The first fifteen minutes with a care robot reveal more than a prepared demonstration. They show whether a person understands what the machine does, can refuse, knows how to obtain help, and retains access to a human relationship. They also expose noise, waiting, errors, and uncertainty that a laboratory did not reproduce. Designing that beginning with curiosity, responsibility, and care puts technology in its proper place: a limited, supervised support rather than a substitute for relationship.

Real arrival is different from a presentation#

A demonstration usually starts with a charged device, an orderly room, and a team that knows every step. Everyday life starts differently. The person may arrive tired, in pain, frightened, hurried, or without having chosen the interaction. They may miss the first instruction, interpret movement as a threat, or be unsure whether they are speaking to a machine, a remote operator, or a system that stores data. The first quarter of an hour reveals those differences.

A demonstration usually starts with a charged device, an orderly room, and a team that knows every step.

Reception should be designed as a critical task. Before activation, the purpose, limits, responsible party, and available human alternative are explained. Information needs understandable formats and time for questions. A screen of terms and conditions or a pleasant voice is not enough. The person should know which actions the robot performs, which decisions it does not make, when it records, how it stops, and what happens on failure. Observation before participation should be an option.

Professional curiosity does not ask why the person failed to adapt quickly. It investigates which feature of the setting or interaction produced uncertainty. Distance, speed, volume, lighting, language, response time, and bodily signals are considered. Silence is not automatically interpreted as consent. A straightforward route to pause, withdraw, or request support is provided without requiring justification of discomfort.

Defining a limited place for the machine#

A useful robot has a specific function. It may transport materials, provide reminders, facilitate communication, or support an exercise. Convenience should not expand its role into assessing emotion, determining clinical priority, or replacing conversation without evidence and authorisation. The more vulnerable the setting, the more important it is to delimit competence. Humanoid appearance and natural language must not confuse technical performance with moral understanding.

Task allocation should be assessed through need, benefit, risk, reversibility, and alternatives. Automating repetitive activity may release team time for relationship. It can also reduce encounters and conceal signs a person would have noticed. Evaluation should not assume that efficiency and care naturally move together. It asks what work disappears, what is displaced, who monitors it, and how users and professionals are affected.

Limits must be visible in both interface and organisation. A robot should not claim certainty when it operates with an estimate. If it does not understand, it should acknowledge that and refer. A person must be able to distinguish an automated suggestion from a professional decision. The human team retains authority and time to intervene; if staffing is removed because the machine supposedly covers a service, supervision becomes fictional. A help button without an available responder is not human control.

Consent requires genuine choice. When a robot is integrated into a care home, hospital, or publicly supported home service, refusal should not mean losing care. A reasonable alternative is offered, while any genuinely essential function is explained specifically. A person can accept some tasks and decline others, change their mind, and know how the preference is recorded. Consent is reviewed when software or purpose changes.

Sensors turn ordinary spaces into sources of data. Cameras, microphones, maps, movement patterns, and interaction logs can reveal health, routines, and relationships. Only necessary data should be collected, for a justified period and with restricted access. People need to know which processing occurs on the device, what leaves it, who receives that information, and when deletion occurs. Privacy for relatives, visitors, and workers is also part of design.

The power-off option should be understandable and physically accessible. Voice, button, gesture, or remote control may be needed, with redundant choices. Stopping should not erase a request for help or leave someone at risk. Once stopped, the robot enters a safe state and communicates what it will do. Emergency logs require protected review. The aim is to preserve agency without constructing a false choice between privacy and safety.

A protocol for the first fifteen minutes#

The opening period can be structured without becoming rigid. Initially, a responsible person introduces themselves, confirms communication preferences, and explains purpose. The robot is then shown while stationary, with sensors and limits described and the stop function demonstrated. The person decides whether to continue. A low-risk task is tried next under close observation, leaving time for questions. Finally, events are summarised and the next step is agreed.

A checklist cannot replace attention. Fear, confusion, pain, or overload should stop the procedure even if the script anticipates another phase. A professional checks understanding through open questions rather than requiring an affirmative response. Infantilising language, attributed emotions, and promises that the machine will learn everything should be avoided. Explanations adapt to language, sensory access, and technological experience without assuming incompetence because of age or disability.

Documentation records configuration, consent, incidents, questions, support, and the final decision. Human expressions should not be turned into unnecessary labels. An accessible summary includes contact details, a complaint mechanism, and a review date. Observations may inform system improvement, but secondary uses need a legitimate basis and safeguards. An introductory session is not unlimited permission for research or model training.

Case 1: a wayfinding robot in a hospital#

A hospital installs a mobile robot to guide visitors. In its presentation it travels accurately through empty corridors. During normal hours, a patient with low vision arrives amid noise and crowds, cannot distinguish the screen, and finds that the robot starts moving before confirming that she is following. Once it loses her, it returns to the start. Staff assume she did not understand the technology. Review identifies failures in contrast, speed, audio cues, detection, and immediate alternatives.

