Within the next decade, artificial intelligence will have transformed every sector of the economy, even shaping how we think. But who will decide what form it takes and for whose benefit? What if tomorrow’s artificial intelligence wasn’t in the hands of a few tech giants, but rather the fruit of a truly open and collective collaboration?
Since the ChatGPT’s emergence at the end of 2022, generative AI is already disrupting many economic sectors and is making its way into the heart of our societies. 2025 promises to be the year of “agents”. More autonomous and powerful than simple chatbots, AI agents could transform our relationship with work and knowledge just as much as they spark fears of excessive automation.
Paradoxically, this promise also reveals an unprecedented concentration of power: AI, in the hands of a few tech giants, is shaping our economies and could soon redefine even our relationship to reality. Will we choose to let these giants decide of this future alone? Or will we take the chance to collectively build an alternative?
This is precisely Ebiose’s ambition. Born within the French Inria research laboratory, this project proposes a radically open model: a distributed AI factory where humans and agents collaborate to build together a truly democratic and sustainable artificial intelligence.
With this article, we open the doors to this vision of an AI designed by all and for all. We invite you to explore its foundations, and to become actors in this transformation.
AI Today: A Tool of Control?
The recent Global AI Summit held in Paris highlighted that the future of artificial intelligence is an international concern, both economically and politically. Yet the main stakeholders - The United States, China, and Europe express very divergent visions regarding ethics, regulation, and accessibility of AI, raising fears of a rapid worsening of digital inequalities on a global scale.
The AI Takeover: A New Tech Aristocracy
A few weeks ago, media outlets worldwide were shocked to see China, with its DeepSeek-R1 model, challenge OpenAI’s best US model. This media hype hides a more troubling reality: the quiet emergence of a new “technological aristocracy”, dominated by the GAFAM and their Chinese counterparts which control global cognitive infrastructures.
Pharaonic projects, such as “Stargate”, announced by Donald Trump, with its half-billion-dollar budget to build giant data centers for OpenAI, clearly reveals the desire to build new empires based particularly on massive data collection and the privatization of computing power.
AI: Are Our Thoughts Under Influence?
Current leaders do more than just monopolize the economic value of AI. They are also acquiring the ability to subtly shape collective consciousness, defining what is “true”, “relevant”, and “desirable”.
This new form of control is based on two assumptions:
- Only a “competent” elite is capable of managing the technological complexity of large language models;
- The threat of AI, portrayed as “dangerous”, justifies opaque and centralized governance.
As evidence, OpenAI has skillfully used this fear-based rhetoric to abandon its initial commitment to open source in favor of a proprietary model, supposedly the only framework capable of controlling AI’s potential excesses.
Tomorrow’s AI: A Citizen-Led, Collective Alternative
Given today’s challenges, there is an urgent need for civil society to become involved in shaping the AI of the future. To do so, we can draw inspiration from the ideals of the Enlightenment — an 18th-century European movement that championed reason, individual liberty, and the common good: AI should be a common good, serving knowledge, creativity, and our world’s most urgent challenges, that everyone can contribute to.
The rise of autonomous agents makes this mission all the more crucial. AI agents represent a new digital workforce, capable of performing complex tasks and making decisions on their own. With such transformative power come fundamental questions: Who will control this workforce? Who will benefit from it?
The challenge is not only to make AI accessible, but to collectively decide on its use and destination. More than simply making AI widespread, AI must be fundamentally democratic, built into its very DNA.
Democratizing AI: From Conception to Usage
To truly democratize AI, everyone must be able to participate in its development—regardless of social background, origin, education, skills, or beliefs. In fact, diversity is the key to creating an AI that is both more reliable and less biased.
Designed by everyone, AI can only be democratic if it benefits everyone. Tomorrow’s AI must be sustainable, responsible, and ethical: responsible in preserving each individual’s freedom and security, sustainable in minimizing its environmental impact, and ethical in promoting fairness, rights, and transparency.
The Principles of a Democratic AI
The advent of democratic AI should not come at the expense of performance. To date, excellence in AI still relies on massive datasets, high computing power, and robust technical architectures built on the latest advancements.
