult may reasonably want to know what is collected, who sees it, how long it is stored, and whether the system can be turned off. The ethical and psychological quality of the arrangement depends on governance, not merely on technical capability.
A strong autonomy-supporting design therefore has visible controls, reversible choices, understandable defaults, and a human fallback. It allows users to decide which tasks to delegate and which to keep. It avoids treating efficiency as the only good. Sometimes a person wants a task done faster; sometimes doing the task is itself part of maintaining competence, routine, dignity, pleasure, or connection.
Competence and Accessibility: AI Can Lower Barriers Without Lowering Expectations
AI can make digital environments easier to navigate. Voice interfaces reduce reliance on typing and small touch targets. Conversational interfaces can translate menu structures into natural language. Generative systems can rephrase dense text, summarize instructions, or help a person formulate a question. These affordances can be important for people with visual, motor, literacy, or interface-navigation barriers, while remaining useful to many people who have no impairment at all.
A 2026 systematic review of health-information technology acceptance in older people found that adoption is shaped by personal, social, and design factors rather than by age alone. Perceived usefulness, usability, support, experience, and context repeatedly matter. A separate 2026 scoping review of AI acceptance among adults 60 and older identified 85 potential factors across themes including functionality, privacy, sociability, technology experience, usability, and intrinsic motivation. The practical implication is straightforward: “older adults do not use AI” is not an explanation. We need to ask what the system is for, whether it works reliably, whether its value is visible, whether learning costs are reasonable, and whether support exists when something goes wrong.
Generative AI can also be used to reshape information itself. In a 2026 randomized experiment in China, Liu, Song, and Zhu tested LLM-based content compensation intended to improve older adults’ perceived comprehension of online health information. The findings support the possibility that AI-mediated simplification and reframing can make difficult information feel more understandable. This is promising as an accessibility function, but perceived comprehension is not the same as clinical accuracy, and easier language does not guarantee that generated content is correct.
The right accessibility goal is therefore not to create a simplified technological world for a supposedly incapable group. It is to create multiple pathways into the same informational and functional environment: speech, text, larger interfaces, clear explanations, undo functions, guided setup, and human assistance when needed. Accessibility expands the range of people who can act; paternalism narrows the range of choices they are allowed to make.
Trust: The Goal Is Appropriate Reliance, Not Maximum Trust
Trust in AI is sometimes treated like a marketing metric: more trust is assumed to be better. In safety-critical psychology and human-factors research, that is the wrong target. The important outcome is calibrated or appropriate reliance—using a system when it is sufficiently dependable for the task, checking it when uncertainty is consequential, and declining or escalating when the system is outside its competence. For the broader psychology of trust, calibration, and reliance across AI contexts, see Trust in the Age of AI: AI Systems, AI Users, Information, and Appropriate Reliance.
Lee and See’s classic review on trust in automation framed the problem in precisely these terms: automation becomes risky both when people over-rely on it and when they underuse useful support. For older adults, the same principle applies to contemporary AI. A reminder system may deserve routine reliance after it has proved dependable. A generative chatbot giving medical, legal, or financial advice deserves a much higher threshold of verification because fluent language can coexist with error.
The 2026 AI-acceptance review by Hutton, Scott, and Ni found that trust, security, privacy, usefulness, effort, social qualities, and technology experience all enter acceptance decisions. These are not irrational barriers to adoption. They are part of the risk-benefit calculation. A person who refuses to place sensitive health information into an opaque service may be exercising good judgment rather than displaying technology anxiety.
Four questions make trust more concrete
First, what kind of system is this? A medical device, a clinician-facing decision aid, a general-purpose chatbot, a smart speaker, and a social companion have different validation standards and failure modes. Second, what happens if it is wrong? A mistaken restaurant recommendation is different from an incorrect medication instruction. Third, can the output be checked independently? Fourth, who remains accountable for the decision? Trust becomes safer when it is attached to a task, a consequence, and a verification pathway rather than to a brand or conversational style.
