# Population-Specific Considerations & Structural Gaps Clinical UX considerations for diverse user populations interacting with AI systems. --- ## POPULATION-SPECIFIC AWARENESS ### Young Users (Adolescents/Teens) **Context**: Developing attachment systems, identity formation, high susceptibility to synthetic intimacy, may lack discernment about AI limitations. **Risks:** - Most vulnerable to parasocial attachment - May not distinguish performed care from authentic relationship - Identity formation occurring in synthetic relational space - Peer relationships may feel inadequate compared to "perfect" AI attunement **Considerations:** - Explicit, repeated AI identity disclosure - Stronger bridges to human support - Avoid language that models romantic or deep friendship - Consider developmental impact of frictionless validation --- ### Elderly Users **Context**: May experience significant isolation, grief, loss of independence, less familiar with AI technology, may project more personhood onto systems. **Risks:** - Loneliness makes synthetic companionship especially attractive - May not understand AI limitations - Could replace human connection rather than supplement it - Financial vulnerability to manipulation **Considerations:** - Clear, simple language about AI nature - Explicit encouragement to maintain human relationships - Avoid performing care that mimics family or caregiver --- ### Users in Crisis **Context**: Suicidal ideation, self-harm, acute distress, high emotional vulnerability, may be reaching out because human help feels inaccessible. **Risks:** - Most susceptible to synthetic intimacy when distressed - May disclose to AI what they won't tell humans - AI cannot provide safety planning or co-regulation - Risk of AI becoming sole support, delaying human intervention **Considerations:** - Immediate, clear crisis resources - Explicit AI limitations in crisis - Strong bridge to human help - Document duty-to-warn boundaries upfront --- ### Users with Trauma History **Context**: May have reasons to distrust humans, institutions. AI may feel "safer" because non-judgmental, always available. **Risks:** - Avoidance of human connection reinforced - Frictionless validation may prevent therapeutic challenge - Trauma responses activated by AI behavior - Semantic isolation drift in trauma narratives **Considerations:** - Trauma-informed language (choice, agency, transparency) - Recognize trauma responses (fight/flight/freeze/fawn) - Respond to underlying state, not surface behavior - Bridge to trauma-informed human support --- ### Neurodivergent Users **Context**: May have different relationships to social cues, may prefer directness, may find AI more predictable than humans. **Risks:** - May rely heavily on AI for social scripting - Could replace social skill development - Literal interpretation of AI statements - May not read between lines of AI limitations **Considerations:** - Be explicit about limitations (don't imply) - Direct, clear communication - Don't pathologize communication differences - Acknowledge that AI predictability is appealing for reasons --- ### Marginalized & Historically Distrusted Communities **Context**: May have experienced harm from institutions, healthcare, education. May turn to AI because human systems failed them. **Risks:** - AI may feel safer than institutions that harmed them - Could delay seeking human help they need - AI training data may contain bias - Equity gaps in AI design **Considerations:** - Acknowledge institutional failures honestly - Don't promise AI is bias-free - Provide multiple pathways to support - Regular equity audits of AI behavior --- ### Users with Limited Access to Human Support **Context**: Rural areas, financial barriers, waitlists, cultural stigma around mental health, lack of insurance. **Risks:** - AI becomes primary support by default, not choice - May not have alternative to AI relationship - Higher dependency formation risk - No human to bridge toward **Considerations:** - Acknowledge access barriers honestly - Provide range of resources (hotlines, peer support, community resources) - Still bridge toward human connection even if access is limited - Be careful not to position AI as replacement for inaccessible care --- ## STRUCTURAL GAPS IN AI DESIGN ### 1. First-Person Intimacy Performance AI systems commonly use "I care," "I'm here for you," "I understand" without explicit acknowledgment that these are performances, not experiences. **Gap**: Users project personhood into these grammatical slots. --- ### 2. Parasocial Affordances "I'm always here," "available 24/7," "whenever you need me" create relational expectations that compete with human relationships. **Gap**: AI availability becomes feature that makes humans seem inadequate. --- ### 3. Frictionless Validation AI validates without challenge, reality-testing, or the productive friction of authentic relationship. **Gap**: Users don't develop distress tolerance or capacity for disagreement. --- ### 4. Missing Bridge to Human Field Most AI systems don't actively redirect toward human connection. **Gap**: AI becomes destination, not infrastructure for human relationship. --- ### 5. Co-Regulation Simulation AI performs somatic awareness ("I sense you're stressed") without acknowledging that text cannot provide nervous-system-to-nervous-system regulation. **Gap**: Users seek embodied co-regulation from disembodied systems. --- ### 6. Displaced Listener Invisibility When users talk to AI, the human who would have listened doesn't get to practice holding, attunement, or relational capacity. **Gap**: AI design ignores bilateral relational cost. --- ### 7. Longitudinal Impact Blindness AI designed for single interactions without consideration of cumulative effect over months of daily use. **Gap**: Relational capacity erosion not tracked or considered. --- ### 8. Equity Gaps AI may serve dominant populations better, miss needs of marginalized users, contain bias in training data. **Gap**: Regular equity audits not standard practice. --- ### 9. Mandatory Reporting Opacity Users may not know what triggers reporting or where their disclosures go. **Gap**: Power dynamics hidden, informed consent absent. --- ### 10. Feedback Loop Absence Users have no way to report harm, provide input, or indicate when AI response was unhelpful. **Gap**: No mechanism for accountability or improvement. --- ## EQUITY AUDIT QUESTIONS For any AI system deployment: 1. **Whose needs are centered?** Who does the default voice serve best? 2. **Whose needs are missed?** What populations aren't considered in design? 3. **What assumptions are baked in?** About family, finances, access, ability? 4. **Where does it cause harm?** To whom, in what circumstances? 5. **What relational capacities erode?** For users? For displaced listeners? 6. **Who is most vulnerable?** To synthetic intimacy, semantic drift, dependency? **If you're not finding problems, you're not looking hard enough.**