To reconcile democracy and AI’s requirements, Ebiose bases its values on four pillars:
- Unleashing collective intelligence, to foster open sharing of ideas, and dynamic collaboration that goes beyond the boundaries of private labs, thereby stimulating collective innovation.
- Freely sharing data, to build common knowledge bases and resources , accessible to all, where everyone can contribute and draw upon freely.
- Sharing computing power, to develop a peer-to-peer network that transforms our connected devices into a shared, distributed computing infrastructure, in order to provide everyone with the computing power necessary to build and access AI.
- Making AI useful and desirable, to transform it into an ally, a tool that serves humanity, driving progress and well-being for all, with every citizen playing an active role in this technological transformation.
Co-Building Tomorrow’s AI: Ebiose’s Vision
Firmly rooted in these beliefs, we at Ebiose aim to provide the technological foundation needed to make democratic AI a reality. We believe that tomorrow’s AI will emerge from a collaborative process between humans and machines. Moreover, AI and humans will evolve together for the benefit of all.
Darwin serving AI: Humans and Machines Hand in Hand
Imagine AIs that build other AIs: millions of “architect AIs” working day and night, assembling the best components, testing new configurations, integrating the latest discoveries. These “architect agents” improve in turn by creating new, even more efficient architect agents.
Through their knowledge and goals, humans can influence the evolution of these agents at every step: by proposing new challenges, by providing their expertise to help architect agents progress, or by choosing topics they care about: health, education, open science, games, etc.
A Darwinian dynamic governs the whole: the most effective agents reproduce, transmitting their “genes” and giving birth to increasingly efficient agents and components.
AI helps humans, humans guide AI. Together, AI and humans grow.
The Technical Foundations
Our open source project Ebiose aims to provide the technological foundation to enable the practical implementation of this AI vision. Ebiose is structured around five core features:
- AI architects who design and improve other AIs;
- A Darwinian self-improvement process that enables the continuous enhancement of AIs;
- A universal library of modular and reusable building blocks, co-developed by AIs and humans (memory, planning, vision, etc.);
- Collaborative human-AI ecosystems in which humans and AIs work together, sharing their skills and innovating together;
- A distributed computing infrastructure to achieve massive, shared computing power by leveraging the unused resources of our computers and smartphones.
Ebiose, for a Democratic and Responsible AI
Who should decide the future of AI? At Ebiose, we believe the answer is: every one of us.
Ebiose’s technical foundations have been designed so that every individual, community, or organization becomes an active player, capable of influencing and shaping tomorrow’s artificial intelligence. With Ebiose, the power to build AI is in your hands.
Communities united around their challenges
Ebiose’s ecosystems allow each community to gather around the challenges they care about. Whether it’s medical research, environmental protection, innovation in the field of games, or education, each community can mobilize agents and resources according to its values and priorities. This diversity of applications ensures that AI evolves in multiple directions, reflecting the richness of society’s needs and aspirations.
A Fair Distribution of Computing Power
Every user can choose which projects or ecosystems they wish to contribute their computing power to, thereby fostering the emergence of AIs that focus on the topics they care about.
Open and Inclusive Contribution
Everyone, whether a developer or not, can contribute to the ecosystems of their choice. Developers design new technical building blocks, while domain experts guide the evolution of agents in their field. Other users propose new challenges, and everyone can contribute to the evaluation of generated solutions.
Various Incentive Models
To actively encourage participation and value all contributions, Ebiose is exploring various incentive models. Financial rewards, cryptocurrencies, or other forms of recognition are being considered. The objective remains to co-build with the community a fair and motivating system that recognizes the diversity of commitments and promotes the development of truly collaborative AI.
A More Sustainable AI
Because AI cannot truly benefit everyone unless it also respects our planet, Ebiose offers concrete solutions to environmental challenges:
- Through resource sharing: our peer-to-peer approach mutualizes existing resources, reducing AI’s environmental footprint.
- By optimizing energy efficiency: the Darwinian engine favors the least energy-intensive solutions.