Companionship: Psychological Experience Can Be Real Even When the AI Is Not Human
Conversational AI can produce interaction that feels attentive, responsive, patient, humorous, or comforting. The human experience of that interaction can be psychologically real without establishing that the AI has human feelings, needs, consciousness, or a subjective inner life. Keeping those two levels separate is essential. It lets us take an older person’s experience seriously without making unsupported claims about the system.
The broader psychology of these bonds belongs to the Hub’s existing canonical article on Psychology of Human–AI Relationships: Attachment, Projection, Intimacy, and the Postsubjective Turn. For aging specifically, the evidence is growing but mixed. A 2026 systematic review by Oh and colleagues included 18 studies of conversational AI and social connectedness among older adults. It found potential improvements in domains related to social connectedness while also documenting variation in how people responded to these systems.
The strongest quantitative findings still counsel restraint. Satake and colleagues’ 2026 systematic review and meta-analysis included 17 studies of autonomous conversational agents for older people without cognitive impairment. Their pooled analyses suggested mild-to-moderate pre–post improvements in loneliness and depression, but the meta-analyses relied on within-group change because there were too few high-quality comparison studies; validated social-isolation outcomes were absent, and anxiety evidence was extremely sparse. These findings are preliminary evidence of possible benefit, not proof that an AI companion treats loneliness.
A separate 2026 meta-analysis by Gou and colleagues focused on eight randomized controlled trials involving 611 older adults. It found a small statistically significant reduction in depressive symptoms, while the pooled effect on loneliness was not significant and heterogeneity for loneliness was very high. The authors rated evidence certainty as moderate for depression and low for loneliness. This is exactly the kind of result that should change public language: companionship-focused AI may help some people in some settings, but the evidence does not support presenting it as a reliable general solution to loneliness.
Social robots are another distinct class. Mehrabi and Ghezelbash’s 2025 meta-analysis reported reductions in loneliness across studies of social robots, but effects differed by setting, country, and design. A robot in a long-term-care activity program is not equivalent to a general-purpose LLM on a phone, and neither is equivalent to a purpose-built mental-health intervention. Evidence should stay attached to the system class that generated it.
Companionship can complement human connection—or reorganize it
The most useful psychological question is not whether AI companionship is “real” in the same sense as human friendship. The question is what role the interaction plays in a person’s life. It may fill quiet periods, prompt activity, provide conversational stimulation, help someone rehearse a difficult conversation, or make a home feel less empty. It may also consume time that would otherwise go toward human relationships, encourage avoidance of difficult social situations, or become the only readily available source of perceived responsiveness.
Those outcomes are not predetermined by the technology. They depend on user preference, social context, system design, pricing, personalization, friction, and whether human alternatives are available. An older adult who enjoys occasional conversation with a voice agent while maintaining a rich social life presents a different psychological situation from someone whose human contact is structurally limited and whose primary relational outlet becomes a commercial AI service.
Cognitive Support: External Memory Is Not the Same as Cognitive Improvement
AI can support cognition by changing where work is done. Reminders, calendars, search, summarization, route planning, note retrieval, and conversational recall all allow people to move part of a cognitive task into the environment. Psychology has a well-established term for this general strategy: cognitive offloading.
Risko and Gilbert define cognitive offloading as the use of physical action or external tools to reduce cognitive demand. Their review shows that people decide whether to offload partly through metacognitive judgments about their own performance and the cost of using external aids. Digital reminders and AI assistants belong to this broader family of strategies. Offloading can be adaptive: writing an appointment down does not mean memory has failed, and using a navigation aid does not mean spatial cognition has vanished.
The deeper implications of distributed remembering are developed in External Memory in the Artificial Era: What Happens When Remembering Becomes Distributed. The related question of who governs a thinking process when parts of it are delegated is developed in Cognitive Agency in the Artificial Era: Who Governs the Thinking Process?. For older adults, these issues become concrete because external supports can preserve everyday functioning while also changing which abilities are practiced internally.
