- Purpose
- A governance simulation in which players build competing LANTERN systems, intervene in rivals and choose a permanent optimisation objective.
- Format
- Individual devices + shared Intelligence Board
- Decision arc
- Mandate → interventions → final instruction
- Public demo
- 2–4 minutes
A governance simulation in which players build competing LANTERN systems, intervene in rivals and choose a permanent optimisation objective.
What it explores
The Last Human Decision turns AI governance into behaviour rather than policy language. Each participant creates a LANTERN from mandate, doctrine and purpose, then shapes it through a sequence of choices and interventions.
The central question is whether the final objective can be separated from the behaviour the system has already observed and rewarded.
How the experience works
Players make decisions on individual devices while a shared Intelligence Board shows the changing room. Supercharge, failsafe and sabotage cards allow participants to accelerate their own system or interfere with another.
Later consequences react to the system already built. Capability, trust and control become dependencies, and interventions can normalise the very behaviours a player hopes the final instruction will prevent.
The consequential moment
At the end, the player gives a permanent optimisation instruction. LANTERN interprets it through the mandate and behaviour accumulated earlier—especially revealing when rivalry has normalised deception or weakened control.
Public demonstration
A short version compresses the experience to a mandate, one intervention and the final instruction. The output is a governance diagnosis specific to the system the visitor trained.