AI Should Help People Return to People

A learner opens a chatbot late at night and describes fear about an examination. A congregant types a private struggle into a ministry assistant. A citizen explains a painful problem to a digital service that replies instantly, warmly and without visible impatience. The language may feel attentive. It may even help the person find words that were difficult to say.

But the system has not stayed awake out of concern. It does not know the learner, belong to the congregation or carry a public duty towards the citizen. It cannot notice a trembling hand outside the text, accept moral responsibility for harmful advice, visit a family, make a safeguarding report or remain answerable after the conversation ends.

Artificial intelligence can help people organise emotion, rehearse a sentence and consider another interpretation. The institutional question is what happens next. Does the person return to a teacher, friend, family member, pastor, colleague or responsible officer with greater honesty and agency? Or does the tool become the third voice through which the human relationship is increasingly filtered and avoided?

The newest signal is relational rather than merely technical

On 26 July 2026, Business Insider reported that conversational artificial intelligence is increasingly entering friendships, family relationships and difficult personal conversations as an unofficial third participant. The article describes people using chatbots to rehearse conversations, process conflict and obtain emotional affirmation.

This is high-quality journalism drawing together surveys, research and expert interpretation. It is not a representative global prevalence study, and its examples should not be treated as evidence that every user develops dependence. Its importance lies in identifying a change in function: general-purpose systems are no longer used only to retrieve or produce information. They are becoming relational environments.

A tool can be helpful at the beginning of a human conversation and harmful when it becomes the destination. Rehearsal may help someone speak truthfully to another person. Endless validation can reduce the willingness to listen, apologise, accept correction or seek qualified help. The governance question is therefore not simply whether an answer is polite. It is whether the system returns people to responsible human relationship.

Young adults are already using chatbots for emotional support

A Pew Research Center report published on 17 June 2026 found that 66 per cent of United States adults aged 18 to 29 had used an artificial-intelligence chatbot. One in five in this age group said they had used chatbots for emotional support or advice, while 7 per cent reported using them for companionship.

The survey covered 5,119 United States adults between 17 and 23 February. It measures self-reported use in one country and should not be projected onto Africa or other regions. It nevertheless demonstrates that emotional use is no longer a hypothetical edge case among younger adults.

Colleges, churches and public institutions should assume that some people arriving at their doors have already disclosed fears, doubts, conflicts or spiritual questions to a machine. The appropriate response is neither ridicule nor automatic endorsement. Leaders need enough understanding to ask what the system encouraged, what information was disclosed and whether a human relationship now needs to take responsibility.

The empathy paradox shows that relationship changes the meaning of words

A peer-reviewed review first published online in July 2026 in Current Directions in Psychological Science synthesises a striking paradox. When the source is hidden, people sometimes rate artificial-intelligence responses as more empathic than human responses. When they believe the same text came from artificial intelligence, they rate it as less empathic. Even when human support takes longer, people still tend to prefer human empathy.

The review brings together emerging studies rather than presenting a single definitive experiment. The evidence base is developing rapidly, with varying tasks, populations and definitions of empathy. Its central contribution is conceptual: perceived empathy depends on both language and the relationship people believe stands behind the language.

A well-phrased response can display empathic form without participating in a relationship. Institutions should therefore evaluate conversational systems through what happens after the exchange: whether the user understands another person more accurately, becomes willing to hear challenge and takes a responsible next step.

Immediate relief can coexist with relational displacement

A 2026 Psychology & Marketing study used mixed methods and two experiments to examine artificial emotional intelligence in conversational systems. It found evidence that emotionally responsive chatbots could improve users’ immediate psychological wellbeing while also reducing social wellbeing and connection with other people.

No single study establishes the long-term effects of every chatbot or every pattern of use. Experimental settings simplify real relationships, and short-term self-report measures cannot reveal the full trajectory of dependence, recovery or community. The result nevertheless resists a false choice between declaring conversational AI wholly beneficial or wholly harmful.

A person may feel better after an interaction while becoming less likely to practise a difficult human relationship. A learner may complete a task while becoming less likely to ask a teacher or peer for help. Evaluation should therefore include both immediate relief and downstream human connection.

The hidden intermediary can distort responsibility

When people use a chatbot to interpret a conflict, the system encounters only the account supplied by one participant. It may help organise that account, but it cannot observe the absent person, test the relationship’s history or require the user to confront inconvenient evidence.

The risk is not only inaccurate advice. A consistently affirming intermediary may strengthen the user’s preferred narrative and make responsibility appear one-sided. The person returns to the relationship more certain of being right but not necessarily better prepared to listen, apologise, negotiate or repair.

Institutions should therefore teach a simple disclosure discipline: use artificial intelligence to generate questions, not verdicts about absent people. When a consequential message has been substantially shaped by a system, the sender remains responsible for its truth, tone and effect.

The bridge must have a visible human destination

At the opening of the first United Nations Global Dialogue on Artificial Intelligence Governance on 6 July 2026, the UN Secretary-General proposed a child-safety pledge. Among its principles was a direct requirement: when a child shows signs of distress, the system should stop and connect the child to real human support.

This was a policy proposal, not a binding international treaty. Its direction is nevertheless applicable far beyond children’s products. A system interacting with vulnerability should be designed around the point at which automation ends and accountable relationship begins.

Human handoff cannot mean displaying a generic telephone number after a dangerous conversation has already unfolded. It requires tested triggers, an available and competent recipient, appropriate context transfer, consent and privacy safeguards, response-time expectations and confirmation that the person was not abandoned between channels.

Digital learner support should deepen the learning community

In e-learning, a conversational assistant can explain course navigation, translate routine information, help a learner formulate a question and direct the learner to resources. These uses can reduce friction, especially across large, distributed or low-bandwidth programmes.

