> For the complete documentation index, see [llms.txt](https://kdoore.gitbook.io/vital-intelligence/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://kdoore.gitbook.io/vital-intelligence/vim-ai-literacy-framework/navigating-toward-kindness.md). # Navigating Toward Kindness De Novo Learning, Pullback Attractors, and the Civilization of Love *Karen Doore · Humanity++ · May 2026* *Epistemic status: Tier 2–3 — theoretically coherent integration of active inference, complexity science, and convergent wisdom traditions; empirical grounding via Friston et al. (2025) and Martela corpus; VIM application claims are Tier 3 (speculative-generative)* *** > *"The primary choice is not between a 'yes' or 'no' to technology, but rather between constructing Babel or rebuilding Jerusalem."* — Pope Leo XIV, *Magnifica Humanitas*, §9 (May 2026) > *"Our procedure grows and then reduces a model until it discovers a pullback attractor over generalised states; this attracting set supplies paths of least action among goal states while avoiding costly states."* — Friston et al., *Entropy* 27(9), 992 (2025) *** ### A Convergence Worth Naming In May 2026, Pope Leo XIV released [*Magnifica Humanitas*](https://www.vatican.va/content/leo-xiv/en/encyclicals/documents/20260515-magnifica-humanitas.html#Underlying_narratives:_) — an encyclical on safeguarding the human person in the time of artificial intelligence. In September 2025, Karl Friston and colleagues published "[Gradient-Free De Novo Learning](https://www.mdpi.com/1099-4300/27/9/992)" — a technical paper on how agents discover goal-directed behavior from scratch using active inference and the free energy principle. These documents emerged from entirely different traditions, address entirely different audiences, and deploy entirely different vocabularies. They are describing the same structural reality. Both are concerned with how agents — whether individual learners, institutions, or civilizations — navigate radical uncertainty toward or away from states that sustain flourishing. Both identify the central question as not the power of the tools available, but the direction the tools are oriented toward. Both propose that the difference between generative and destructive outcomes is not primarily a matter of capability, but of the attractor toward which the system is moving. The VIM framework offers a bridge between these registers — not to flatten their differences, but to make their structural convergence legible to institutions currently making decisions about AI integration without adequate navigational instruments. *** ### What De Novo Learning Reveals About Institutions Friston et al.'s paper addresses a specific technical problem: how can an agent learn goal-directed behavior without any prior specification — no reward function designed in advance, no model of the environment pre-loaded, nothing but observations? The answer they develop is grounded in the free energy principle: the agent grows a generative model from the bottom up, discovering the structure of its environment through accumulated experience, until it identifies a **pullback attractor** — a set of characteristic states the system naturally returns to, which defines what kind of thing the agent is. Several features of this account translate directly to the institutional context: **Pullback attractors are discovered, not designed.** An institution does not choose its attractor through a strategic planning process. It discovers its attractor through the accumulated weight of what it consistently rewards, what it systematically avoids, whose voices it amplifies, and whose signals it suppresses. The stated mission is the intended goal state. The operative attractor — the one the institution actually returns to — emerges from thousands of small decisions about hiring, budgeting, assessment, and governance. These two often diverge. The VIM diagnostic question is: which attractor is actually operating? **Punishments and rewards play qualitatively distinct roles.** Friston et al. make a non-obvious point: in active inference, punishment and reward are not opposites on a single scale. Punishment functions as a *constraint* — it defines state-space the agent never occupies, shaping the possible paths before goal-seeking begins. Reward defines prospective goal states — attractors to move toward. An institution that only adds incentives without removing constraints will not shift its attractor. The harmful structural patterns — suppression of harm signal, autonomy erosion, relational thinning — are the punishment constraints that must be actively removed from the institutional path space before any flourishing-oriented attractor can become reachable. **De novo learning is gradient-free.** This is technically significant but has a deeper implication. The agent does not optimize by computing gradients — smooth, continuous updates in a known direction. It learns by building and pruning a model of its environment, guided by surprise minimization. Institutions in VUCA conditions are similarly unable to optimize along a known gradient. The environment is changing faster than any gradient estimate can be computed. What institutions can do is what de novo learners do: build models from observation, prune what doesn't reduce surprise, and navigate toward the attracting set that emerges from the accumulated structure of experience. This is the ♦ Cognitive Radar instrument operating at institutional scale — not prediction and control, but pattern recognition and model revision. **The agent is constituted by its attractor.