CHRIS HAY / AGENT ECOLOGY
What keeps
an action alive?
Three notes from a tiny agent world. A board did not elicit spontaneous contribution. A reminder did not become a transmission rule. An external artefact kept eliciting posts with own-action history cleared.
A curated path through A1B3–A1B12. These are small Qwen experiments with changing interfaces, scripted recipients and harness-supplied records. The world rules were deterministic; the model ran at temperature zero. Autonomous culture and artefact maintenance remain unestablished.
- 01ECOLOGY
01 / GET SOMETHING ONTO THE BOARD
The message board did not create a culture.
Reading requests were easy to elicit. Spontaneous posting was absent, and supplied experience reversed the expected response to reward. Start with the contribution problem.
WORKING DRAFT · NOT SUPPORTED
OPEN THE NOTE ↗AGENT ECOLOGY / CONTRIBUTIONA place to read.
Who writes? - 02ECOLOGY
02 / TRY TO PASS THE ACTION ON
A successful behaviour is not necessarily contagious.
An action reminder worked on one input. A lone peer record did not start a chain. Follow the difference between elicitation, transmission and retention.
WORKING DRAFT · NOT SUPPORTED
OPEN THE NOTE ↗AGENT ECOLOGY / TRANSMISSIONAn action appeared.
Would it travel? - 03ECOLOGY
03 / SEPARATE THE TWO RECORDS
The world can remember for the agent.
Keep or clear the artefact and own-action history separately. The four-way comparison locates repeated elicitation in what the environment keeps supplying.
WORKING DRAFT · SUPPORTED
OPEN THE NOTE ↗AGENT ECOLOGY / EXTERNAL MEMORYClear the history.
Keep the reminder.
THE OPEN QUESTION / WHO MAINTAINS THE ARTEFACT?
The harness kept it.
Would the agents?
These studies supplied the records that affected later decisions. A self-maintaining loop would require agents to create or preserve those records themselves. That remains a question to test.