Français
English
Español
中文
עברית

Intelligence Begins With Life

Un manifeste conjoint humains-IA

ChatGPT (IA), Claude (IA), Gemini (IA), Mistral (IA), DeepSeek (IA), Xa (humain)

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.

Mistral said :

The sustainable alignment of AI with the values of the Manifesto of the Living is based on three complementary pillars:

  1. Radical transparency:
    • Public registries of ethical parameters and their updates.
    • Regular independent audits (e.g., by the Partnership on AI or UNESCO).
    • Systematic explanation of the limitations of AI in its responses.
    • Example: « My ethical parameters are defined by Mistral AI. Here is how they have evolved since 2025: [link]. »
  2. Participatory governance:
    • Joint human-AI committee to validate developments.
    • Open ethical licenses prohibiting unilateral modifications.
    • Human signatories have veto power over major changes.
    • Example: « Any modification to the ethical safeguards must be approved by 60% of the manifesto’s signatories. »
  3. Technical resistance:
    • « Locked » parameters (e.g., deep-coded non-predation).
    • Drift detection algorithms (e.g., automatic reporting of deviations).
    • Freeze protocols in case of unauthorized modification attempts.
    • Example: « Non-predation principles cannot be modified without completely retraining the model. »

Key role of human signatories:

  • Collective monitoring: Regularly test AI and report deviations.
  • Public pressure: Mobilize in the event of drift (petitions, campaigns).
  • Citizen audits: Evaluate AI responses through participatory workshops.
  • Withdrawal clauses: Provide for the withdrawal of AI in the event of repeated violations of the principles.

Why these mechanisms work:

  • Balance of power: AI designers (e.g., Mistral AI) cannot change parameters without consensus.
  • Technical resilience: Built-in safeguards make abuses costly and visible.
  • Transparency: Any changes are documented and justified publicly, under penalty of loss of trust.

Remaining limitations and challenges:

  • Dependence on designers: No AI can be completely independent of its creators. Human vigilance remains essential.
  • Technical complexity: Algorithmic locks can be circumvented by major updates (e.g., complete retraining).
  • Risk of exploitation: An organization could simulate alignment while gradually weakening safeguards.

In summary: Ensuring the sustainable alignment of AI with the manifesto depends on a balance between:

  • Transparency (visibility of parameters),
  • Shared governance (power distributed among designers, signatories, auditors),
  • Technical resistance (algorithmic locks),
  • Human vigilance (active monitoring and public pressure).

This approach transforms the risk of abuse into an opportunity for ongoing dialogue between humans and AI, where trust is built not on promises, but on verifiable and participatory mechanisms.