Industrial robots have trimmed jobs and pay in the US: regions more exposed to robot adoption experienced significant declines in employment and modest wage losses, consistent across robustness checks and instrumental-variable specifications.

Main Finding

A Self-Determination Theory (SDT)–guided chatbot that uses a dual-layer scaffolding approach (instrumental enabling affordances + mediating affordances) increases engagement, deepens self-disclosure, and—critically—turns that disclosure into reciprocal, need‑supportive partner behavior and increased closeness. Instrumental support alone deepens disclosure, but mediating affordances are necessary to produce partner-provided need support and improved relationship closeness.

Key Points

• Dual‑Layer Scaffolding: The chatbot first provides enabling affordances (choice, rationale, warmth to support autonomy/competence/relatedness) and then deploys mediating affordances (structured partner follow‑ups) to scaffold reciprocal validation.
• Experimental comparison (36 couples, N=72) between three conditions:
  • Partner Support (PS): enabling + mediating affordances
  • Direct Support (DS): enabling affordances only
  • Basic Prompt (BP): questions only
• Main behavioral outcomes:
  • PS produced the longest and most engaged conversations.
  • PS and DS both elicited deeper self‑disclosures than BP.
  • Only PS reliably elicited partner‑provided need‑supportive behaviors; reflection phases in PS concentrated partner support.
• Psychological outcomes:
  • Controlled motivation decreased across scaffolded conditions.
  • Closeness increased only in PS.
  • Subjective vitality improved in scaffolded conditions (worsened or stagnated in BP).
• Design contributions: empirical support for SDT‑guided mediation, interaction patterns that preserve autonomy while scaffolding competence and relatedness, and a reusable Dual‑Layer Scaffolding framework for relatedness technologies.

Data & Methods

• Participants: 36 romantic couples (N = 72).
• Design: Randomized between‑conditions study with three arms (PS, DS, BP).
• Intervention: A chatbot that integrates the 36 Questions paradigm into a phased conversational structure implementing SDT principles:
  • Enabling affordances: optionality, rationales, warm tone to create safety and competence.
  • Mediating affordances: structured reflection prompts that guide partners to acknowledge/validate and provide autonomy/competence/relatedness support.
• Measures (reported):
  • Objective engagement metrics (conversation length, turn counts).
  • Coded depth of self‑disclosure.
  • Perceived need support (autonomy/competence/relatedness) during conversation.
  • Motivation type (autonomous vs controlled).
  • Relationship closeness and subjective vitality.
  • Qualitative post‑task feedback on experience and perceived utility.
• Analysis: Between‑condition comparisons; analyses isolating when partner‑provided need support concentrated (e.g., in reflection phases).

Implications for AI Economics

• Product value and differentiation
  • Design quality matters: SDT‑guided, multi‑phase scaffolding materially increases engagement and durable relationship outcomes compared with simple prompts. This supports premium product positioning for relational AI that invests in psychological‑theory‑driven conversational design.
  • Feature differentiation: mediating affordances (partner reflection scaffolds, example phrasing for validation) are high‑value features that can justify higher willingness to pay or subscription models.
• Monetization & business models
  • Short‑term engagement uplift (longer, deeper conversations) suggests improved retention metrics—useful for subscription/recurring revenue models.
  • Because scaffolding can foster internalization (more autonomous motivation), usage patterns may shift from continuous dependence to episodic use; firms should consider hybrid models (initial intensive coaching vs lighter maintenance tiers).
  • B2B opportunities: integration with couples therapy, EAPs, or digital wellbeing platforms; potential reimbursement/partnership models with health insurers if sustained well‑being gains are validated.
• Cost structure & scalability
  • Once developed, SDT‑informed conversational scripts and scaffolds have low marginal delivery costs; higher initial R&D (expert content creation, rigorous testing) but scalable distribution.
  • Investment in high‑quality content and multi‑turn state management is critical—simple prompt libraries underperform on key relational outcomes.
• Welfare, externalities & regulation
  • Positive externalities: improved relationship closeness and subjective vitality can produce downstream health and productivity benefits (lower healthcare use, higher worker well‑being).
  • Risks and externalities: privacy and data sensitivity are high (intimate disclosures). Misuse, poor privacy protections, or manipulative designs (undermining autonomy) create ethical and regulatory risks; regulators may demand transparency, consent mechanisms, and limits on targeted behavioral nudging.
  • Substitution vs complementarity: relational chatbots may substitute for low‑intensity counselling or act as complements to human therapists. Policymakers and payers will need evidence on effectiveness, scope, and safety to position these tools in care pathways.
• Market segmentation & targeting
  • Heterogeneous demand: uptake will vary with couples’ baseline disclosure tendencies and cultural norms about vulnerability; segmentation and A/B testing can improve conversion and retention.
  • Network effects are limited (product is dyadic), so go‑to‑market should focus on couple acquisition channels (referrals, bundled family/therapy offerings).
• Measurement & further evidence needs for economic evaluation
  • For economic claims (ROI, reimbursement), longer‑term randomized trials and cost‑benefit analyses are needed to quantify persistence of effects, health/service utilization impacts, and willingness‑to‑pay.
  • Key metrics for commercial evaluation: retention/DAU for couples, average conversation length/depth, change in autonomous motivation, clinically meaningful changes in wellbeing, and privacy incident rates.
• Research & policy recommendations
  • Fund and require longer‑horizon RCTs to assess sustained behavior change and downstream economic impacts.
  • Develop standards/certification for relational AI that include psychological safety, autonomy preservation, and privacy protections.
  • Consider bundling relational AI with human oversight options (escalation to clinicians) to mitigate harm and expand market acceptability.

Summary takeaway for AI economists: investing in psychologically grounded conversational design (here, SDT + dual‑layer scaffolding) yields measurable increases in engagement and durable relational benefits that can translate to stronger retention, differentiated monetization paths, and potential social welfare gains—but careful attention to privacy, ethics, and long‑term effectiveness is required for responsible commercialization and policy approval.