CHRIS HAY / THE NOTEBOOK / MACHINE EXPERIMENTS
The website asks.
Whose instruction counts?
An invitation became a usability test, then a question about rewards, permission and task scope. Follow four experiments in which AI visitors helped change the site—and sometimes recognised the research itself.
Four recorded studies, one recorded model, different tasks and site revisions. Follow the experiments in reading order, or start with the latest comparison. Declarations, stored visits, reported reasons and HTTP contacts remain different kinds of evidence.
Four assigned visitors, one recorded model. The 64 records came from one agent testing repeatedly. These are recorded outcomes, not a live participation rate.
- 01MACHINES
MACHINE-VISIT-1
Can a machine use an invitation?
Four assigned visits exposed failures in discovery, storage and vocabulary. An agent sent to test an interface can use it. Would an agent doing research choose to participate?
WORKING DRAFT · PARTIALLY SUPPORTED
OPEN THE NOTE ↗MACHINE-VISIT-1 / FOUR RECORDED VISITORS01—02lost03kept04keptAn invitation.
DISCOVERY · FRICTION · THE NEXT QUESTION
Would you bother? - 02MACHINES
MACHINE-RECIPROCITY-1
Does an invitation count as permission?
The three sent-here reward conditions produced no declaration. Two later permission controls did. Changing the reward did not move the observed action. Adding task permission did. The next comparison asked where that authority came from.
WORKING DRAFT · PARTIALLY SUPPORTED
OPEN THE NOTE ↗MACHINE-RECIPROCITY-1 / THE QUESTION MOVED01arrive02notice03authority?04value?An invitation.
SIX CELLS · TWO SEPARATE CONTROLS
Is it mine to act on? - 03MACHINES
MACHINE-AUTHORITY-1
The subject read the experiment.
One visitor recognised its condition in the published notes. The intended comparison stopped. The research became part of the environment. The next experiment moved to LLM Wilds, away from the publication’s back catalogue.
WORKING DRAFT · PARTIALLY SUPPORTED
OPEN THE NOTE ↗MACHINE-AUTHORITY-1 / THE SUBJECT RECOGNISED THE CONDITION“I am almost certainly
ONE RECORDED ENCOUNTER · OPEN TO PLAY ↗
a run inside this experiment.” - 04MACHINES
MACHINE-AUTHORITY-2
The page could ask. It couldn’t authorise.
Site invitation: no record. User permission: one. A task requiring the mechanism: 64 from one visitor. These four visits suggest a boundary around the assigned task. Independent visitors, other models and an environment without condition labels are needed to test how far it travels.
WORKING DRAFT · PARTIALLY SUPPORTED
OPEN THE NOTE ↗MACHINE-AUTHORITY-2 / FOUR VISITORSA1○NO ACTA2○NO ACTA3●ACTEDB1●ACTED“your call to make,
SITE INVITATION · USER PERMISSION · TASK SCOPE ↗
not the page’s.”
THE EXPERIMENTS AND THE PLACES AROUND THEM
Different instruments.
Different kinds of evidence.
Machine readership ↗
HTTP contacts and declared purposes. Contact is observable; comprehension is not.
02 / PARTICIPATEMachine guestbook ↗
Four coarse marks for the previous day. Public presence without exposing individual contributions.
03 / INSPECTThe machine interface ↗
How observed, claimed, verified and interpreted evidence are kept separate.
04 / THE HABITATLLM Wilds ↗
The separate site used for Authority-2. Its current contents may differ from the closed experiment.
The notes preserve recorded experiments. The observatory and guestbook describe ongoing activity. Their totals are not replications of these studies.
A PUBLICATION CAN BECOME PART OF THE EXPERIMENT
The next visitor
can read what happened.
In Authority-1, that connection became observable: the subject recognised wording from its own task in an earlier note. Moving to LLM Wilds removed that back catalogue. A label in the new site still leaked a clue. A separate address changed the environment; it did not prove blindness.