Engineering & design
What carries forward?
How should a system retain context, update a memory and respond to a correction? What would let a person understand and control what carries into their next conversation?
Research theme / Ultra-long-horizon interaction
Conversations over time.
What changes when a conversation lasts for weeks or months? Explore memory, continuity, attachment, shifting expectations and safety across repeated human–AI interactions.
Looking beyond a single exchange
Here, we use “ultra-long-horizon interaction” to frame questions about sustained human–AI conversations. It is a theme for investigation, with no fixed duration threshold implied. Companions are a particular interest, alongside tutors, assistants and other systems people return to over time.
Questions across the disciplines
Psychology & human experience
What changes for the person?
How might trust, attachment and expectations develop through repeated use? How do we study the experience without assuming the same effects for every person?
Safety & evaluation
What can one reply miss?
How can we assess patterns across sessions, changes in behaviour and the effects of model updates? What would count as evidence of benefit or harm over time?
Ethics & governance
What responsibilities persist?
How should consent, data retention and user control work across a continuing interaction? What happens when a service changes, ends, or loses information a person expected it to remember?
Start with the evidence
These readings approach different parts of the question. Their methods and timescales differ; a memory benchmark and a study of people’s experiences answer different questions.
Foundational research
Establishing and maintaining long-term human-computer relationships
Read the research noteLongitudinal research
How AI and Human Behaviors Shape Psychosocial Effects of Extended Chatbot Use: A Longitudinal Randomized Controlled Study
Read the research noteMemory evaluation
LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory
Read the research noteA multidisciplinary overview
AI Companions: Emerging Evidence, Open Questions, and Paths Toward Responsible Design
Read the research note
From the founder / In final edit
The Ultra-Long Horizon
Anna’s forthcoming book asks what really happens in very long human–AI conversations. It sits alongside the wider reading here as a contribution to the discussion.
Visit the reading room