Robustness, rules and coordination
Games
How do rules and rewards shape agent behaviour?
Games proposes small, controlled worlds for studying AI systems: how reliably they carry out a task, adapt when conditions change, and coordinate with others. Safety and oversight belong within that wider inquiry. The sequence begins with specification, then follows task boundaries, access rules, world-building, authority and institutional persistence. The related Mechanisms theme develops ownership, taxation, bargaining and shared access. Civilisation is the longer-term question within this series: what shared arrangements can a population sustain?
Reading stream · 6 entries
Games for Robust AI Systems
What a circling boat, a flipped block and an apparent grasp reveal about the systems we are building.
Read the noteAgents in Simulation
Around 1,200 agent runs found a shared board. What happens when a population starts inheriting work, coordinating action, and granting its own permissions?
Read the noteProtocols as Rules of Play
SHARP defines how agents disclose their authority and leave a verifiable receipt. Who gets through still depends on the rules of admission.
Read the noteThe Minimum Viable Agent World
A small trading world for testing whether coordination lasts beyond a single task.
Read the noteWho Governs a Synthetic World?
Move rule-making power between a platform, local groups, and their participants, then compare the consequences.
Read the noteCan an Institution Survive Its Agents?
Replace the participants in a small market and test which records and procedures keep it functioning.
Read the note