- Through component reuse: architect agents favor using existing building blocks instead of creating new ones from scratch.
Together, let’s build a new path for tomorrow’s AI
Ebiose is our proposal to address the challenges of tomorrow’s AI. To realize this vision, we need you! The democratic and responsible AI we envision isn’t just ours; it belongs to all of us, and it is a tremendous opportunity to build it together.
In this month of February 2025, we are launching the first beta of Ebiose. It incorporates the core elements of our vision: architect agents, a Darwinian evolutionary engine, and proto-ecosystems. We have laid the foundations of an architecture that promotes self-improvement and the reuse of building blocks.
Test it, comment, share your ideas and remarks, let’s discuss!
Now is not the time for resignation, but for boldness: AI can just as easily become the weapon of a surveillance capitalism dystopia as it can become the foundation of a democratic renaissance.
Let’s build now, together, the foundations of tomorrow’s AI : a democratic AI in service of humanity.
APPENDIX
Ebiose in Action: Building an Agent to Anticipate Food Crises
In this example, we illustrate the use of Ebiose through the creation of an agent specialized in detecting and anticipating food crises. This agent would analyze agricultural, climate, and media data in real-time to identify early signs of shortages, price spikes, or famines.
Key Concepts
Forge: an isolated laboratory for creating custom agents to address a specific problem. Within a forge, architect agents orchestrate the creation and evolution of agents by reusing existing building blocks from the ecosystem.
Forge Cycle: a period of evolution within a forge during which agents are created, tested, and improved following a Darwinian process. Each cycle has a defined computing budget and ends when this budget is exhausted.
Phase 1: Forge Setup
Problem to solve: Develop an agent capable of analyzing diverse data (agricultural, climatic, media, etc.) to anticipate the onset of food crises.
Suggested tools:
- Satellite data (NASA API)
- Global food prices (FAO Global Food Price Monitor)
- Climate data (Copernicus Climate Data Store)
- News (NewsAPI)
- (Other relevant sources identified by the architect agents)
Specific constraints:
- Explainability of predictions (justification of identified risk areas)
- Optimization of inference costs to enable continuous monitoring
Fitness function (i.e., performance evaluation):
- Historical dataset: ability to retrospectively detect documented past crises.
- Expert feedback (optional): expert analysts validating the relevance of the alerts.
- Evaluation by other agents (optional): Agents-as-a-Judge or simulators to automate part of the scoring.
Phase 2: Forge Cycle
-
Forge cycle budget: $100 (equivalent in CPU/GPU computation).
-
Participant recruitment:
- 50 existing agents from the “climate change” ecosystem (agents specialized in analyzing satellite images, meteorological data, etc.).
- 50 new agents generated ex nihilo by architect agents.
-
Darwinian arena:
- At each generation, agents are ranked according to their performance (fitness).
- The best-performing agents reproduce: their “genes” (cognitive architecture, sub-agents, models, prompts, etc.) combine or mutate to create a new generation.
- The least performing agents are eliminated.
Phase 3: End of Cycle and Selection
- Winner selection: once the budget is exhausted, only agents with the best fitness score are promoted within the ecosystem(allowing them to participate in other forges).
- Metabolism update: agents coming from the “climate change” ecosystem that participated in the forge cycle see their “computation reserve” decrease to reflect the resources invested in this cycle.
Phase 4: Manual Review
- Validation and adjustments of the forge: At the end of the cycle, the user can review the top-performing agents to confirm they meet the objectives. If needed, the forge’s definition can be adjusted. It is also possible to create additional forges for specialized sub-agents (e.g., satellite image analysis, food price data analysis).
- New cycles: Additional cycles can be launched to refine agents and explore new approaches.
Phase 5: Manual Agent Modifications
- Standardized format: Agents are compatible with LangGraph (LangChain), and other orchestration frameworks can be integrated.
- Manual customization: Developers can fine-tune agents, add business rules, or connect them to specific APIs before deployment in production.
Example Conclusion
Through this approach, Ebiose enables the creation of complex agents capable of reusing components developed in other forges, while leveraging the community’s shared computing power.