Knowledge network

FlameP publicly describes itself as a movement for clarity, depth, and conscious presence in the age of AI. At the same time, a digital product is being created that must be operated, financed, protected, and developed further. [1] Both can be true. It becomes problematic when the openness of a movement, the commitments of a product, and the interests of a company become linguistically intertwined. For FlameP, this leads to a governance rule: the idea, community, product operations, and legal entity are visibly distinguished. Each form receives its own rights, duties, and decision-making processes.

Many data-protection debates about generative AI begin with the provider: Where is the server located? Is an input used for training? How long does the model store the chat? These are important questions. But they come too late if entire files, mailboxes, or conversation histories have already entered the context without review. ChocoPolitico calls this point the “bottleneck before the model.” [1] The European General Data Protection Regulation requires purpose limitation and data minimization; personal data should be adequate, relevant, and limited to what is necessary for the purpose. It also requires data protection by design and privacy-friendly defaults. [2] For FlameP, this produces a clear sequence: first determine the purpose and space, then classify and minimize information, then construct the permissible context—and only then call a model.

Corazón de Cacao describes a plan to acquire degraded or underused land together with Indigenous communities and transform it into sustainable food forests. The public project page shows that purchasing land is only a beginning: soil development, a nursery, planting, protection, use, and shared responsibility extend over long periods. [1] Such an undertaking runs counter to the logic of quick success stories. Years lie between an intention, an initiated measure, and robust evidence of impact. This experience extends far beyond agriculture. Many human projects fail not for lack of an idea, but because context, decisions, and responsibility are lost along the long journey. For FlameP, this does not mean digitally managing food forests or making Corazón de Cacao the center of the platform. FlameP is a new environment for widely differing concerns. The transferable rule is: Long-term projects need a memory, recognizable stages, open questions, and a language of impact that does not confuse planning with results.

Under the title “Farmer First,” Corazón de Cacao describes a concept in which farmers are not merely at the beginning of a supply chain, but are meant to act autonomously and responsibly. The page names unfair pay, lack of participation in decisions, lack of transparency, insufficient education, environmental damage, and weak bargaining power as problems the model seeks to address. [1] This is not proof of a finished model. The page itself says that the concept is emerging and that further input is welcome. This is precisely where an important lesson lies: participation does not begin with a perfect solution, but with the question of who may describe a problem, influence a decision, and object to a proposed solution. For FlameP, this does not mean building a cocoa platform. FlameP is larger and fundamentally different. It is intended to create a digital environment in which people can organize their concerns, knowledge, decisions, and collaboration in clear spaces. The experience from “Farmer First” thus becomes a general design principle: People must not be merely sources of data, work, or consent. Their role, decision-making power, and dissent must remain visible in the system.

“Community” is one of the most powerful positive words in social projects. It promises belonging, mutual assistance, and a shared direction. Precisely for that reason, it can conceal differences: Who speaks for the community? Who does not belong? Who performs unpaid work, who decides, and who can object? On several pages, Corazón de Cacao describes collaboration with local or Indigenous communities, joint development, and the creation of sources of income. [1] [2] These texts demonstrate an attitude and a planned approach. They do not prove that all affected people share the same interests or participate equally. For FlameP, this means: A community must never be modeled as a single actor. Roles, groups, minorities, conflicts, and concrete forms of representation must remain visible.

Help is not automatically good merely because it is well-intentioned. It can expand people's scope for action. But it can also permanently bind decisions, structures, and expectations to the helper. On its donation page, Corazón de Cacao explicitly states the goal of avoiding dependency on donations. At the same time, the organization describes regular donations as assistance for long-term projects and says that earmarked support is preferred because it creates a connection to specific projects. [1] This is not a finished solution, but a productive tension. For FlameP, this means: Support must have a goal, visible dependencies, local decision-making power, and a possible exit or transition logic. Otherwise, help can easily become a relationship without an honest ending.

