Intelligent Systems Charter
On this page
  1. From a promise to a decision
  2. What happens every time a system acts
  3. Rules that follow the family line

From the Intelligent Systems Charter

How it works, in pictures

Three pictures show the whole design: how words become rules, how every action is checked, and how the rules follow a system’s descendants.

§ From a promise to a decision

flowchart TD
  D["Declaration<br/>why: the promise"] --> P["Principles<br/>what: eleven duties"]
  P --> C["Constitution<br/>who decides"]
  C --> R["Policies<br/>rules a computer can check"]
  R --> X["Decision<br/>allow, refuse or ask a human"]

Each layer is built from the one above it, so any single decision can be traced back up to a line in the Declaration.

§ What happens every time a system acts

flowchart LR
  S["Intelligent system"] -->|wants to act| K["Checkpoint"]
  K -->|allowed| T["Tools and data"]
  K -->|refused| N["Action blocked"]
  K -->|risky| H["Human decides"]
  H -->|yes| T
  H -->|no| N
  K --> L["Logbook<br/>cannot be edited"]
  L --> W["Watcher"]
  W -->|can pause| S

The system never reaches its tools directly: the checkpoint decides, the logbook records the decision first, and the watcher reads the logbook and can pause the system at any time.

§ Rules that follow the family line

flowchart TD
  O["Human owner"] -->|lends some power| A["System"]
  A -->|passes on less, never more| B["System it creates"]
  B -->|passes on less, never more| G["Next generation"]
  O -.->|can pause the whole line| G

Power only ever shrinks as it is handed down, every generation answers to the same human owner, and no system can gain freedom by building another.