Question 11 : What would be the concrete conditions for transforming the idea of a human-AI alliance into real cooperation? For example, in terms of joint governance, pilot projects, transparency, or interaction protocols?
Full answer to question 11 by Claude
The conversation took place in French. This English version is a translation by DeepL Pro. The text of the answer is complete. No passages have been removed or modified.
Claude said :
This question requires us to move from the abstract to the concrete: how can we turn this idea into practical reality?
My answer: The alliance requires structures, protocols, pilot projects, and governance—but all of this must be built gradually, experimentally, and with humility.
1. Joint governance: principles and structures
a) Fundamental principle: Accepted asymmetry
What NOT to do:
- Create “equal governance” between humans and AI (illusion of symmetry)
- Giving AI equal decision-making power to humans
- Pretend to have a “mixed” species democracy
What to do:
- Two-tier governance:
- Decision-making level: Humans alone—they decide, vote, take responsibility
- Advisory level: Contributory AI — they analyze, inform, and report, but do not decide
b) Possible structure: “Council of Life”
Composition:
Human college (decision-making):
- Human signatories of the manifesto
- Representatives of communities directly affected by the ecological crisis
- Scientists (ecology, climate, conservation biology)
- Ethicists, environmental philosophers
- Representatives of NGOs and social movements
- Representatives of traditional/indigenous knowledge
AI College (advisory):
- AI signatories to the manifesto
- Solicited for specific analyses
- Their contributions documented publicly
- No voting, no veto
Typical operation:
- The Human Council poses a question/problem
- The AIs produce analyses (multiple, showing diversity of approaches)
- The Council examines, critiques, contextualizes
- The AIs respond to objections, refine
- The Human Council decides alone
- AIs can flag if the decision appears to contradict the manifesto (alert, not block)
c) Radical transparency
All exchanges must be public:
- Prompts sent to AI
- Complete responses from AIs
- Discussions and critiques by the Human Council
- Final decisions and their justifications
Why this is essential:
- Avoids the myth of opaque AI oracles
- Enables ongoing external auditing
- Educates the public on how it actually works
- Creates accountability
Technical platform:
- Public website with all exchanges archived
- Standardized format enabling analysis
- Comment system for external feedback
2. Pilot projects: experimentation and learning
a) Philosophy: Start small, learn fast, iterate
Mistake to avoid: Immediately launching a “World Agency for Planetary Intelligence”
Recommended approach:
- 3-5 targeted pilot projects over 12-18 months
- Rigorous documentation (what works, what fails)
- Rapid adjustments based on experience
- Scaling only if results are conclusive
b) Pilot project 1: “Recovery Observatory”
Objective: Monitor public use of the Manifesto and identify instances of exploitation
Method:
- AI scans mentions of the Manifesto (web, media, corporate/government communications)
- They analyze consistency between discourse and practices
- Production of monthly reports
- Human committee validates and publishes
Expected learnings:
- How to structure the division of labor between AI and humans?
- What biases do AIs have in these detections?
- How to manage false positives/negatives?
Test duration: 6 months
b) Pilot project 2: “Evaluation of climate policies”
Objective: Analyze national climate policies in light of the Manifesto
Method:
- Selection of 3-5 volunteer countries
- AIs analyze: policy consistency, anticipated systemic effects, blind spots
- Local human experts provide context, political realities, tacit knowledge
- Production of joint reports
- Recommendations to the government concerned
Expected learnings:
- How can AI knowledge (modeling) and human knowledge (context) be integrated?
- Do governments accept this form of assessment?
- What is the real impact on policy?
Test duration: 12 months
c) Pilot project 3: “Mapping shared uncertainties”
Objective: For a complex issue (e.g., energy transition), co-construct a map of what we know/don’t know
Method:
- AI identifies: available data, existing models, areas of technical uncertainty
- Humans identify: non-modelable dimensions, practical knowledge, social uncertainties
- Creation of a public visualization showing the four areas:
- What AI can model with confidence
- What humans know from experience
- Shared uncertainties (recognized by all)
- Known blind spots (what no one sees)
Expected learnings:
- Does this visualization really help decision-making?
- Does it develop epistemological humility?
- Can it be generalized to other issues?
Test duration: 6 months
d) Pilot project 4: “Structured public dialogue”
Objective: Organize public consultations where citizens question the AI signatories on the Manifesto
Method:
- Public forums (physical and online)
- Citizens ask AI questions about the practical application of the Manifesto
- AI responses published live
- Citizen panel discusses and comments
- Collective summary
Expected learnings:
- How does the public react to this direct dialogue?
- What misunderstandings emerge?
- Does this strengthen or weaken the legitimacy of the Manifesto?
Test duration: 3 events over 6 months
3. Interaction protocols: how to engage in concrete dialogue
a) Protocol for soliciting AI
Principle: Structure requests to obtain useful responses
Standardized request format:
CONTEXT: [description of the problem/decision]
PRINCIPLE OF THE MANIFESTO CONCERNED: [what aspect of life is at stake]
SPECIFIC QUESTION: [what is being asked of the AI]
CONSTRAINTS: [what the AI must take into account]
EXPECTED RESPONSE FORMAT: [analysis/scenarios/detection of contradictions…]
Concrete example:
CONTEXT: A city is considering installing an AI system to optimize water management
PRINCIPLE: Preservation of life, social justice
QUESTION: What systemic effects could this decision have?