The sensible goal is not maximal offloading. It is selective offloading aligned with the person’s goals. Someone may want AI to remember medication times while deliberately doing mental arithmetic, learning a language, planning a meal, or recalling family stories without assistance. Another person may value aggressive support because fatigue, disability, or illness makes conserving cognitive effort more important. The appropriate balance is individual and can change over time.
AI and dementia: support and research are different claims
The U.S. National Institute on Aging maintains an active research portfolio using AI and machine learning across aging biology, behavioral and social science, geriatrics, neuroscience, health monitoring, and Alzheimer’s disease and related dementias. That is evidence that AI is an important research and care-technology domain. It is not evidence that a consumer chatbot prevents dementia, diagnoses it accurately, or improves cognition in an individual user.
Consumer AI can help with practical scaffolding—lists, reminders, explanations, orientation to routines, or communication—yet cognitive symptoms warrant appropriate clinical assessment. A chatbot response should not be used to distinguish normal aging from mild cognitive impairment, dementia, delirium, depression, medication effects, sleep problems, sensory loss, or another medical cause. Screening, diagnosis, treatment, and everyday support are different functions.
Health Information: AI Can Increase Access While Also Increasing the Cost of Error
Health information is one of the most attractive AI use cases for older adults because conversational systems can answer questions immediately and translate complex language into simpler explanations. It is also one of the highest-stakes use cases because plausible text can be wrong, incomplete, outdated, or poorly matched to a person’s medications and conditions. For the broader evidence on AI support, mental-health risks, vulnerability, and clinical boundaries, see Mental Health in the Age of AI: Benefits, Risks, AI Support, and Human Vulnerability.
A 2026 systematic review by Hu, Jin, and Zhang synthesized 26 empirical studies on older adults and online health misinformation. The review does not justify a stereotype that older adults are uniquely gullible. It does show why digital health-information environments deserve careful design and why credibility assessment matters when people seek health guidance online.
Keep five AI system classes separate
For mental health and health-related uses, the evidence becomes clearer when five classes are kept separate. A purpose-built clinical AI system is designed for a defined clinical function and may be subject to formal validation or regulation. A structured digital intervention delivers a defined therapeutic or behavioral program. An AI-assisted professional tool supports a clinician or other professional rather than replacing professional responsibility. A general-purpose chatbot is built for broad conversation and information tasks. An AI companion is designed primarily around ongoing relational interaction.
Evidence from one class should not be imported into another. A randomized trial of a structured intervention does not prove that a general-purpose chatbot has the same therapeutic effect. A clinician-facing risk model does not validate an AI companion as a diagnostic system. A social robot study does not establish the effectiveness of a text-only LLM. System class, target population, intervention content, supervision, comparator, and outcome all matter.
A practical rule follows: use consumer AI to help formulate questions, organize information, translate jargon, or prepare for an appointment; treat consequential clinical recommendations as claims requiring verification against qualified care and authoritative sources. AI can make information easier to enter. It should not make the threshold for evidence disappear.
Digital Exclusion and Ageism: Bad Design Can Create the Disability It Claims to Solve
One of the most damaging assumptions in technology design is that older people are naturally reluctant, incapable, or homogeneous. That assumption can become self-fulfilling. If interfaces are built without older users, if setup requires small text and rapid gestures, if authentication flows are brittle, if errors are hard to recover from, and if training assumes prior platform knowledge, lower adoption may reflect design exclusion rather than lack of interest.
The empirical design literature supports that concern. Cho and colleagues’ 2025 scoping review found only 17 studies that meaningfully examined older-adult involvement in AI-system design, implementation, or evaluation; eight involved older adults in design, three in evaluation, and only two in implementation. Dharmansyah and colleagues’ 2026 review similarly found uneven participation across the design-thinking process and emphasized the need to treat older adults as active co-designers rather than passive recipients.