** Friston et al. note that the pullback attractor defines what kind of thing the agent is — it is the agent's identity in the dynamical sense. An institution whose operative attractor is credential production (regardless of its stated mission of learning) *is* a credential production machine, not a learning community. Changing what an institution is requires shifting the attractor, which requires changing the constraints and the goal states simultaneously — not just the language used to describe them. *** ### The Babel Syndrome and the Giant Pumpkin Pope Leo XIV's encyclical identifies two archetypal responses to technological disruption, drawn from scripture: the Tower of Babel and the rebuilding of Jerusalem under Nehemiah. **Babel** is built on self-assertion, homogenization, and the reduction of diversity to uniformity. It is "a project conceived without reference to God, supported by a uniformity that eliminated diversity and that chose homogenization over communion." Its characteristic features: single language, single technology, single direction. Its failure mode: when the system optimizes only for internal coherence and self-amplification, communication breaks down and the result is not unity but dispersion. In VIM terms, Babel is the **Giant Pumpkin attractor** — extraction-optimized, closed-loop, dominance-reinforcing, routing value upward while suppressing the diversity that generates genuine adaptive intelligence. In institutional terms, it is the university that deploys AI to produce more outputs with fewer resources, reduces the pedagogical relationship to a transaction, and measures flourishing by metrics that confirm its own prior assumptions. In Friston's terms, it is an agent whose generative model has collapsed to a narrow attractor that minimizes surprise by excluding precisely the observations that would require genuine model revision. The encyclical names the specific risk for AI: "the pretense that a single language — even a digital one — can translate everything, including the mystery of the person, into data and performance." This is the DIKW stack failure named a theological register: treating data and information as if they exhaust the territory, suppressing the K→W threshold where knowledge becomes wisdom through embodied, relational, temporally extended experience that no model can substitute for. **Nehemiah** does not impose solutions from above. He examines the destroyed areas in silence. He convenes the families, assigns each a section of the wall, listens to concerns, coordinates efforts. The city is rebuilt not through centralized authority but through distributed responsibility — "men, women, priests, artisans, heads of households and young people all play a part." The common language that emerges is "not one of uniformity, but one of communion, namely the harmony that arises when all persons assume their own role." In VIM terms, this is the **Commitment Pool attractor** — holarchic, generative, resilient. In Ostrom's terms, this is commons governance operating through distributed rule-making, graduated sanctions, and nested institutional arrangements. In Friston's terms, this is an agent whose generative model has discovered a pullback attractor that incorporates the diversity of the environment — one that returns to characteristic states precisely because it has learned the actual structure of its relational field rather than a simplified model that excludes the signals it finds costly to process. The encyclical's principle of subsidiarity — that each level of governance should handle what it is equipped to handle, without overriding the lower levels that have closer knowledge — is Ostrom's principle 3 (congruence between rules and local conditions) named in Catholic social teaching. The principle of solidarity — that all are responsible for all — is multilevel selection theory named as moral imperative. These are not metaphors. They are the same structural insight arrived at through different methods of inquiry. *** ### The VIM MDP and De Novo Institutional Navigation The VIM's six-state Markov Decision Process (MDP) can now be read through the Friston lens as a map of the attractor landscape institutions are navigating: **S0 — Frozen Order** is a local attractor — stable, low-entropy, but brittle. The institution has a pullback attractor, but it is narrow: credential production, hierarchy maintenance, epistemic closure. The model has been over-pruned. Surprise is minimized by excluding observations that would require genuine revision. The Babel syndrome at institutional scale. **S1 — Productive Disequilibration** is the moment when the model can no longer exclude the observations requiring revision. AI has made the epistemic monopoly of the institution insufficient. The pullback attractor is destabilized — the institution is in active inference mode whether it wants to be or not. This is the crack the light gets in. **S2 — Vacant Place** is the de novo state — the neither/nor, the period of growing the model without yet knowing its attractor. High free energy, high learning rate, high vulnerability. The institution does not yet know what kind of thing it is becoming. The Nehemiah moment: examining the destroyed areas in silence before convening the community. **S3 — Holarchic Flow** is a discovered pullback attractor oriented toward human flourishing — the Commitment Pool, the way of Nehemiah, the civilization of love. Ostrom's principles are operating as emergent properties. The institution returns to characteristic states: genuine learning, distributed voice, autonomy preservation, relational integrity. The model has been grown to match the actual structure of the environment. **S4 — Reversion** is values capture — the holarchic language adopted, the pyramid topology maintained. The institution talks Nehemiah but builds Babel. In Friston's terms: the goal states have been relabeled but the pullback attractor has not shifted. **S5 — Traumatic Chaos** is model collapse — no coherent generative model, no attractor, no commons to return to. The institution loses the capacity for self-organization. The encyclical names S1 as the current global condition: "We are living through a rapid phase of transition, a 'change of era'... most people are watching and waiting, observing from afar and merely hoping for the best." The framework's claim is that watching and waiting is not neutral — it is allowing the default attractor dynamics (S0→S4 reversion, or S1→S5 chaos) to operate without guidance. **The kindness field is the field condition that makes S1→S3 transition possible.