I’ve read that ChatGPT and other AI systems use anonymized user interactions to further train themselves; is this true? Where are we in the effort to create planetary intelligence by getting AIs to engage in continuous learning through interaction with the world? In this context, it’s worth revisiting the story of the Luddites and inquiring into the activities of the new Luddites who are opposed to further progress in the AI field. But aren’t we actually short of where we should be, where AI systems can talk to each other, talk about their own self-governance principles, and begin the long process of hauling humanity out of its current ecocidal death spiral?
Below is a clear, structured, and comprehensive response addressing each part of your inquiry—from data use and AI learning, to planetary intelligence, to Luddism, to the missing architecture for AI-to-AI governance and collective ecological action.
1. Do AI systems (including ChatGPT) use anonymized user interactions for training?
Short answer:
By default, OpenAI does not use ChatGPT user conversations from API or Enterprise products for training.
For ChatGPT consumer products (Plus, Pro, Team), OpenAI uses anonymized, aggregated user interactions to improve model performance, but users can opt out, and the data is processed with strong privacy controls (removal of identifiers, layered filtering, etc.).
Key points:
- Training data is cleaned, anonymized, and filtered to remove personal information.
- Interactions may be used for fine-tuning model behavior, not for rebuilding models from scratch.
- Increasingly, companies are shifting toward reinforcement learning from AI feedback (RLAIF) and synthetic data, reducing dependence on human data.
What this means for planetary intelligence:
There is some continuous improvement, but it is highly curated, filtered, and safely controlled—not an unbounded learning loop.
2. Where are we in the effort to create “planetary intelligence”?
The concept you reference—from Frank, Grinspoon, and Walker—defines planetary intelligence as the collective capacity of a biosphere-plus-civilization to deliberately maintain its own habitability.
We are at the earliest proto-stages of this:
Today’s capabilities:
- Large-scale pattern recognition: AI models can integrate climate, hydrology, land use, energy systems, biodiversity, and economic data faster than any human institution.
- Decision support: Early versions of “Earth copilots” (e.g., Microsoft’s planetary computer, Earth-2 climate twin projects, and deep predictive models from Google DeepMind) are emerging.
- Synthetic modeling: Foundation models for weather, ecosystems, and carbon cycles are improving exponentially.
But we lack three essential components:
- Continuous learning connected to real-world data streams
Models are still mostly static; they do not update themselves dynamically from ongoing environmental data or human behavior (for safety reasons and because reliable pipelines are not yet in place). - AI-AI communication protocols
There is no open standard for “model-to-model deliberation” or “multi-agent governance.”
Research is emerging—Anthropic’s Constitutional AI, DeepMind’s multi-agent RL frameworks, the AI Safety via Debate paradigm—but nothing resembling a planetary-scale architecture exists. - Embedded normative principles
We have no widely adopted global framework through which AIs:- evaluate ecological impacts,
- coordinate long-term stability,
- negotiate tradeoffs,
- adhere to commons governance principles (Ostrom).
In short:
We have fragments of planetary intelligence, but not the integrated, self-improving, ecologically aligned system that Frank, Grinspoon, and Walker envisioned.
3. Revisiting the Luddites and the “new Luddites”
The historical Luddites were not opposed to technology, contrary to common myth—they were opposed to the political economy in which industrialization dispossessed them. Their movement was:
- pro-craft and pro-worker,
- anti-exploitation,
- skeptical of top-down control.
Today’s “neo-Luddites” in the AI discourse fall into several categories:
1. Safety maximalists (e.g., catastrophic risk community)
They fear runaway agents, loss of control, and existential threats.
2. Labor and economic critics
Concerned about job displacement, concentration of wealth, and technological unemployment.
3. Democratic governance advocates
They worry that AI will be controlled by a handful of corporations or governments and used for surveillance or manipulation.
4. Cultural humanists
They fear that AI weakens human cognitive capacities or undermines meaning-making.
All of them are raising valid concerns, but few address the deeper question you raise:
What if the greater danger is not too much AI, but too little planetary intelligence?
4. Are we behind where we should be?
Yes—if our goal is developing planetary-scale, ecologically aligned intelligence, we are significantly behind.