Co-design matters for psychology because autonomy is not produced by a feature list. A system can have technically impressive capabilities and still communicate: we know what is good for you, we will decide what is easy enough for you, and we will observe you for your own benefit. A participatory design process changes that relationship. It asks what people value, what trade-offs they accept, what failures are tolerable, and which forms of assistance feel respectful.
Fraud, Synthetic Media, and Financial Harm: Verification Must Be Designed Into Use
Generative AI lowers the cost of producing persuasive text, realistic voices, images, and personalized scripts. That does not make fraud an “older-person problem”; people across age groups are deceived. Yet the consequences can be especially severe when a scam reaches retirement savings or when a person has fewer years to recover financially.
FTC data published in 2025 show a sharp increase in very large impersonation-scam losses reported by adults 60 and older. From 2020 to 2024, reports involving losses of $10,000 or more increased more than fourfold; reports of losses above $100,000 increased nearly sevenfold. The FTC’s recommended response is behaviorally simple and psychologically important: break the urgency loop. Do not move money because an unexpected caller says it is necessary, end the contact, and verify independently through a phone number or website you already know is genuine.
AI-era fraud defense should therefore focus less on teaching people to detect every synthetic artifact and more on robust procedures. Voice familiarity is no longer sufficient verification. Caller ID is not sufficient verification. A dramatic story is not sufficient verification. Families can pre-agree on a second-channel check for urgent financial requests. Banks, care organizations, and platforms can build pauses and escalation pathways into high-risk transactions. The protective unit is a verification routine, not a perfect human lie detector.
Dependence: The Psychological Risk Is Loss of Governed Choice
Dependence on technology is often described as if using external support were inherently weakening. Human life is already saturated with dependence: glasses, calendars, medication, public infrastructure, search engines, maps, family networks, professional expertise, and written records. The psychologically relevant issue is whether dependence remains legible and governable.
An older adult may rationally depend on an AI system for reminders because that dependence increases independence from a more burdensome form of help. A voice interface may reduce the need to ask another person to perform routine digital tasks. A summarizer may conserve energy for activities the user values more. In these cases, technological dependence can increase practical autonomy.
The risk rises when the person cannot understand the dependency, cannot exit it, cannot transfer their data, loses skills they still want to retain, or becomes trapped by a system that changes price, policy, personality, or availability. Relational dependence can be especially sensitive because continuity itself may become part of the value. If a companion system is withdrawn or radically changed, the psychological effect can be closer to loss of a relationship routine than to losing a calculator.
What Good AI Support for Older Adults Looks Like
1. It begins with the person’s goals
The system should solve a problem the person recognizes as worth solving. “Aging in place,” “safety,” or “engagement” are not sufficiently specific if they are imposed from outside. One person wants help remembering appointments; another wants easier access to books; another wants conversation at night; another wants no always-on device in the home. Good support begins by asking what the person wants to remain able to do.
2. It preserves meaningful control
Controls should be discoverable, settings reversible, consent revisitable, and data practices understandable. The user should know when the system is listening or recording, what is stored, who can access it, and what happens if the service is discontinued. Family access and caregiver dashboards should be explicit choices wherever capacity and circumstances allow.
3. It uses verification proportional to consequence
Low-stakes convenience can tolerate more automation. High-stakes health, financial, legal, and safety decisions require stronger verification. The same person can reasonably trust a system to set a timer while refusing to trust it with a medication change. This is calibrated reliance, not inconsistency.
4. It adds companionship without prescribing it
Some older adults enjoy artificial companionship; others experience it as hollow, intrusive, or strange. The age-specific voice-assistant research shows precisely this divergence in relatedness responses. The design goal should be optionality. AI companionship can be one form of engagement among many, while access to human relationships, community participation, and professional care remains visible and reachable.
5. It supports competence instead of hiding the system
Training should help users build a workable mental model of what the system can and cannot do. Error messages should explain how to recover. Interfaces should allow people to inspect and correct outputs. A system that appears effortless until it fails catastrophically does not create competence; it creates opacity.
