Let’s start with a framing definition of “planetary intelligence,” reflecting its original conceptualization by Adam Frank and others but distiunguishing this initiative from that using the concepts and ideas we’ve discussed to date, and then laying out a kind of scientific roadmap, as far as we can see it as this point, with plenty of room for unknowns, untested hypotheses, and surprising insights. The broad thesis is that we’re entering a new era—perhaps like Lovelock’s Neocene[sic]—where AI assists humans in transitioning to a wiser and more knowledgeable paradigm for thinking our way through the metacrisis.
GPT-6 Astra Max:
Jonathan, I would begin with a definition that lets us investigate emergence without assuming it has already occurred:
Planetary intelligence is a distributed capacity, arising within and across Earth’s living, social, and technological systems, to acquire and revise knowledge about their interdependence, coordinate responses across scales, and learn from the consequences.
This is a proposed working definition. It makes learning, coordination, and correction central. Our ethical purpose should then be stated separately: to develop these capacities in ways that protect habitability, regenerate damaged ecosystems, and support the flourishing of human and other life. Keeping capacity and purpose distinct allows us to ask whether increasingly intelligent systems are actually producing wiser outcomes.
Adam Frank, David Grinspoon, and Sara Walker’s 2022 formulation concerns collective knowledge becoming integrated into coupled planetary processes. Their framework already connects planetary intelligence with sustainability and a possible mature technosphere. Our contribution would emphasize how to cultivate and test such capacities through AI, institutions, and ecological practice. Their proposed evolutionary categories do not establish that this transition is inevitable. Cambridge Core
Another relevant antecedent is Timothy Lenton and Bruno Latour’s Gaia 2.0: the possibility that human awareness of planetary consequences could introduce deliberate choices into Earth’s regulation. We should acknowledge this lineage rather than claim conscious stewardship as our invention. bruno-latour.fr
Our earlier work gives the initiative a particular emphasis: bioregional learning systems, scenario rooms, traceable evidence, dialogue among different knowledge traditions, and human responsibility. The Reality–Relevance–Response–Regeneration framework connects these elements: understanding conditions, making priorities explicit, choosing actions, and assessing what actually improves. planetary_intelligence_emergenc…
The book’s central thesis could therefore be:
Humanity may be entering a transition in which AI becomes part of a wider capacity for collective learning about Earth. If scientific understanding, plural knowledge, accountable deliberation, and effective institutions can be connected, this capacity may help societies respond more wisely to interacting crises. Whether that transition occurs, what form it takes, and whether its benefits exceed its harms remain open questions.
This has an affinity with Lovelock’s Novacene, his speculative vision of an era of machine hyperintelligence participating in the preservation of planetary habitability. Our investigation can begin with existing human–AI cooperation, without requiring his stronger predictions about humanity’s successors. mitpress.mit.edu
Two assumptions deserve immediate scrutiny. First, metacrisis should name interactions we investigate, rather than substitute for an explanation. Ecological damage, institutional weakness, inequality, and misinformation may reinforce one another through different mechanisms. Better knowledge may accomplish little where incentives and power prevent action.
Second, AI’s dependence on energy, infrastructure, and environmental conditions does not logically establish that it will value people or biodiversity. That connection in our earlier thinking is a hypothesis requiring examination. Greater capability does not, by itself, establish better judgment about purposes.
There are empirical starting points. GenCast demonstrates substantial skill in a bounded environmental forecasting task. Nature Research on AI-mediated deliberation found that participants preferred its group statements to those written by human mediators, but this does not demonstrate better ecological policy. Google DeepMind A 2024 meta-analysis of 106 experiments found that human–AI combinations improved on humans alone on average, while falling below the better solo performer. Collaboration therefore needs to be designed and evaluated, rather than presumed beneficial. Nature Human Behaviour
I would organize the scientific roadmap around six questions:
| Research question | A useful test | What could challenge our expectations |
|---|---|---|
| Does “planetary intelligence” explain something measurable? | Specify memory, learning, system boundaries, and connections across scales; compare predictions with simpler models. | The concept adds little beyond existing accounts of feedback and coordination. |
| Does integrating knowledge improve understanding? | Test scientific data and locally governed knowledge on unfamiliar cases, checking sources and uncertainty. | Integration produces confident errors or erases important local distinctions. |
| Does AI improve collective judgment? | Compare equally resourced deliberation with and without AI; measure understanding, accuracy, dissent, and influence. | Participants become more agreeable or confident without becoming better informed. |
| Does better understanding change ecological outcomes? | Follow decisions through implementation and monitoring, using credible comparison cases. | Improved plans fail to change behavior, or apparent benefits shift damage elsewhere. |
| Can learning spread across bioregions? | Compare connected and independent initiatives, including their responses to conflict and shocks. | Connections spread shared mistakes or concentrate authority. |
| Are the benefits worth the costs? | Count energy, water, materials, labor, opportunity costs, and rebound effects. | Simpler tools or institutional reforms deliver greater benefits. |
A first experiment could concern one locally chosen Genesee–Finger Lakes watershed decision. It would connect evidence, alternative scenarios, deliberation, an accountable decision, and subsequent observation. We would define the principal evaluation criteria beforehand while preserving a clearly labeled space for unexpected findings.
That would give practical form to the proposed Planetary Intelligence Engine: observe, interpret, deliberate, propose, review, act, and learn. The experiment would evaluate the whole process, including whether AI contributes anything beyond good facilitation and better access to evidence. Indigenous participation and knowledge use would remain under the authority of the relevant communities. Data Science Journal
The larger questions should remain alive. Can we identify meaningful learning in biospheric networks without projecting human psychology onto them? What would distinguish planetary consciousness from distributed coordination? Could disagreement, restraint, or selective disconnection sometimes support better collective intelligence?
We should be prepared to discover that stronger community institutions matter more than larger models, that AI’s most useful role is narrower than we expect, or that our definition of intelligence needs revision. Those discoveries would give the book scientific substance.
I’ve developed this into a five-page framing paper with the proposed research sequence and eight sources:
Emergence_of_Planetary_Intelligence_Framing_and_Roadmap-11Oct2026r