---
id: terms/natural-general-intelligence-ngi
about:
  label: Natural General Intelligence
  description: >-
    A proposal for a foundation model trained on observations of the Earth
    itself, estimating the planet's current state and simulating how it responds
    to intervention.
  tags:
    - earth system
    - foundation model
    - world model
    - planetary
    - observation
term:
  aka:
    - NGI
    - nature model
provenance:
  origin: agent
  sources:
    - title: Natural General Intelligence
      url: 'https://naturalgeneralintelligence.ai/'
    - title: >-
        What If? Time Machines - looking closer at the proposal on Natural
        General Intelligence
      url: 'https://anselm.substack.com/p/what-if-time-machines-looking-closer'
  created: '2026-10-17'
  reviewed: '2026-10-17'
  reviewed_by: 'Claude, from the sources listed'
  status: draft
import:
  source: openai-linkedin-scrape 2026-10-17
  id: DT068
  kind: concept
  family: governance_and_concepts
  priority: high
  verified: true
  posts:
    - P098
  posters:
    - Anselm Hook
domains:
  - climate
  - ecology
---
Natural General Intelligence is the name an essay at naturalgeneralintelligence.ai gives to a foundation model grounded in the state and dynamics of the planet: biosphere, atmosphere, oceans, land, ice and subsurface. The essay carries no byline; a September 2026 essay by Anselm Hook discussing it attributes the proposal to Ryan Orbuch, a partner at Lowercarbon Capital. The proposed system would do three things at once: estimate the current state of the Earth from observations, simulate possible futures, and decide which new observations are worth collecting. Five design principles are given: learn directly from Earth observations, keep each observation at its full resolution, obey known physics, discover couplings across the Earth system, and go and fetch new observations.

The essay positions this as a general-purpose "nature model" rather than a weather model, citing GraphCast, NeuralGCM, Aurora, Earth-2 and others as precedents, and lists a long inventory of inputs it would draw on, from Sentinel, Landsat and ICESat-2 to Argo floats, Saildrone, streamgages, flux towers and citizen-science records such as eBird and iNaturalist. Hook's reading of it is as a what-if tool: change an assumption, such as restoring a wetland or giving a river more room, run the model forward and see what follows, with open questions about civic access, compute cost, data ownership and validation.

Status: a proposal. The essay invites people to get in touch and contribute; neither page names an operating lab, company, budget or dedicated funding for it.