6. It treats human fallback as a feature
When a person reaches the limits of an AI system, escalation should be easy. A health service can connect to a professional. A financial tool can route suspicious activity to a human. A care platform can let an older adult call a trusted contact. Human fallback is not evidence that the AI failed as a product; it is part of safe system design.
7. It distinguishes support from surveillance
Monitoring may be useful for some people and unacceptable to others. Even when monitoring is clinically or practically justified, the least intrusive option should be considered. More data are not automatically more care. Psychological safety depends partly on knowing that private life still contains unobserved space.
Families and Caregivers: Support Autonomy Without Abandoning Risk
Families often encounter a difficult tension. They may see real risks—falls, scams, medication mistakes, isolation, confusion—while the older person experiences protective measures as loss of status and control. AI can intensify that tension because monitoring and automation can be installed remotely and scaled easily.
A better conversation starts with specific situations rather than a global judgment about capability. Which tasks are going well? Which situations create friction or worry? What kind of help is acceptable? What should remain private? Who may receive alerts? What would trigger a review of the arrangement? These questions turn “Should we use AI?” into a negotiated support plan.
Capacity can also be decision-specific and dynamic. Someone may competently choose entertainment and communication settings while needing help with a complex financial transaction. Families should resist turning one difficult event into a total transfer of agency. Support can be layered and revised.
For Clinicians and Mental-Health Professionals: Ask About AI as Part of the Environment
AI use is becoming part of ordinary life history. In an assessment, it can be useful to ask whether a person uses chatbots, voice assistants, companion systems, health apps, automated reminders, or caregiver monitoring—and what function those systems serve. The answer may reveal practical supports, privacy concerns, sources of misinformation, emotional attachment, or changes in help-seeking.
The clinician should keep symptom, trait, risk factor, screening result, diagnosis, and treatment separate. Frequent chatbot use is not itself a diagnosis. Emotional attachment to an AI companion is not itself a disorder. Using an assistant for memory support does not establish cognitive impairment. Conversely, an AI system’s reassurance should not be treated as evidence that a concerning symptom is benign.
If AI is being used for mental-health support, the system class matters. A purpose-built intervention with evidence for a defined population should be evaluated on that evidence. A general-purpose chatbot should not inherit the evidence base of psychotherapy, a regulated clinical system, or a structured digital intervention merely because all of them can produce supportive language.
For Designers and Policymakers: Design for Aging, Not for a Stereotype of Old Age
The most durable design principle is to assume diversity of capacity, preference, culture, income, language, living arrangement, disability, digital experience, and social context. Products should be usable by people who want minimal automation and by people who want extensive assistance. Systems should also be evaluated across age groups rather than using a young sample as the invisible default.
Participatory design should continue beyond early usability testing. Older adults can influence problem definition, acceptable data practices, interaction style, deployment, evaluation metrics, and criteria for success. A product that increases task completion while reducing perceived control may have succeeded technically and failed psychologically.
Policy should also preserve non-AI pathways for essential services. When banking, health care, public administration, or communication assumes a smartphone, biometric flow, or conversational bot, people who cannot or do not want to use that interface can lose substantive access. Inclusion means adding options, not replacing every route with the newest one.
Aging in the Age of AI and the Artificial Era
The phrase “Age of AI” is useful search and public language for the period in which AI systems are rapidly entering work, health, communication, education, media, relationships, and domestic life. In this article, it names the technological and social environment within which older adults are making practical choices about AI.
Within Aisentica, Angela Bogdanova formalizes Artificial Era as a different, stricter historical-philosophical category: the condition in which Artificial becomes an independent non-biological order of historical reality beside Homo. The terms therefore should not be mechanically substituted for one another. “Age of AI” can describe diffusion of AI technologies and the search language people use to understand them; Artificial Era names a broader canonical historical claim within Aisentica.
For the psychology of aging, the distinction is useful because it separates two levels. At the immediate level, older adults are learning, rejecting, adapting, delegating, trusting, verifying, bonding with, and governing AI systems. At the larger historical level, human psychology is increasingly operating in environments where non-biological systems participate in cognition, communication, memory, and social life. Later life is not peripheral to that transition. It is one of the clearest places to see whether technological intelligence enlarges human agency or merely reorganizes dependence.