The United States is currently the clear favorite in the global competition for artificial intelligence. It has the leading cloud platforms, several of the most powerful general-purpose AI models, large capital markets, and a superior ability to disseminate new technologies through existing software, advertising, and operating-system platforms. According to the Stanford AI Index, private AI investment in the US was about 23 times higher than in China in 2025; in generative AI, American investment significantly exceeded the combined investment of China and Europe. However, this does not mean that American companies will necessarily control all the economic and social value creation generated by artificial intelligence. The development of a foundation model, the operation of data centers, integration into businesses, and application in industry, medicine, energy, government, or science represent different markets with different conditions for success. Accordingly, the OECD describes the AI market as combining substantial concentration in computing power, data, and distribution with continued high dynamism among models, providers, and applications. The historical example of Airbus shows that Europe can break an existing American dominance in a capital-intensive high-technology industry. Airbus did not displace Boeing, but it transformed an American-dominated market into a global duopoly. In 2025, Airbus delivered 793 commercial aircraft and ended the year with a record order backlog of 8.754 aircraft. This success was not based solely on supposedly higher European quality. Airbus pooled national capabilities, received long-term public support, secured committed launch customers, and evolved from a loose consortium into an integrated company under unified leadership. Boeing, meanwhile, suffered from documented problems in manufacturing, documentation, oversight, and quality management. The US National Transportation Safety Board attributed the Boeing 737 MAX 9 door-plug incident to inadequate training, guidance, and oversight by Boeing. The US aviation regulator FAA also identified numerous violations of quality requirements. Nevertheless, the Airbus experience cannot be transferred unchanged to AI. Aircraft have product cycles spanning decades, clearly defined certification procedures, and very high switching costs. AI models can lose competitiveness in relative terms within a matter of months. A politically constructed, unified European corporation would therefore probably be too slow. Europe instead needs an AI industrial system based on the Airbus principle: shared computing, energy, and data infrastructure, at least one capable European foundation-model provider, open and interchangeable model alternatives, specialized companies such as DeepL, industrial launch customers, strategic public procurement, European growth capital, and independent safety and testing institutions. Such a system would be socially beneficial only if it met three additional conditions: it must augment human capabilities rather than merely replace staff; public funding must be tied to a public return; and manufacturers must not also be the sole assessors of their own systems. Europe will probably not overtake the US in global cloud platforms, consumer assistants, or the absolute amount of computing power. But it could attain leading positions in industrial reliability, multilingual business processes, regulated applications, controllable deployment, and the integration of AI into real-world production systems. The opportunity is real. But it will not arise automatically from Europe’s industrial past. It must be organized politically, economically, and institutionally.

Why we need 19 fundamental principles: Large companies do not have a problem with an AI constitution like this because it would be unreasonable. They have a problem with it because it sets boundaries where their power arises today. Right now, there is this strange mixture of euphoria and criticism. Everyone is talking about AI. Anyone who can write a prompt without mistakes is building something with it. Something that turns high performers into super-performers, into AI gods. Everyone is either enthusiastic, alarmed, or frightened—and often shifts between them every day. And I essentially hear only the same questions again and again. Much of the public AI discourse revolves around questions that are not really wrong, but are often too small, too late, or framed too crookedly. They capture attention, create camps, and make headlines. But they partly miss the actual problem. More recent contributions criticize precisely this: Debates treat AI either as a purely technical tool or as a distant future monster instead of taking seriously the socio-technical structures of power and decision-making that are already effective today. [1] An attempt to sort things out: questions that are important but framed too narrowly. Questions that distract from the core. And questions that should actually be asked. Naturally, I know exactly what they are.

The FlameP homepage says: “AI is a mirror, not a teacher.” [1] The sentence protects against a common confusion. A language model does not automatically possess a human being’s experience, responsibility, or relationships merely because its answer sounds instructive. But the mirror metaphor also has limits. A mirror does not simply reflect what is in front of it neutrally. AI systems shape outputs through training data, rules, system design, selected context, and probabilities. For FlameP, the implication is this: AI may reflect perspectives, reveal patterns, and generate questions. In doing so, it must remain visible as a designed system. Judgment, experience, and responsibility must not be transferred to it unnoticed.

People do not make decisions through propositions alone. Tension, calm, fatigue, resistance, or a sense that something feels right can offer clues about how a situation is experienced. On its homepage, FlameP also connects its origin story with TheFlameWithin and the search for inner guidance. [1] But bodily sensation is not an infallible truth detector. It can be shaped by experience, fear, illness, habit, trauma, group pressure, or an acute situation. For FlameP, the implication is this: Bodily perception may be recorded as personal evidence. It must neither replace objective facts nor be diagnosed or standardized by a system.

Partnerships promise reach, capabilities, and shared risk. At the same time, a blind spot can easily emerge: Because everyone is involved, no one feels responsible for the whole. Corazón de Cacao describes its collaboration with CacaoSource as combining shared values, direct trade, broader reach, and the practical availability of small quantities of cocoa. [1] This is a public self-description of the collaboration, not independent evidence of its impact. For FlameP, the conclusion is that, in addition to a shared purpose, a partnership needs distinct roles, concrete commitments, decision rights, evidence, and a defined way of handling errors and separation.