The hospital changes arrival procedures: reception offers a choice between human and robotic guidance, describes the service, and configures pace and channel. The device confirms each turn, waits for a response, and refers to staff after lost contact. Diverse users test routes, functions are limited, and abandonment is treated as design evidence rather than personal error. The technology provides simple direction and never handles clinical priority. First minutes become part of routine evaluation.

Case 2: a social robot in residential care#

A care home introduces a robot that recalls activities and proposes conversation. One resident believes it reports directly to family members and avoids mentioning discomfort. Another feels required to answer because the device repeats prompts after silence. Interaction rates look high, but they do not reflect freedom. The team rehearsed functions without examining how authority, listening, and privacy would be understood in a place that is also a home.

Automatic use is suspended and accessible individual conversations take place. Existing data is explained, unnecessary logs are deleted, and residents choose functions, times, and stop words. Repetition is limited and human contact remains available. In-person activities are not reduced. Outcomes concern choice, wellbeing, and concerns rather than minutes of interaction. A committee including residents, families, and workers reviews changes. The machine assumes a place subordinate to each person's life plan.

Case 3: robotic assistance in rehabilitation#

A service introduces a robotic device to support walking. During the first session, the technical protocol dominates attention and a patient does not understand how to report pain. The harness is correctly positioned, but pace increases before she can formulate a question. She completes the exercise and it is classified as successful. Later she avoids another appointment because of fear and discomfort. Repetition counts had concealed perceived safety and continuing consent.

The team redesigns introduction around visual explanation, unloaded demonstration, an accessible stop signal, and jointly decided progression. Pain is recorded before, during, and after use, while a professional remains available. Conventional alternatives stay open. Device decisions are visible to the team, and parameters do not change without confirmation. Follow-up includes confidence and willingness to return as well as physical outcomes. Repair acknowledges the initial experience and offers another session only if the patient wishes.

Responsibility before, during, and after use#

Responsibility cannot dissolve among manufacturer, supplier, management, and professional. Every function needs an identified party who authorises use, supervises, maintains, responds to incidents, and communicates with the person. Contracts specify updates, cybersecurity, accessibility, logs, support, and withdrawal. A purchasing organisation retains duties even where a technical decision belongs to a supplier. Known risks are translated into practice rather than archived in an inaccessible manual.

Testing before deployment should use realistic scenarios: noise, irregular connectivity, obstacles, different bodies, and varied communication methods. Failure monitoring during use should not turn life into constant surveillance. Afterwards, incidents, refusals, human referrals, staff workload, and unexpected effects are reviewed. A low complaint rate is insufficient because a channel may be unknown or feared. Evaluation needs accessible methods and independent observation.

When harm occurs, repair includes stopping use where risk remains, addressing consequences, communicating clearly, correcting records, and changing the system. The person is not blamed for an unanticipated movement, and failure is not hidden behind the phrase user error. Investigation distinguishes technical, organisational, and communicative causes. Learning is shared proportionately and verified before restart.

Caring for the human team#

A robot changes work even where no post disappears. It may add configuration, cleaning, monitoring, explanation, and error-management duties. If these tasks receive no time or recognition, technology shifts care into invisible labour. Deployment should consult those who will operate it, assess competencies, and adjust schedules. Training covers ethical limits and responses to uncertainty, not only buttons and maintenance.

Staff need freedom to challenge a recommendation and stop a device without punishment for lower throughput. Alerts should prevent fatigue and prioritise genuine situations. Cooperation with users helps uncover effects but does not transfer quality assurance to them. Knowledge is distributed across care, engineering, accessibility, privacy, security, and lived experience. No single perspective represents the entire system.

Leadership should resist simplistic claims of substitution. If projected savings depend on removing human contact before safety and acceptance are demonstrated, the robot occupies the wrong place. Benefits can include lower physical workload, greater accuracy, or availability, measured alongside relationship, autonomy, and equity. Productivity is a legitimate criterion but not the only one where a person depends on a service.

Institutional sources for decision-making#

The World Health Organization offers guidance on ethics and governance of artificial intelligence in health, supporting attention to safety, autonomy, and accountability. The UNESCO Recommendation on the Ethics of Artificial Intelligence provides principles concerning rights, oversight, and diversity. These frameworks must be translated into specific tasks and the applicable regulatory context.

The European Commission maintains institutional information about the regulatory framework for artificial intelligence. The NIST AI Risk Management Framework supports structured identification, measurement, management, and monitoring. No source replaces required clinical, technical, or accessibility testing. They help teams frame questions, assign responsibility, and retain evidence throughout the life cycle.

The useful truth of the first quarter-hour#

The first fifteen minutes tell the truth because they bring technology, environment, expectations, and power together. They reveal whether explanations are understood, refusal has consequences, human support is real, and the device responds to an actual body. Designing this period well does not guarantee the whole service, but it exposes assumptions before they become repeated harm.

A robot that knows its place does not claim to replace judgement, affection, or responsibility. It performs a bounded task, shows its limits, accepts being stopped, and refers when it cannot understand. Curiosity examines experience; responsibility assigns owners and boundaries; care preserves freedom and relationship. Only then can automation extend capacity without dehumanising those who receive care or those who provide it.