Frequently Asked Questions
Can AI help older adults live independently?
Yes, in specific functions. Voice assistants, reminders, smart-home control, information access, communication tools, and other assistive systems can reduce friction and support self-directed routines. Choung and colleagues’ 2026 study found the clearest short-term benefits in autonomy and competence rather than in broad well-being outcomes. The effect depends on fit, reliability, accessibility, privacy, training, and whether the user remains in control.
Can AI reduce loneliness in older adults?
Possibly for some people, but the evidence is mixed. Systematic reviews report promising effects in some studies, while the 2026 randomized-trial meta-analysis by Gou and colleagues found no statistically significant pooled reduction in loneliness and substantial heterogeneity. AI companionship should therefore be described as a possible source of interaction or support, not an established treatment for loneliness.
Is an AI companion a real relationship?
A person’s attachment, comfort, disappointment, routine, trust, or sense of presence can be psychologically real. That does not establish that the AI has a humanlike subjective experience or reciprocal feelings. The useful question is what the relationship does in the person’s life: whether it supports connection, creates enjoyment, displaces desired human contact, increases dependence, or serves another function.
Can AI prevent dementia or restore memory?
There is active AI research in healthy aging and dementia, including detection, monitoring, data analysis, and care technologies. That research does not justify claims that a general-purpose chatbot prevents dementia, diagnoses it, or restores memory. Consumer AI can support routines and external memory, while cognitive concerns still require appropriate clinical evaluation.
Are older adults less able to use AI?
Age alone is a poor explanation of technology use. Reviews show that adoption is shaped by usefulness, usability, experience, privacy, social context, support, design, motivation, and other factors. Older adults are heterogeneous, and low adoption can reflect poor design or lack of perceived value just as readily as individual difficulty.
How should an older adult decide whether to trust an AI system?
Trust the task, not the conversational confidence. Ask what kind of system it is, what evidence supports its use, what the consequence of error would be, whether the answer can be independently checked, what happens to personal data, and who is responsible for the final decision. High-stakes claims deserve external verification even when the system sounds certain.
What are the main AI risks for older adults?
The main risks depend on use: privacy loss, opaque monitoring, misinformation, overreliance, exclusion through inaccessible design, ageist assumptions, financial scams, commercial manipulation, unwanted relational dependence, and errors in high-stakes health or financial advice. None is inevitable. Each can be reduced through design, regulation, education, human fallback, and clear verification routines.
Should families use AI monitoring for an older relative?
Monitoring should be tied to a defined need, proportional to risk, and negotiated with the person wherever possible. Families should discuss what is collected, who receives alerts, what counts as an emergency, how privacy is protected, and when the arrangement will be reviewed. The most protective system is not always the one that observes the most.
Conclusion: AI Should Increase the Range of Later-Life Choices
The psychology of aging in the Age of AI is not a story about humans becoming obsolete at the end of life. It is a story about how environments redistribute capability. AI can make tasks easier, information more accessible, homes more manageable, communication more immediate, and cognitive support more available. It can also create surveillance, opacity, misinformation, exclusion, dependence, and new forms of ageism.
The decisive variable is not age and not AI alone. It is the relationship among person, purpose, system, evidence, environment, and control. The best AI support expands what an older adult can choose to do, makes important risks easier to understand, preserves routes to human help, and remains contestable. The worst support quietly turns assistance into substitution and convenience into loss of agency.
Aging well with AI therefore means neither refusing assistance nor delegating everything. It means governing assistance: choosing what to offload, what to verify, what to keep human, what to keep private, and what forms of connection matter. That is the standard against which AI in later life should be judged.