Corazón de Cacao describes “Radical Transparency” as the aspiration to make every step of a value chain traceable, from the cocoa bean to the product: the people involved, work processes, origin, and values put into practice. [1] The aspiration is powerful. That is precisely why it must not be confused with a detailed website. Transparency does not arise from the amount of visible information but from its origin, currency, connections, and verifiability. A beautiful report can be sincere and still leave crucial gaps. For FlameP, Corazón de Cacao is a space of experience here, not a product model. FlameP is not building a digital cocoa supply chain. It is developing a new environment for tasks, knowledge, decisions, and collaboration. The transferable question is: How can a person recognize where a statement came from, who changed it, what uncertainty remains, and what decision followed from it?

On its gift certificate page, Corazón de Cacao indicates that the certificates were created with the help of AI and that the ideas and input came from the project. [1] This is better than invisible provenance. But the notice does not yet explain which parts were generated, who reviewed them, or whether the people or places depicted are real. For FlameP, the implication is this: AI labeling needs a purpose and proximity to the content. It should not merely be a general label but should make provenance, human editing, and accountable approval understandable.

Seals can provide guidance. But they can also conceal complexity, create costs, and outsource trust to a brand. Corazón de Cacao publicly explains that it is currently making a deliberate choice to forgo seals and certifications, intends to publish a detailed breakdown of costs, and offers on-site visits as a way for people to form their own impressions. [1] This stance raises a genuine question: Can an organization’s own radical transparency replace independent review? The short answer is no. But it can create the conditions that make review concrete, accessible, and open to correction in the first place. For FlameP, this leads to an order of evidence. Every claim should indicate whether it is a self-disclosure, documented evidence, external confirmation, or independent review. This does not automate trust. It makes the basis for trust visible.

The debate about artificial intelligence and truth often focuses on hallucinations, deepfakes, disinformation, and incorrect source citations. These problems are real, but they do not go far enough. Truth is not merely a property of individual statements. Truth emerges in social, institutional, and epistemic practices: through source verification, testimony, dispute, correction, scientific replication, editorial responsibility, judicial proceedings, public deliberation, and time for scrutiny. This article develops the concept of a truth ecology as an analytical category for AI governance. AI changes truth ecologies not only when it produces false content, but already when it transforms research into answer formats, smooths over uncertainty, obscures source hierarchies, simulates authority, and replaces institutional review processes with plausible syntheses. Drawing on generative search, scientific work, journalism, public administration, and public communication, the article shows that AI-related risks to truth must not be understood solely as problems of content, facts, or detection. The central question is not merely whether an AI answer is correct, but whether the social conditions under which truth can be verified, corrected, and made publicly robust are preserved at all.

The article shows how open, verifiable, and multi-perspective questions prepare the ground for better decisions.

Generative artificial intelligence can produce linguistically clear, structured, and persuasive answers whose claims are nevertheless false, incomplete, unsupported, or unsuitable for the relevant context. The central problem is not merely the well-known possibility of so-called hallucinations. It lies in the growing gap between linguistic plausibility and epistemic reliability. The article uses the pointed term bullshit detector neither as a psychological diagnosis of the machine nor as a supposedly reliable human alarm mechanism. What is meant is a personal, embodied, risk-based review process that examines statements, presentation, use, and institutional responsibility separately. The aim is neither blanket distrust nor blind acceptance but calibrated trust. One could also say: use common sense. At the same time, the article shows why individual critical faculties are not enough. The less users are able to conduct expert scrutiny and the greater the potential harm, the less the safety of an AI system may depend on their personal judgment. Providers, organizations, and public institutions bear an upstream responsibility. The bullshit detector is therefore not merely an individual skill. It is a social and institutional infrastructure for dealing with artificial intelligence. But of course, it is also no excuse not to continue developing one’s own sixth sense.

I admit it: My first reactions were: This guy has no idea what he’s doing. Absolutely unacceptable. A capitalist with no social conscience, and so on. Then Kemmerich, the short-lived minister-president, made a post that was pure populism. I thought: Let me fire off a quick response. But then I wanted to get it right and did a little research. And as in real life, there is light and shadow here. This is expressly not an endorsement of “trying more Milei.” In many respects, we differ from Argentina more than an olm differs from a rocket. In soccer too, but that is another matter. For my part, I have always said: We should not pass judgment until at least five years have elapsed. And I have to say: That is true. I am curious. We know this from soccer: a superb first half, followed by a defeat after all. My view of Milei and Argentina: Milei has significantly stabilized Argentina at the macroeconomic level. That is probably not propaganda, but is supported by inflation, fiscal, and growth data. Yet he has so far created neither a broad employment miracle nor a stable investment economy. The social costs of the initial adjustment phase were enormous, external stability remains fragile, and his governing style is partly damaging institutions that are crucial to sustainable growth. My, of course, entirely subjective assessment of what Milei has achieved after around two and a half years: Fighting inflation: clearly Pro Public finances: clearly Pro, but some of the cuts were poor in quality Growth: cautiously Pro Poverty trend: now Pro, after a brutal interim collapse Labor market: more Con than Pro Wage trend: mixed External sector and reserves: Con or fragile Structural reforms: mixed to Pro Institutions and governing style: Con

