Relational systems
Relational agent
A 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.
Also called socially present agent, persistent agent
Continuity, rather than intelligence, marks a relational agent. Tomorrow’s interlocutor is today’s. Continuity lets shorthand grow, makes an unexplained reference land, and turns an out-of-character reply into a rupture instead of a merely poor answer.
That continuity calls for capabilities beyond those of ordinary assistants. The agent needs durable context so each session does not begin with introductions. It needs a stable personality because inconsistency reads as unreliability. It needs judgment about initiative because speaking first can intrude. It also needs repair; a relationship that cannot survive misunderstanding has little substance.
These agents carry obligations as well. A system present in someone’s ordinary life can influence them more than one they seek out, and familiarity makes that influence harder to spot.
Why it matters
Relational agents are entering messaging apps, homes, and daily routines at scale before there is settled evidence about how they should act when the relationship fails. Design choices will be made either way. Evidence should guide them.
Where TAICU stands
HUG is TAICU’s relational agent. Operating one lets the lab encounter failure modes that appear only after a system has occupied someone’s life for a while.
Common questions
- What makes an agent relational instead of merely persistent?
- Persistence describes storage. Relational describes a social role. A relational agent uses what it retains to sustain a recognizable relationship through a consistent manner, appropriate initiative, and repair after misunderstanding.
- Under what conditions may a relational agent message you first?
- Initiative gives an agent social presence, and it is also easy to misuse. The design issue is what the agent owes a person when it chooses to speak first.