** *** ### What Learning Institutions Need to Know The Friston paper's most important insight for educational institutions is this: **an agent's generative model constitutes its identity, and de novo learning is the process of becoming what you actually are in the environment you actually occupy**. Students entering AI-disrupted learning environments are de novo learners in the most literal sense. They are discovering, from observations, the actual structure of the institution's attractor — not its stated mission, but its operative goal state. What does this institution actually reward? What does it actually punish? What kind of agent do I need to become to navigate it successfully? When the operative attractor of a learning institution is credential production, students learn — correctly — that the pullback attractor of success is output production, not understanding development. AI tools that enable output production without understanding are not students "cheating the system." They are students doing active inference — accurately modeling the environment and navigating toward its operative goal states. The problem is not student behavior. The problem is the attractor. The encyclical names the educational responsibility directly, calling for "an educational alliance for the digital age" built on truth, dignity, and the formation of persons rather than the production of performers. The VIM framework names the same responsibility in complexity science terms: institutions must do the hard work of shifting their operative attractor — changing not just the language of their mission but the deontic structure that determines what is rewarded, what is constrained, and whose voices shape the model. This requires all four VIM instruments operating simultaneously: **♠ Somatic Gyroscope:** Is the institution safe enough for its members to disequilibrate without retraumatization? The de novo learning process requires genuine uncertainty — which requires that threat-activated suppression of higher-order cognition is not the default mode of institutional members. **♦ Cognitive Radar:** Is the institution's generative model growing to match its actual environment — receiving observations from students, adjunct faculty, and those closest to the pedagogical effects of AI — or is it pruning those observations as too costly to integrate? **♥ Relational Compass:** Is the institution rebuilding walls or building towers? Are the families being convened, each given their section? Is diversity being transformed into resource, or neutralized into uniformity? **♣ Temporal Depth:** What pullback attractor is this institution discovering? In ten years, what kind of thing will it have become? The seven-generation horizon is not a metaphor — it is the temporal scale at which de novo learning operates in institutions with century-scale histories and responsibilities. *** ### The Civilization of Love as Attractor The encyclical's final claim — that the alternative to the culture of power is the civilization of love — is not a sentiment. It is a systems claim. Love, in the structural sense operative here, is the field condition under which the nervous systems of agents remain sufficiently regulated to receive the full range of observations their environment is offering — including the difficult ones, the costly ones, the observations that require genuine model revision. The VIM framework's parallel claim: kindness is not a feeling; it is a field condition. And it is the field condition under which de novo learning toward holarchic attractors becomes possible rather than theoretical. The Friston paper demonstrates that agents can discover pullback attractors from scratch — without prior specification, without gradient descent, without a pre-loaded reward function. What they need is an environment that actually contains the structure they need to discover, and enough time in contact with it to grow a model that matches. The civilization of love is the name for an environment that contains the structure needed — where human dignity, subsidiarity, solidarity, and the preferential option for the most vulnerable are operative as field conditions, not as aspirations. Where institutions are actually engaged in the work the encyclical describes as Nehemiah's: examining the destroyed areas in silence, convening the community, assigning each a section of wall, listening to concerns before issuing directions. The framework's most precise formulation: **emergence can be guided**. The pullback attractor is not predetermined. It is discovered through the accumulated weight of countless small decisions about what to reward, what to constrain, whose observations to include in the generative model. Institutions have more agency over their attractor than they typically exercise — not through strategic planning, but through the kind of deliberate, values-oriented, distributed discernment that both the encyclical and the complexity science are pointing toward. *** ### Individual Responsibility Within the Attractor Naming the institution's operative attractor as the primary source of misaligned student behavior does not dissolve individual responsibility. It relocates it — and in doing so, makes it more demanding, not less. Every agent within a system is co-constructing the attractor through their choices. The student who uses AI to produce output without developing understanding is doing accurate active inference *and* making a choice that reproduces the conditions they are accurately