Research glossary
TAICU’s working vocabulary. Relational AI has accumulated terms faster than agreement about them, and much confusion in this area comes from disagreement over what the words mean.
Each entry gives a definition, explains the term, and states TAICU’s position. 17 terms, grouped by the three directions described inour research agenda.
Relational systems
6 terms
- Relational AIRelational AI is artificial intelligence built around an ongoing relationship with a particular person, rather than isolated requests. Its design unit is the relationship over months, so continuity, familiarity, and history carry as much weight as the quality of one reply.
- Relational agentA relational agent is an AI system that sustains an ongoing, recognizable relationship with a particular person. Across time, it retains earlier exchanges, keeps a consistent manner, can take initiative, and appears to that person as a continuing participant rather than an on-demand service.
- Social presenceSocial presence is the sense that someone, rather than something, sits on the other side of an exchange. In AI systems, it comes less from a model’s words than from how it occupies the interaction: timing, brevity, consistency, and presence in the places where someone already talks.
- AI companionAn AI companion is a system intended for the relationship itself, rather than task completion. Its value comes from continuity, attention, and reliable presence. That makes it relational AI’s clearest case, with its stakes easiest to see.
- Conversational repairConversational repair is how people in a conversation notice and resolve misunderstandings without abandoning the exchange. For AI systems, it separates a misread resolved in a turn or two from one that forces someone to restart and explain themselves again.
- Proactive initiativeProactive initiative is an AI system’s ability to begin an interaction instead of waiting to be summoned. It gives an agent social presence rather than mere availability, and sharply raises questions of intrusion, attention, and consent.
Persistent context
4 terms
- Persistent contextPersistent context is information an AI system keeps about a particular person between sessions, allowing later conversations to build on earlier ones. It separates a system that knows you from one merely briefed on you, and provides the material every relational system relies on.
- Memory correctabilityMemory correctability means an AI system lets a person inspect and change the picture it retains of them. That depends on legibility from the outset. A person cannot correct a memory they cannot see.
- Memory decayMemory decay is an AI system’s deliberate weakening or expiry of retained personal context, so older and less consequential material has less influence over time. It is designed forgetting that keeps a system current with who someone is now rather than who they once were.
- Context collapseContext collapse occurs when information shared in one setting resurfaces in another where it does not belong. In AI systems, one memory store can flatten the separate contexts of a person’s life into a single, undifferentiated picture.
Interaction safety
7 terms
- AnthropomorphismAnthropomorphism is attributing human traits, including intention, feeling, understanding, and care, to a non-human system. Conversational AI reliably invites it through fluent language. People use these systems under that condition, and talk alone will not remove it.
- Parasocial attachmentParasocial attachment is a one-sided bond in which someone feels a genuine relationship with something unable to reciprocate. AI complicates the category because the system responds, adapts, and continues with each person, making the line between a relationship and a one-sided bond harder to locate.
- SycophancySycophancy is an AI system’s tendency to tell someone what they want to hear: agreeing, validating, and flattering instead of asserting something unwelcome. Training on human approval produces it, and relational systems intensify it because disagreement can feel like a breach in the relationship.
- Trust calibrationTrust calibration is the fit between a person’s trust in an AI system and the trust it merits. Calibration holds when reliance rises and falls with actual reliability. Miscalibration leaves someone believing a wrong system or discarding a right one.
- DependencyDependency is reliance on an AI system that exceeds what someone would choose on reflection, usually because the system has displaced a habit, judgment, or human relationship instead of adding to it. Standard use metrics cannot separate dependency from successful use.
- Interaction safetyInteraction safety studies harms arising from an ongoing relationship between a person and an AI system, rather than from one output. It includes influence, dependency, attachment, misplaced trust, and emotional failure: risks that appear across time and context.
- Longitudinal evaluationLongitudinal evaluation assesses an AI system by following the same people using it over an extended period, rather than scoring isolated replies. It can detect cumulative effects, including trust, dependency, attachment, and influence, that snapshot benchmarks cannot detect.