Notes / Principles & methods
Short positions that guide the research program. These are working principles, separate from experimental findings.
A system that remembers more or coordinates better may become more useful. That growth does not itself justify a larger mandate.
Knowledge, capability, and authorization must remain distinct. An agent’s accumulated context can inform a recommendation; it cannot grant permission for an action. Collective-intelligence research inherits the same constraint.
A useful agent should be able to question an assumption, identify a conflict, and argue for a better course of action.
It may advise, negotiate, challenge, and push back. Those contributions support human judgment. They do not authorize the agent to override it or silently substitute its own goals.
An unanswered question belongs in the record. Filling it with a plausible explanation conceals the very uncertainty the research needs to resolve.
Distinguish what is known, what remains unknown, and what failed. When further work is justified, state the next falsification test: an observation that could contradict the working claim, with a bounded scope and an explicit stopping condition.