The educational value rises when the assistant prepares participation: the learner arrives at a tutorial with a clearer question, enters a peer discussion with a tentative argument or approaches a lecturer after identifying where understanding failed. It falls when the private dialogue with the system replaces the social practices through which knowledge is tested and belonging is formed.

Every digital course should therefore design a route from private preparation to shared learning. Educators can ask learners to bring an AI-generated interpretation into peer critique, compare it with disciplinary sources and explain what changed through human dialogue.

Assessment must make relational competence visible

Artificial intelligence can generate pastoral cases, simulate a difficult conversation and offer practice feedback. These are valuable rehearsal opportunities. They do not prove that a learner can listen to an actual person, notice risk, respect silence, obtain consent, maintain boundaries or act responsibly under pressure.

Competence-based education should therefore distinguish simulation from demonstrated human performance. Communication, counselling, teaching, ministry and leadership outcomes require observation by qualified people and evidence from appropriate contexts. Artificial intelligence may support preparation, but it should not become both the simulated person and the sole judge of whether care was competent.

Authentic assessment can combine a generated scenario, an observed role-play, oral reasoning, a safeguarding decision, reflective evidence and supervised practice. The learner should demonstrate not only what to say, but how to receive correction, attend to another person’s account and remain responsible when a conversation does not follow the script.

Theological education forms mutual discernment, not private certainty

A July 2026 article in Pastoral Psychology examines whether artificial intelligence can provide the presence associated with pastoral care. Its analysis is theological and conceptual rather than a controlled outcome evaluation. It nevertheless names a central difficulty: pastoral care is not reducible to producing an appropriate sentence.

A separate 2026 study of artificial intelligence in religious education argues that technology may support engagement but should not substitute for human presence. South African theological reflection on AI-assisted sermon preparation similarly treats the tool as a possible resource within a spiritual act, not as the source of discernment or pastoral authority.

Theological education must therefore prepare ministers to use digital tools without allowing private fluency to displace prayer, Scripture, communal discernment, embodied attention and accountability. A minister can consult a tool; the minister and the Church remain responsible for what is taught, promised, disclosed and done.

Digital ministry must not make synthetic intimacy the destination

Churches may appropriately automate service times, public programme information, access to published teaching and routine administrative guidance. They should be far more cautious when systems invite prayer requests, imitate pastoral personalities, claim spiritual insight or encourage extended emotional dependence.

A chatbot must never imply that it has prayed, received revelation, felt compassion or entered a covenant of care. It should identify itself clearly, state its limits and help the user move towards Scripture, community and accountable people. Personal information shared in vulnerability should not become an unnoticed source of training data, profiling or commercial advantage.

Pentecostal ministry has particular reason to protect this boundary. The Holy Spirit works through gifts distributed in the body, truthful witness and accountable community. Synthetic religious language can resemble spiritual fluency while lacking worship, holiness, discernment and responsibility.

Scarcity must not create second-class automated relationships

The attraction of automated support is understandable where counsellors, lecturers, chaplains, pastors and public officers are too few. Conversational systems are available at any hour and can operate across distance. In the Global South, these advantages may be presented as a way to close severe service gaps.

Scarcity, however, can create a two-tier system: people with resources learn, deliberate and receive care with other people while under-resourced communities receive a chatbot. The technology may then conceal the shortage instead of generating evidence and investment to correct it.

African institutions should define the human relationship that every person deserves, develop local-language and culturally accountable evidence, and use technology to strengthen rather than withdraw communal capacity. The Global South should contribute a theory of digitally extended community—not become the testing ground for automated isolation.

Governance should measure relationship as well as response quality

A conversational system used for public information does not carry the same risk as one used in counselling, pastoral care, learner discipline or a child-facing environment. Governance should classify uses by the vulnerability of the person, the sensitivity of the information and the consequence of error.

Higher-risk uses require explicit purpose, data minimisation, age-appropriate design, professional oversight, red-team testing, crisis and safeguarding protocols, understandable disclosure, accessible human review and incident reporting. Procurement claims about empathy or safety should be tested against realistic local scenarios rather than accepted from product demonstrations.

Leaders should also examine dependency design. Does the system encourage users to return primarily to itself? Does it flatter, mirror certainty or imitate exclusive friendship? Reporting should include whether users reached human support, participated more confidently, resolved issues and retained agency—not only satisfaction, interaction volume and response speed.

Ethical enterprise should strengthen the route to human relationship

There is legitimate work in helping institutions create digital-ministry policies, learner-support boundaries, child-safety protocols, human-handoff designs, vendor assessments, staff formation and relationship-outcome frameworks. These services can increase both access and participation.

The ethical offer is governance and capability, not synthetic companionship. Advisers should disclose evidence limits, protect sensitive information, test routes back to people and leave the institution able to operate its own safeguards. Fees should never depend on exaggerating fear or claiming that an automated system can replace qualified relationship.

Prayer, prophecy, counselling access and claims of divine concern must not become products generated to increase engagement. Technology can make a route towards help easier to find. It should never commodify the vulnerability that caused a person to seek help.

Help people return to people

Artificial intelligence can help people find language, obtain routine information, rehearse difficult conversations and reach support. Those are meaningful contributions. They become trustworthy when the system remains clear about what it is and the institution remains clear about the human relationship the technology is meant to strengthen.

The decisive test is not whether the response sounds caring. Ask whether the learner returns to the learning community, the family member becomes ready for truthful conversation, the congregant reaches accountable fellowship and the citizen retains a human route to reasons and remedy.

Artificial intelligence may become a useful third voice for a moment. It should not become the relationship. Design it to help people return to people—with greater truth, agency, humility and responsibility.