What happened before?
Preserve facts, events, preferences, and previous interactions.
AICO · LONGITUDINAL ALIGNMENT
Beyond memory,
toward alignment.
AICO explores how what accumulates across interactions can remain organized and evolve over time. By connecting person, context, topic, and strategy, AICO allows past interactions to shape how the system interprets and handles what comes next.
ACCUMULATED EXPERIENCEEVOLVING
PERSONgoals, preferences, experience
CONTEXTsituations, people, relationships
STRATEGYwhat worked, what comes next
Longitudinal alignment for what comes next.
Remember. Understand. Organize. Act. Update.
01 / THE NEXT STEP OF LONG-TERM MEMORY
"A thousand readers, a thousand Hamlets."The same thing can mean something different to different people—and to the same person in a different context.
An interpretive maxim inspired by Shakespeare's Hamlet
The same memory does not
mean the same thing.
Long-term memory can preserve what happened before. But past information has no fixed meaning on its own. What matters depends on who the AI is aligned with, what is happening now, what they are trying to achieve, and what has happened around it before.
AICO therefore goes beyond recalling past interactions. It maintains an evolving understanding of person, context, topic, and strategy, so the same memory can be interpreted and used differently as situations change.
What does it mean here?
Interpret what matters for this person, in this context, toward this purpose—and determine what should carry forward.
02 / FROM MEMORY TO FUTURE ACTION
What should memory
change next?
AICO does more than retrieve relevant history. It turns accumulated experience into explicit structures that can guide the next response, recommendation, plan, or action, then evolve as experience continues.
Understand
Build a working view of the person, their goals, and what has changed over time.
Read context
Identify the situation, people, constraints, relationships, and past experiences that matter now.
Organize strategy
Continue an existing path, adapt it, or create a new one when the situation differs.
Guide action
Use that organized understanding to make the next interaction more appropriate and useful.
03 / LONG-TERM MEMORY NEEDS STRUCTURE
More than a
memory archive.
AICO organizes accumulated experience around the person, the current context, recurring topics, and strategies, so that past experience can be reused without treating every new situation as the same.
NOWcurrent situation
SELFperson
CTXcontext
PASTexperience
Local relevance first. AICO reads the part of long-term experience that can inform the present situation.
Personal memory
Goals, preferences, experiences, boundaries, and changes that matter over time.
Context & relationships
Relevant people, situations, events, constraints, and relationship state when another person matters.
Dynamic topic graph
Recurring purposes and topics connected to the earlier situations that give them meaning.
Strategy tree
Paths that can organize multi-turn reasoning, next moves, transitions, and revision.
RUNTIME MECHANISMS
Dynamic topic matching · local graph retrieval · multi-source RAG · strategy execution · feedback calibration
04 / SITUATED MEMORY
The same memory,
different paths.
There is no single best way to act for everyone, or even for the same person in every situation. The same past experience can support different strategies when context, purpose, or relationship changes.
CURRENT QUESTION
Should I follow up?
One apparent problem. More than one reasonable path.
Continue directly
Shared history suggests a direct follow-up is normal and likely to be understood as care.
Branch the strategy
A more formal relationship and a different purpose call for a different message and timing.
PERSON × CONTEXT × PURPOSE × HISTORY = STRATEGY
05 / EXPERIENCE BECOMES FUTURE BEHAVIOR
A response is part
of a longer trajectory.
Each interaction is not an isolated answer. Past experience provides context for the current step; the result of the current step becomes evidence for what should be preserved, changed, or reconsidered next.
Read
Locate the current person, purpose, context, and relevant experience.
Match
Find related topics and strategy structures without assuming the situation is identical.
Shape
Reuse, extend, or branch a path for the situation at hand.
Act
Respond, recommend, plan, assist, or support a cooperative next step.
Calibrate
Use new evidence to decide what should become part of future understanding.
EXPLICIT STRUCTURE EVOLUTION
Reuse, extend, branch, revise, or reject. AICO keeps change connected to the experience and context from which it emerged.
07 / RESEARCH MATERIALS
Read the code.
Follow the argument.
AICO is developed in public. The source repository contains the research framework and implementation; the technical report will set out the long-term memory model, its evolving structures, and the rationale behind them.
OPEN SOURCETECHNICAL REPORT
AICO repository
Framework, algorithms, backend, and interface code for long-term AI memory, evolving context, strategy, and future action.
View on GitHubAI-human Alignment and Cooperation
Enabling AI to learn cooperation through long-term alignment. The report will formalize AICO's long-term memory model, context reasoning, strategy evolution, and maintenance regimes.
Coming soon.08 / CITATION
Citing AICO.
If AICO contributes to your work, please cite the project using the entry below. Until the technical report is published, this software citation is the recommended reference.
BIBTEX
@software{aico_lab_aico_2026,
author = {Pengcheng Zhou},
title = {AI-human Alignment and Cooperation: enabling AI to learn cooperation through long-term alignment},
year = {2026},
url = {https://github.com/PKQZPC/AICO},
note = {Open-source research framework}
}