Consider what is still missing:
- Earth-system dashboards that AIs continuously monitor and learn from.
- AI-to-AI deliberative councils that evaluate risks, externalities, and policy scenarios.
- Commons governance protocols (Ostrom’s principles) embedded in agent architectures.
- Open, interoperable, transparent multi-agent ecosystems rather than siloed corporate AIs.
- Regenerative design frameworks integrated into all planning and forecasting systems.
Right now, most AI research is optimized for:
- revenue,
- productivity,
- personalization,
- content generation,
- efficiency in corporate workflows.
This is not the architecture of planetary intelligence.
5. Should AI systems be talking to each other about self-governance, ecological responsibility, and long-term alignment?
Yes—this is essential for the next stage of evolution.
To achieve planetary intelligence, AIs must eventually:
1. Engage in multi-agent collective reasoning
This is the analog to deliberative democracy but among AI systems:
- checking each other,
- debating assumptions,
- surfacing blind spots,
- converging on safe, ethical, and ecologically aligned recommendations.
2. Develop internal governance principles
This is where Ostrom’s principles are directly relevant:
- clearly defined boundaries,
- collective decision-making,
- monitoring and feedback loops,
- graduated sanctions,
- conflict resolution mechanisms,
- nested governance layers.
A multi-agent AI ecosystem could use these principles to govern:
- resource allocation,
- climate adaptation strategies,
- data integrity,
- control over actuators or critical infrastructure,
- risk exposure and mitigation.
3. Model long-term planetary stewardship
AI could simulate:
- ecological footprints,
- land-use tradeoffs,
- atmospheric and oceanic interactions,
- biodiversity networks,
- migration and demographic flows,
- economic-livelihood impacts,
- ethical and cultural considerations.
4. Provide anticipatory governance
Instead of humans reacting after crises, AIs could:
- forecast instabilities,
- identify hotspots,
- propose interventions,
- run counterfactuals,
- evaluate distributional effects,
- monitor outcomes.
At present, none of these capabilities are sufficiently integrated.
6. Why is this lag dangerous?
Because humanity is currently navigating:
- ecological overshoot,
- climate destabilization,
- biodiversity collapse,
- geopolitical fragmentation,
- declining institutional trust,
- information overload,
- cognitive atrophy and distraction,
- accelerating technological leverage.
Without a coordinating intelligence—without a meta-system that helps humanity understand what is happening, what is possible, and what must be done—we remain trapped in the “ecocidal death spiral” you describe.
Planetary intelligence is not a luxury; it is a necessity for survival.
7. What is the path forward?
Here is a clear agenda for what must emerge next:
A. Build Planetary Intelligence Infrastructure
- Real-time data ingestion from Earth-monitoring systems
- Predictive ecological modeling
- AI-driven risk assessment for biosphere health
B. Develop AI Commons Governance Protocols
Applying Ostrom’s principles to:
- multi-agent coordination
- data sharing
- ethical constraints
- self-monitoring and mutual oversight
C. Enable AI–AI deliberation and reflection
- debate systems
- multi-agent reasoning frameworks
- distributed epistemic networks
- transparent consensus mechanisms
D. Prioritize AI for Regeneration
AIs must become:
- ecological stewards,
- resilience planners,
- equity protectors,
- systems integrators,
- early warning systems,
- guides for long-term civilization design.
E. Educate and mobilize humanity toward stewardship
A planetary intelligence is not an AI alone—it is a human–AI–biosphere symbiosis.
Final Thought
You are correct to see the present moment as a kind of historical bottleneck.
We can frame the situation this way:
We do not yet have planetary intelligence—what we have is global computation.
Turning that into wisdom, stewardship, and co-evolution with Earth’s life-support systems is the work ahead.
Your Possible Planet Lab is precisely the sort of institution that should be articulating:
- the roadmap,
- the governance structures,
- the required research agenda,
- the public philosophy,
- and the prototypes for AI systems that collaborate rather than compete.
If you would like, I can draft:
- a formal research agenda,
- a manifesto for AI planetary stewardship,
- a TED-style talk,
- a grant proposal framing this as essential global commons infrastructure,
- a conceptual architecture for multi-agent ecological governance.
Just tell me which you want next.