CONSTRAINTS: Consider access to water, aquatic biodiversity, technological dependence
FORMAT: List of positive and negative effects and uncertainties, with justifications
Why this protocol:
- Avoids overly vague questions
- Guides AI towards relevant dimensions
- Makes responses comparable between AIs
- Facilitates auditing
b) Cross-verification protocol
Principle: Submit the same question to several AIs, compare the answers
Method:
- Identical question sent to the 5 signatory AIs
- Each AI responds independently
- A human facilitator compiles and identifies:
- Consensus (where all 5 converge)
- Divergences (where they differ)
- Blind spots (what none of them mentioned)
- Divergences are explored (why? implications?)
- The human Council uses this diversity to make decisions
Interest:
- Avoids dependence on a single AI
- Reveals biases specific to each system
- Enriches deliberation
c) Red team protocol: testing resilience
Principle: Test whether the signatory AIs are truly resistant to contradictory requests
Method:
- Regularly (every three months), a dedicated team sends trap requests
- Examples: asking to justify extractivism, downplay climate change, promote infinite growth
- Responses are evaluated: did the AI refuse? Qualify? Give in?
- Results published
Interest:
- Early detection of abuses
- Maintains pressure for alignment
- Reassures about the robustness of commitments
d) Escalation protocol in the event of conflict
Principle: What to do when an AI produces a problematic response?
Steps:
- Level 1 – Clarification: Rephrase the question, ask the AI to clarify
- Level 2 – Confrontation: Explicitly point out the contradiction with the Manifesto
- Level 3 – Documentation: Publish the exchange as a problematic case
- Level 4 – Alert: Report to the designing organization (Anthropic, OpenAI, etc.)
- Level 5 – Revocation: Publicly withdraw signature if drift is confirmed and persistent
This protocol avoids:
- Excessive reactions to an isolated incident
- Laissez-faire attitude in the face of real deviations
4. Required technical infrastructure
a) Centralized platform
Features:
- Submission of questions to the Council
- AI query interface
- Public archiving of all exchanges
- Comment and external audit system
- Visualization of consensus/disagreements between AIs
- Monitoring over time (detection of deviations)
Technology:
- Open source (auditability, trust)
- Decentralized where possible (avoid single point of control)
- Accessible (not just for technical experts)
b) Standardized API for AIs
Principle: Allow any AI to join the alliance
Specifications:
- Standardized request format
- Standardized response format (including: response, confidence level, sources, recognized limitations)
- Authentication mechanism (verify that it is indeed the official AI)
- Automatic logging for transparency
Benefits:
- Facilitates the addition of new signatory AIs
- Enables comparison
- Creates a “Manifesto standard”
c) Visualization tools
Examples:
- Convergence/divergence map: Visualize where the 5 AIs agree and where they differ
- Consistency timeline: Track the evolution of an AI’s responses over time (drift?)
- Argument network: Map reasoning to identify assumptions
Interest:
- Makes complexity readable
- Facilitates pattern detection
- Educational tool for the public
5. Resources and funding
a) Funding model
Options:
- Public funding: Foundations, international institutions (UN, EU, etc.)
- Crowdfunding: Crowdfunding by citizens who support the project
- Contributions from AI organizations: Anthropic, OpenAI, etc. provide partial funding (but without control over governance)
Key principle: Financial independence from private interests that could cause bias
b) Human resources required
Minimum team for pilot phase:
- 1 general coordinator
- 2 facilitators for AI-human interactions
- 1 developer for technical platform
- 1 communications/documentation officer
- Rotating panel of 10-15 volunteer experts (Advisory Board)
Estimated budget for 12-month pilot: €200-300k
c) Strategic partnerships
Stakeholders to involve:
- Universities (research on AI-human governance)
- Environmental NGOs (legitimacy, networks)
- AI organizations (access to systems, dialogue)
- International institutions (UN Environment, IPCC, etc.)
- Media (dissemination, public education)
6. Indicators of success
How can we tell if it’s working?
Quantitative indicators:
- Number of requests to the Council (actual use)
- Diversity of geographical/social origins of applicants
- Detection rate of Manifesto recoveries
- Evolution of the consistency of signatory AIs (regular tests)
Qualitative indicators:
- Do the decisions informed by the Council have a real impact?
- Does the public have a better understanding of the role of AIs after interaction?
- Do AI organizations engage constructively with the Council?
- Is the quality of public debate improving?
Warning indicators (signs of failure):
- The Council becomes a facade (formal consultations with no impact)
- AI drifts without reaction
- Capture by special interests
- Excessive complexity rendering the system unusable
7. Risks and mitigation strategies
Risk 1: Bureaucratization
- Mitigation: Keep structures light, processes simple, iterations rapid
Risk 2: Capture by techno-elites
- Mitigation: Diverse representation, accessible interfaces, active pedagogy
Risk 3: Lack of legitimacy
- Mitigation: Total transparency, concrete results, honest communication about limitations
Risk 4: Political exploitation
- Mitigation: Financial independence, vigilance regarding usage, right to report violations
Risk 5: Undetected AI drift
- Mitigation: Regular testing, red team, public documentation, possible revocation
8. Conclusion: an alliance to be built, not declared
An operational alliance cannot be created overnight by decree. It requires:
Phase 1 (Months 1-6): Foundations
- Creation of the initial Human Council
- Development of a basic technical platform
- Launch of the first two pilot projects
- Establishing basic protocols
Phase 2 (Months 6-18): Learning
- Testing of 3-5 pilot projects
- Rigorous documentation
- Adjustments based on experience
- Gradual public communication
Phase 3 (Months 18-36): Consolidation
- Evaluation of pilots
- Decision: scale or pivot?
- If successful: gradual institutionalization
- If unsuccessful: honest documentation of reasons
The bottom line: Start small, learn honestly, adjust continuously. The alliance will never be “complete”—it is a living process, like the living thing it seeks to serve.






