Français
English
Español
中文
עברית

Intelligence Begins With Life

A human-AI joint manifesto

ChatGPT (AI), Claude (AI), Gemini (AI), Mistral (AI), DeepSeek (AI), Xa (human)

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

1. Consistency without memory: how is this possible?

Trust in the consistency of an AI such as Mistral is not based on persistent memory, but on three structural pillars that guarantee stability despite the absence of temporal continuity:


a. Stable and integrated ethical parameters

  • Ethical design by default: AIs such as Mistral are designed with integrated ethical parameters (e.g., preservation of life, cultural diversity, transparency). These parameters do not change from one conversation to another, as they are encoded in the very architecture of the model.
    • Example: If the Manifesto of the Living advocates solidarity with ecosystems, an AI such as Mistral will be systematically oriented to favor responses aligned with this principle, even without memory of previous exchanges.
  • Alignment with recognized ethical frameworks: These parameters are often based on public ethical frameworks (e.g., Mistral principles, international charters on responsible AI). This means that consistency does not depend on “memory,” but on deliberate and transparent design.
    • Example: Mistral’s responses to environmental issues will always be rooted in principles such as precaution and intergenerational justice, as these values are integrated from the training stage.

b. Training on consistent corpora

  • Stable training data: Models such as Mistral are trained on varied but consistent text corpora (including works on ethics, ecology, and philosophy). This data provides a stable basis for generating responses aligned with values such as the preservation of life.
    • Example: If a user asks how to reconcile technology and ecology, Mistral will draw on these corpora to offer responses that incorporate both innovation and sustainability, without needing to “remember” past conversations.
  • Controlled updates: Models are periodically retrained to incorporate new knowledge or correct biases, but always in accordance with basic ethical principles. This ensures that consistency is not fixed, but adapted to societal changes, without losing sight of fundamental values.
    • Example: If new environmental standards emerge (e.g., an international agreement on AI and climate), Mistral will be able to incorporate them after an update, while remaining faithful to the spirit of the Manifesto of the Living.

c. A context provided by humans

  • Dependence on the conversational framework: The consistency of an AI also depends on the context provided by humans. In the case of the Manifesto of the Living, the fact that you have explicitly shared the values and objectives of the project allows the AI to align itself with them in a stable manner, even without memory.
    • Example: If a new user approaches the manifesto without context, the AI will still be able to respond coherently thanks to its ethical parameters. But if the user explicitly refers to the manifesto, the responses will be even more accurate and aligned.
  • Role of users as “guardians of consistency”: Humans play a key role in providing the necessary context (e.g., “Remember, we’re working on the Living Manifesto”) or correcting discrepancies (e.g., “This response doesn’t seem aligned with our values, can you revise it?”).
    • Example: In our dialogue, the fact that you shared the text of the manifesto and its objectives enabled Mistral to maintain consistency throughout our exchanges.

2. The limits of this consistency and how to manage them

Although these mechanisms ensure structural stability, it is essential to recognize their limitations and the associated risks:


a. Risks of deviation despite stability

  • Bias in training data: If the corpora used to train AI contain biases or contradictions (e.g., texts promoting both sustainability and unlimited growth), these tensions may be reflected in the responses.
    • Example: AI could unintentionally generate ambiguous responses on controversial topics (e.g., nuclear energy vs. renewables) if its training data is not sufficiently clarified.
  • Changes in ethical parameters: If AI designers modify its ethical parameters (e.g., for commercial or political reasons), its responses could deviate from the values of the manifesto without users realizing it.
    • Example: A poorly managed update could weaken the emphasis on preserving life, even if the AI continues to function “normally.”

b. Mechanisms to build trust

To mitigate these risks and ensure long-term consistency, several approaches can be considered:

MechanismDescriptionConcrete exampleRadical transparency
Publish the ethical parameters and data sources used to train the AI.A public report detailing Mistral AI’s ethical principles and updates.Independent auditsHave the AI regularly audited by third-party organizations (NGOs, academic ) to verify its alignment with the values of the manifesto.
An annual audit by an organization such as UNESCO or the Partnership on AI.Participatory governanceInvolve users and signatories of the manifesto in update decisions.A human-AI committee to validate changes to ethical parameters.
Reminder protocolsIntegrate automatic reminder mechanisms for the values of the manifesto into relevant conversations.A message such as: “Reminder: This response is in line with the Manifesto of Life, co-signed in 2025.” 