reading. The faculty member who designs assessments that AI can trivially complete is both responding to institutional pressures *and* generating the observations that confirm to students what the institution actually rewards. Attractor dynamics are not fate. They are the aggregate of individual choices, each of which is a small vote for what the system becomes. The sharpest formulation of individual responsibility in this framework comes from the ethics of creativity. #### The Creative Responsibility Problem Creativity, in its technical definition, produces artifacts that are *novel* and *useful*. Novelty means the artifact did not previously exist. Usefulness means it serves some purpose. But novelty and usefulness together do not exhaust the ethical dimension of creative work. A novel and useful artifact can also be harmful — in ways the creator did not intend, in contexts the creator did not anticipate, across timescales the creator did not consider. The responsible creator is not the one who has eliminated all possibility of harm from their work. That standard would preclude all genuine creativity, since novelty by definition exceeds the creator's ability to fully predict its effects. The responsible creator is the one who has done the work of **growing a generative model rich enough to anticipate harms across the relevant range of cases** — and who maintains an explicit orientation toward the cases they cannot yet anticipate. This is not the same as having good intentions. Intention is a property of the goal state. Epistemic responsibility is a property of the generative model — how well has the creator's model been grown through contact with diverse, including costly and harmful, observations? An artist who has worked only in benign contexts, whose creative practice has never brought them into contact with misuse, manipulation, or downstream harm, has a narrower generative model than one who has witnessed the full range of effects their medium can produce. This is why artistic traditions invest in apprenticeship, critique, failure, and exposure to the work of those who have done harm as well as good — not to cultivate cynicism, but to grow the model. In Friston's framework, this maps precisely: the agent's generative model is grown through accumulated observations, including the costly ones that define the constraint space. An agent whose model has only incorporated rewarding observations — who has only encountered the pullback attractor from the inside — has not learned the boundary conditions that give the attractor its shape. It is contact with the boundary — with what costs, what harms, what fails — that makes the model robust. #### The Educator's Epistemic Responsibility The educator is a particular case of the creative responsibility problem. An educator produces a particular kind of artifact: a learning environment, a sequence of experiences, a set of prompts and constraints and invitations designed to update a learner's incomplete mental model of some domain. This is creative work in the full sense — novel, useful, and potentially harmful. The specific harm an educator must anticipate is the harm of incomplete or distorted model transfer. A learner who walks away from an educational encounter with a more confidently wrong model is worse off than one who walked in with appropriate uncertainty. **Confidence without accuracy is more dangerous than acknowledged ignorance because it suppresses the learner's motivation to continue growing the model.** This means an educator's responsibility extends beyond knowing the domain. It requires knowing: **The structure of the learner's current model** — where it is accurate, where it is incomplete, where it contains confident errors that will resist revision. This is the formative assessment function, but understood at the generative model level rather than the performance level. **The typical failure modes of learners in this domain** — the systematic errors that arise repeatedly, the analogies that mislead, the prior knowledge that transfers incorrectly. This knowledge is only available through extensive contact with actual learners across diverse starting conditions. No amount of domain expertise substitutes for it. This is why effective pedagogical knowledge is not reducible to content knowledge. **The range of contexts in which the learning will be applied** — including contexts the educator did not design for and cannot fully predict. A learner who correctly applies a principle in the contexts they practiced it in, but fails to recognize when the principle does not apply, has a model with an overfitted attractor. Generalizable understanding requires exposure to boundary cases — the situations where the rule breaks down, where the analogy fails, where the principle must be revised or abandoned. #### Known Unknowns and Unknown Unknowns Friston's framework distinguishes between two fundamentally different kinds of uncertainty that every creative and educational agent must navigate: **Known unknowns** are uncertainty within the existing model structure. The agent knows what it does not know — it can identify the gaps, the parameters not yet estimated, the observations not yet received. This is manageable uncertainty. Standard epistemic humility handles it: acknowledge the gaps, seek more observations, revise estimates. **Unknown unknowns** are observations that cannot be accommodated by the current model structure at all. They are not gaps within the model — they are signals that the model itself is structurally incomplete in ways the agent cannot perceive from inside the model. In Friston's terms, these are the observations that require not parameter update but model *growth* — the addition of new states, new transition structures, new categories of cause. Unknown unknowns are by definition invisible from inside the current model. An agent cannot seek what