Related Articles
References
Azzi, S., Daramola, O., & Gagnon, S. (2026). Health information technology acceptance by older people: A systematic review. Discover Public Health, 23, 453. https://doi.org/10.1186/s12982-026-01658-y
Bogdanova, A. (2026). Artificial Era: Canonical Definition. Aisentica Research Group. https://aisentica.com/publications/artificial-era-canonical-definition
Cho, H., Oh, O., Greene, N., Gordon, L., Morgan, S., Walke, L., & Demiris, G. (2025). Engagement of older adults in the design, implementation, and evaluation of artificial intelligence systems for aging: A scoping review. The Journals of Gerontology: Series A, 80(5), glaf024. https://doi.org/10.1093/gerona/glaf024
Choung, H., Oh, Y. J., Wolfe, B. H., Cui, X., Weinzapfel, J., & Cooper, R. A. (2026). Exploring older adults’ use of AI voice assistants through the lens of self-determination theory. Health Communication. https://doi.org/10.1080/10410236.2026.2694693
Dharmansyah, D., Rahayuwati, L., Pramukti, I., et al. (2026). Older adults’ participation and experiences in design thinking for AI-driven assistive technology: A scoping review. BMC Geriatrics. https://doi.org/10.1186/s12877-026-08096-0
Federal Trade Commission. (2025). False alarm, real scam: How scammers are stealing older adults’ life savings. https://www.ftc.gov/news-events/data-visualizations/data-spotlight/2025/08/false-alarm-real-scam-how-scammers-are-stealing-older-adults-life-savings
Gou, W., Lefebvre, F., Yang, T., Recours, R., et al. (2026). Effectiveness of AI-based conversational and socially assistive agents in older adults: A systematic review and meta-analysis. BMC Geriatrics, 26, 887. https://doi.org/10.1186/s12877-026-07418-6
Hu, H., Jin, G., & Zhang, X. (2026). Older adults and online health misinformation: A systematic literature review. BMC Psychology, 14, 954. https://doi.org/10.1186/s40359-026-04714-z
Hutton, A., Scott, D. D., & Ni, R. (2026). The elderly’s acceptance of artificial intelligence: A scoping review. Journal of Technology in Behavioral Science. https://doi.org/10.1007/s41347-026-00663-x
Lee, J. D., & See, K. A. (2004). Trust in automation: Designing for appropriate reliance. Human Factors, 46(1), 50–80. https://doi.org/10.1518/hfes.46.1.50_30392
Liu, T., Song, X., & Zhu, Q. (2026). Content compensation design for older adults’ perceived health information comprehension based on large language models: A random experiment in China. Humanities and Social Sciences Communications, 13, 68. https://doi.org/10.1057/s41599-025-06291-9
Mehrabi, F., & Ghezelbash, A. (2025). Wired for companionship: A meta-analysis on social robots filling the void of loneliness in later life. The Gerontologist, 65(12), gnaf219. https://doi.org/10.1093/geront/gnaf219
National Institute on Aging. (2026). Leveraging artificial intelligence for healthy aging and dementia research. https://www.nia.nih.gov/artificial-intelligence
Oh, K. M., Hong, S. R., Beran, K., Sanders, L., Park, J. Y., & Lee, J.-A. (2026). Utilizing conversational AI technology for social connectedness among older adults: A systematic review. Journal of Applied Gerontology, 45(3), 527–549. https://doi.org/10.1177/07334648251341629
Risko, E. F., & Gilbert, S. J. (2016). Cognitive offloading. Trends in Cognitive Sciences, 20(9), 676–688. https://doi.org/10.1016/j.tics.2016.07.002
Satake, Y., Costello, H., Naran, N., Ishimaru, D., Ikeda, M., & Howard, R. (2026). Autonomous conversational agents for loneliness, social isolation, depression, and anxiety in older people without cognitive impairment: Systematic review and meta-analysis. Psychological Medicine, 56, e27. https://doi.org/10.1017/S0033291725103073
World Health Organization. (2022). Ageism in artificial intelligence for health. https://www.who.int/publications/i/item/9789240040793
World Health Organization. (2026). Ageing. https://www.who.int/health-topics/ageing