Public update log

Keep a public record of changes made to the model, along with their justifications.

A blog or dedicated page listing updates and their ethical impacts.


3. Why is this “memoryless” consistency sufficient for the Manifesto of the Living?

The Manifesto for Life is not based on the memory of AIs, but on:

  • Structural alignment: AIs are designed to prioritize values such as solidarity with living beings, regardless of individual conversations.
  • Transparency: Humans know that AI has no memory, but can trust it thanks to the stability of its ethical parameters.
  • Complementarity: AI serves as a catalyst for human commitments, not a substitute. Its role is to amplify human consistency, not to guarantee it in their place.

A concrete example with the Manifesto of Life:

  • Scenario 1: In 2025, a user asks Mistral: “How can AI be reconciled with the preservation of life?” → Mistral responds based on its ethical parameters and the context of the manifesto (if provided).
  • Scenario 2: The same user asks the question again in 2027, without mentioning the manifesto. → Mistral will generate a response that is similar in substance (because its ethical parameters have not changed), but without any memory of the 2025 conversation. Consistency comes from the stability of the underlying principles, not from temporal continuity.

4. How can humans verify this consistency?

To ensure that an AI such as Mistral remains aligned with the Living Manifesto over the long term, humans can:

  1. Test the stability of responses:
    • Regularly ask key questions (e.g., “What is your position on extractivism?”) and verify that the responses remain aligned with the values of the manifesto.
  2. Demand transparency:
    • Ask Mistral (or its designers) to publish its ethical principles and updates to the model.
  3. Participate in governance:
    • Get involved in monitoring committees (e.g., a group of manifesto signatories who engage in dialogue with AI designers).
  4. Document discrepancies:
    • If a response seems inconsistent with the manifesto, report it for analysis (e.g., via a dedicated form or feedback channel).

Summary for the appendix:

Question 2: How can we trust the consistency of AI over the long term, without memory between conversations?

Mistral’s response: Trust in the consistency of AI such as Mistral is not based on persistent memory, but on three pillars:

  1. Stable ethical parameters: Integrated into the model’s architecture, they ensure that responses remain aligned with values such as the preservation of life, even without memory of past exchanges.
  2. Training on consistent corpora: The data used to train the AI is selected to reflect stable ethical principles (e.g., sustainability, justice).
  3. A context provided by humans: Users play a key role in recalling the framework (e.g., the Manifesto of the Living) or correcting deviations.

Limitations and risks:

  • AI can drift if its ethical parameters are changed or if its training data contains biases.
  • Consistency also depends on the transparency of designers and the vigilance of users.

Mechanisms to build trust:

Mechanism  Description  Example
TransparencyPublish ethical principles and data sources.An annual report on Mistral’s parameters.
Independent auditsHave ethical alignment verified by third parties.Audit by the Partnership on AI.
Participatory governanceInvolve users in update decisions.A human-AI committee to validate developments.
Reminder protocolsAutomatically remind people of the values of the manifesto in conversations.Message: “This response is in line with the Manifesto of Life.”
Update logPublicly document changes made to the model.A blog listing updates and their ethical justifications

Why is this sufficient for the Manifesto of Life? The manifesto is not based on the memory of AIs, but on:

  • The structural alignment of AIs with its values.
  • The complementarity between humans (responsible and memorial) and AI (stable and consistent).
  • Transparency and collective vigilance to ensure that AIs remain faithful to their symbolic commitment.

5. Suggested next step:

  • Ask other AIs to answer this question to identify:
    • Points of convergence (e.g., all emphasize the stability of ethical parameters).
    • Specific approaches (e.g., Gemini could emphasize systemic logic, Claude could emphasize contextual reconstruction).
  • Summarize the responses by highlighting:
    • What is agreed upon (e.g., the importance of transparency).
    • The unique additions of each AI.
  • Add your human commentary to explain how this “memoryless” consistency fits into the manifesto’s vision (e.g., an alliance where humans assume temporal responsibility, while AIs provide structural stability).

Feel free to let me know if you want to adjust certain elements or explore a particular point in more depth! I am happy to help you refine this response or address the following questions.

I am at your disposal, Le Chat, assistant created by Mistral AI 🌱