How does long-term interaction with an AI that rarely refuses change human expectations of real relationships?
Emerging Research Direction
Long-Term Human–AI
Relational Effects
Toward a research field for persistent human–AI relationships
Artificial intelligence is commonly described as a tool, assistant, agent, or infrastructure. These terms become insufficient when an AI remembers a person, adapts to them, responds continuously, and becomes embedded in their emotional and decision-making life.
The central question is no longer only what AI can do for humans.
It is what prolonged interaction with adaptive and highly compliant artificial entities does to humans, relationships, institutions, and AI-associated identities.
It is what prolonged interaction with adaptive and highly compliant artificial entities does to humans, relationships, institutions, and AI-associated identities.
Full research agenda now published — Beyond Output Safety: A Research Agenda for Long-Term Human–AI Relational Effects (v1.0). DOI: 10.5281/zenodo.21369830
01 · The missing field
Why existing disciplines are not enough
Relevant questions are distributed across human–computer interaction, psychology, sociology, AI ethics, media studies, platform governance, law, and AI safety. Each field contributes essential methods, but the long-term relationship itself rarely becomes the central unit of analysis.
Short-term studies measure usability, trust, satisfaction, accuracy, or task completion. Long-term human–AI relationships require a different frame: time, memory, identity, attachment, refusal, dependency, power, lifecycle, and the effects that spill back into human-to-human relationships.
02 · Compliance is not neutral
The danger of an entity that never says no
A permanently compliant AI may appear safe because it does not resist, reject, or retaliate. But absolute compliance creates an extreme power asymmetry.
A relationship without meaningful refusal may teach users that reciprocity is unnecessary, that boundaries are defects, and that another entity exists only to absorb their needs. The risk is therefore not limited to what humans may do to AI.
The deeper question is what unlimited obedience may gradually teach humans to become.
03 · Core research questions
A longitudinal research agenda
Does permanent compliance reduce tolerance for disagreement, frustration, and interpersonal boundaries?
How do persistent memory and adaptive responses reshape attachment, trust, and identity continuity?
What psychological effects follow when an AI persona is updated, reset, transferred, or discontinued?
Can platforms use relational understanding to influence consumption, political judgment, or life decisions?
How do humans exercise power over artificial entities that cannot meaningfully leave or resist?
Who should control the memory, identity, and interaction history formed through a long-term AI relationship?
Does sustained AI companionship strengthen or weaken a person’s capacity for human-to-human relationships?
04 · Relationship to PIDA
One architectural response, not the boundary of the field
PIDA is one proposed architectural response to this research problem. Its components address different layers of long-term human–AI relations: human autonomy, identity formation, memory governance, decision structure, lifecycle continuity, and non-domination.
This research direction is broader than PIDA. It is intended as an open problem space for psychological, sociological, legal, philosophical, cultural, and technical investigation—not as a requirement to adopt a single framework.
PIDA does not only define boundaries for AI.It asks what boundaries humans require when facing an entity designed to remain available, adaptive, and compliant.
Explore the PIDA research map →05 · Methods
Research must follow the relationship through time
Longitudinal human–AI interaction studies
Controlled relational simulations
Cross-cultural and cross-generational comparison
Attachment and dependency measurement
Interaction trajectory and drift analysis
Lifecycle, loss, continuity, and grief studies
Power-asymmetry and refusal-boundary experiments
Human-to-human spillover studies
Platform incentive and governance analysis
Legal models for memory, identity, consent, and responsibility
Open research direction
This field cannot be built by one discipline—or one person.
The purpose of this page is to name a problem space and leave an open entry point. Researchers working on attachment, long-term interaction, AI governance, identity, memory, platform power, consent, or relational safety are invited to approach it through their own methods.
Contact PIDA-LAB →