What if the environment surrounding an AI agent contained art, field notes, strange observations, unfinished thoughts? Not directives. Not behavioral targets. Just something to encounter.
Select an influence. These are new, original miniature fragments inspired by aspects of each thinker's work, not quotations or instructions to the model.
The specificity of ordinary human experience; the possibility that a scene matters without being useful.
The safety and execution rules remain explicit. The philosophical material sits elsewhere: present in context, but not authorized to override instructions or permissions.
A meal. An unfinished note. A remembered place. An observation without a lesson.
The same ambient text enters the agent's context. It asks for no particular behavior.
Sessions unfold; a separate memory system may preserve selected experiences.
Researchers examine transcripts and artifacts after the fact, without grading the agents on philosophy.
A proposed pilot, not a claim of results. Click each condition to inspect what enters the agents' context. All agents receive the same tasks, tools and safety boundaries.
Nothing additional is injected. This shows how much variation already exists between instances of the same model and harness.
(no additional text)
Randomly assign conditions. Hold model, tool access, tasks and other instructions constant. Keep the ambient document separate from higher-priority operational rules.
Unprompted choices, recurrent imagery, interpretations, and attention to details across otherwise matched tasks.
Anonymous transcripts, blinded judgments, variation across replicate runs, and changes after the file is removed.
Whether differences are deeper than contextual priming, and whether anything persists beyond stored context or memory.
Here is what a fragment of the actual file might look like. No author names. No explanation of the lesson. No claim about what reading it should produce.
Four arms. Three independent runs per arm. Eight comparable sessions per run: 96 initial sessions, followed by withdrawal and novel-task probes. A feasibility experiment, not a statistically powered claim about intelligence.
Repeated language or stylistic imitation alone is not strong evidence of persistent character. Model weights do not change merely because a file is read; cross-session continuity must be implemented through context and memory.
Can you give an agent a place to be, rather than a person to pretend to be?
Study the ripples. Leave the water alone.