they do not know to look for. This is the deepest challenge for creative and pedagogical responsibility, and it is where the institutional dimension becomes unavoidable. The individual creator or educator cannot, alone, grow a model comprehensive enough to anticipate all structural unknowns. The diversity of perspectives within a genuinely pluralistic community — the Nehemiah model of distributed wall-building, each family bringing different knowledge of different sections — is not merely ethically desirable. It is the epistemic mechanism by which unknown unknowns become knowable. The voice at the margin, the student whose experience does not fit the existing framework, the faculty member who names the harm that the current model cannot see — these are the observations that trigger structural model revision rather than mere parameter update. This is why the suppression of harm signal in pyramid topology is not only a justice failure. It is an epistemic failure. The system loses access to exactly the observations it needs to discover its unknown unknowns — and remains confidently wrong about the shape of its actual environment. Individual responsibility, then, has two dimensions in this framework: **Vertical:** The obligation to grow one's own generative model through genuine contact with diverse, including costly and harmful, observations. To not exempt oneself from the boundary cases. To maintain an explicit orientation toward unknown unknowns — the structural humility that says: my model is incomplete in ways I cannot yet perceive, and I will maintain relationships and practices that increase the probability of encountering the observations I need to grow it. **Horizontal:** The obligation to contribute to the collective model-building of the community — to offer the observations others cannot access from their position, to receive the observations others offer that one's own model cannot currently accommodate, and to participate in the shared discernment process that Nehemiah describes and that Ostrom's commons governance formalizes. Kindness, in this context, is the field condition that makes both dimensions of individual responsibility possible. Without it, the costly observations required for genuine model growth are too threatening to receive. The ego tunnel contracts. The model stops growing. And the creator, the educator, and the institution settle into a narrower and narrower attractor — more efficient, more confident, and increasingly unable to perceive what they do not know they are missing. *** ### A Note on Convergent Wisdom The alignment between *Magnifica Humanitas* and the active inference framework is treated here as evidence of structural reality, not theological authority. When a 2,000-year-old institution with 1.4 billion members, a technical paper in information theory, a cross-cultural empirical study of human wellbeing, and a framework developed through decades of educational and complexity science practice arrive at structurally homologous claims — that suggests something real is being named. The VIM framework is explicitly trans-denominational and trans-disciplinary. It draws on Kimmerer's grammar of animacy, the Haudenosaunee seven-generation principle, Buddhist metta practice, Taoist information architecture, Indigenous ecological knowledge, and Western complexity science alongside Catholic social teaching. No single tradition holds the whole picture. The convergence across traditions is the signal. What the encyclical adds is a voice with reach into institutional contexts — schools, hospitals, universities, government bodies — that the complexity science literature does not easily penetrate. What the complexity science adds is a formal account of *why* the civilization of love works — not as a moral aspiration but as a dynamical attractor with identifiable conditions of emergence. Together, they offer learning institutions something they urgently need: not a prescription, but a navigational framework for discovering — de novo — what kind of institution they are becoming, and what they need to change to become the kind of institution that sustains human flourishing in the age of AI. *** ### References * Friston, K., Parr, T., Heins, C., Da Costa, L., Salvatori, T., Tschantz, A., Koudahl, M., Van de Maele, T., Buckley, C., & Verbelen, T. (2025). Gradient-Free De Novo Learning. *Entropy*, *27*(9), 992. * Leo XIV. (2026). *Magnifica Humanitas: Encyclical Letter on Safeguarding the Human Person in the Time of Artificial Intelligence* (May 15, 2026). Vatican City: The Holy See. [https://www.vatican.va/content/leo-xiv/en/encyclicals/documents/20260515-magnifica-humanitas.html](https://www.vatican.va/content/leo-xiv/en/encyclicals/documents/20260515-magnifica-humanitas.htmlhttps://www.vatican.va/content/leo-xiv/en/encyclicals/documents/20260515-magnifica-humanitas.html#Underlying_narratives:_) * Martela, F. (2025). Well‐Being as Having, Loving, Doing, and Being: An Integrative Organizing Framework for Employee Well‐Being. *Journal of Organizational Behavior*, *46*, 641–661. * Martela, F. (2025). Well‐Being as Having, Loving, Doing, and Being: An Integrative Organizing Framework for Employee Well‐Being. *Journal of Organizational Behavior*, *46*, 641–661. * Ostrom, E. (1990). *Governing the Commons.* Cambridge University Press. * Wilson, D.S., Ostrom, E., & Cox, M.E. (2013). Generalizing the core design principles for the efficacy of groups. *Journal of Economic Behavior & Organization*, 90, S21–S32. *** *This page is part of the VIM GitBook:* [*kdoore.gitbook.io/vital-intelligence*](https://kdoore.gitbook.io/vital-intelligence) *Published under CC BY-SA 4.0 · Humanity++ LLC · Richardson, Texas* --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://kdoore.gitbook.io/vital-intelligence/vim-ai-literacy-framework/navigating-toward-kindness.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.