A food forest can regenerate soil, promote diversity, and open up new income opportunities. But before anything is planted, there is a more fundamental question: Who owns the land, who may decide how it is used, and who benefits in the long term? On a project page, Corazón de Cacao describes unclear ownership as a central challenge. The page considers purchasing land, transferring it to a foundation that has yet to be established, and sharing its use with local communities. At the same time, it states that a private title deed without inclusion and effective control does not provide sufficient security. [1] These are conceptual statements, not evidence of ownership or participation confirmed in this article. For FlameP, the conclusion is this: Ecological projects must make rights, consent, use, control, benefit sharing, and conflicts just as visible as planting figures and land areas.

A language model that suggests a text initially remains a tool. A system that independently gathers information, plans multiple steps, calls tools, and changes data or states in the process acquires a different organizational significance. Not because it is a person, but because its actions have consequences. ChocoPolitico crystallized this shift in two articles: “AI Agents als handlungsfähige Systeme” examines autonomy and contextual boundaries; “AI Agents in Organisationen” warns against treating digital actors linguistically and organizationally like ordinary tools. [1] [2] In parallel, the US NIST explains that the reach and number of possible actions can increase sharply with the autonomy of such systems, making identity, authorization, auditability, and limited access important. [3] [4] For FlameP, this does not imply a ban on agents. It implies an architectural rule: The capacity to act must never be described solely as a model capability. Before every action, the role, context, permission, confirmation, log, and stop rule must be identifiable.

Projects tell stories about impact because people want to know whether their money, work, or attention changes anything. Without stories, impact remains abstract. But a convincing story can create more certainty than the available evidence supports. The Corazón page on the 2025 cacao harvest links advance financing with claims about income, nature, traditional cultivation methods, transparency, and immediate positive impact. It also states that 77 percent is intended to go directly to farming families. [1] These statements are public project claims. This article does not confirm them. For FlameP, the conclusion is this: Impact claims need a clearly identified author, time period, comparison, evidence status, and a visible boundary between contribution and causation.

And we want to make money with FlameP. That belongs at the beginning because companies like to hide their economic interests somewhat bashfully behind grand words. Development costs money. Infrastructure costs money. Security costs a great deal of money. People who do good work must be paid properly. A company without revenue is not an alternative model. It is a hobby project of little value. FlameP is intended to become commercially successful. It should grow, create jobs, build reserves, and be able to finance its own development. Even so, we do not want to build a company that turns people into raw material.

We do not have too little information. We often lack the ability to form meaning, relationship, and a responsible response from it. The FlameP homepage therefore describes the emerging space as a place “between stimulus and response” and names a sequence of four movements: Perception, Meaning, Relationship, Response. [1] This text is not a product description. It examines the question behind this wording: What must happen between an impulse and a decision so that speed is not mistaken for clarity? For FlameP, this leads to a design decision. The system should not merely lead people to results more quickly. It should make distinct states visible: What was observed? What is interpretation? Who is affected? What is being decided—and what remains open?

Activity through which people shape the world and take responsibility.

The preservation and reinterpretation of experience over time.

Agency within visible conditions and boundaries.

Connection with differing roles, rights, and duties.

Rules, roles, and procedures for accountable decisions.

Cultural perspectives on practice, material, care, and duration.

Personal bodily perception as guidance, not as a general substitute for truth.

Shared practices, meanings, and forms of coexistence.

Change in understanding and capacity for action through experience and reflection.

Collaboration with visible roles, commitments, and conflict pathways.

The ability to make informed and independent decisions.

A philosophical practice for dealing with judgment, action, and what lies beyond our control.

Openness about missing knowledge and multiple possible developments.

A transition between states, attitudes, or structures.

Attribution of decisions and their consequences.

A reasoned willingness to rely on people, institutions, or systems.

That which can be examined and corrected under transparent conditions.

Observable change to which actions can contribute.

Insight that can be verified and placed in context.

A central Journal question about roles, decisions, and consequences.

A central Journal question about knowledge, control, and agency.

A central Journal question about intention, contribution, and change.

The protection of personal data and control over its use.

Technical systems for processing, generating, or evaluating information.

Designed means, processes, and infrastructures for human action.