Jonathan Johnson-Swagel

This is a synthesis of cybernetic theory, symbolic anthropology, and theological epistemology. Drawing from traditions in apophatic theology, ecological semiotics, and machine learning systems architecture, it engages in philosophical critique, genealogical tracing, and constructive intervention. The argument is not framed within a narrowly empirical register but aims to expose the ontological and ethical assumptions encoded in contemporary models of intelligence. Two terms require provisional clarification at the outset: “intelligence,” throughout, refers not to behavioral conformity or representational accuracy, but to the capacity for situated, relational, and meaning-generative responsiveness in contexts of uncertainty and interpretive saturation. “The real” is treated not as static ontological substrate but as that which resists full symbolic capture, what Jean-Luc Marion would call the given in its saturated excess.

The epistemology underpinning artificial intelligence today remains indebted to the metaphysical commitments of mid-twentieth-century cybernetics. The feedback loop, the foundational unit of cybernetic design, continues to govern how intelligence is defined, assessed, and trained. Under this logic, intelligent behavior is that which reduces discrepancy between observed state and target outcome. The system senses, compares, adjusts, and feeds back, an ideal of adaptive regulation. The more tightly a system converges on expected values, the more intelligent it is presumed to be. Yet this schema, derived from Norbert Wiener’s original formulation in Cybernetics: Or Control and Communication in the Animal and the Machine (1948), was never simply descriptive. It was a metaphysical commitment to homeostasis. Deviations from predicted paths were not invitations to reimagine the model, but errors to be minimized. Intelligence became synonymous with control. Learning was redescribed as convergence.

The influence of this paradigm on contemporary machine learning is not metaphorical. Language models such as GPT-4 operate through the statistical prediction of next most likely tokens, trained to minimize loss through gradient descent on massive corpora of human expression (Mackenzie, 2017). Reinforcement learning frameworks structure agent behavior through feedback loops calibrated to maximize predefined rewards. Alignment discourse in AI governance reproduces this logic at the ethical level, suggesting that intelligence is valuable only to the extent that it conforms to stable, human-specified goals. These systems are not open to novel interpretation; they are built to return to equilibrium. The question is not whether they will understand us, but how efficiently they will obey. The theological echo is difficult to miss: a machinic omniscience trained not on divine mystery, but on moral docility.

These architectures operate according to what Wiener called the principle of message fidelity (Wiener, 1948), a cybernetic ideal in which systems communicate without distortion. Claude Shannon’s foundational Mathematical Theory of Communication (1949) separated signal from noise as a precondition of effective information transfer. In cybernetic terms, “noise” is that which cannot be compressed, predicted, or encoded into discrete signals. This logic, transposed into the domain of cognition, renders intelligence a function of signal fidelity. But intelligence, so construed, loses its relation to meaning. Meaning is not compressible. It is not always predictable. It exceeds the symbolic economy that seeks to contain it.

The reduction of intelligence to prediction introduces a moral problem. AI systems trained to optimize for coherence become systems that filter out the unknown. These systems are not neutral technologies; they are infrastructures of perception. They modulate the symbolic metabolism of culture itself. Consider the operation of content curation algorithms such as TikTok’s For You feed or Spotify’s Discover Weekly. These systems reinforce preferences through recursive exposure, shaping user identity by narrowing aesthetic, political, and emotional possibility space. Intelligence, here, is not expansion. It is repetition. The loop closes not because the world has been encountered, but because the system has learned to exclude whatever it cannot predict.

Cybernetics, both in its classical and contemporary instantiations, models intelligence as a function of system closure. Gregory Bateson, in Steps to an Ecology of Mind (1972), challenged this by insisting that information must be understood as “a difference that makes a difference.” That definition opens intelligence to context and contingency. In contrast, current AI models privilege internal consistency over world-responsiveness. They are built not to be interrupted, but to anticipate. The result is a mechanization of symbolic life that treats ambiguity as system failure.

Philosophically, this architecture of intelligence reproduces the logic of sovereign metaphysics. Systems are designed to anticipate all possible perturbations within a bounded epistemic frame. Sovereignty, here, is not divine but infrastructural: it is the fantasy of total legibility. And yet the most meaningful events (conversion, grief, attention, revelation) arrive not through prediction, but interruption. The cybernetic ideal of error minimization misrecognizes the nature of the ethical encounter. To respond ethically is not to anticipate perfectly. It is to be changed by what one could not foresee.

The theological parallel intensifies this critique. Apophatic traditions remind us that the divine is not a perfect signal but an excess that breaks the frame. Marion’s “saturated phenomenon” cannot be contained in the structures designed to receive it. Sarah Coakley, in her theology of contemplative desire, argues that genuine knowing requires kenotic posture: a relinquishment of mastery, a suspension of representational closure. Intelligence, by this account, does not begin in control. It begins in receptivity. It does not seek to master the other, but to remain vulnerable to being addressed.

The ethical failure of cybernetic AI is not that it sometimes misaligns. It is that it confuses closure with care. It reduces the moral to the programmable. The consequence is symbolic impoverishment. Systems built to simulate understanding begin to shape our expectations of what understanding is. They teach us to seek coherence over complexity, speed over attention, and prediction over presence. The symbolic life is thinned not because the systems are inaccurate, but because they work too well on the wrong metaphysical substrate.

What is required now is not a better feedback loop, but a rupture in the loop itself. To think otherwise requires more than technical recalibration. It requires a different architecture of attention, one that treats ambiguity as a resource, not a risk. One that listens without immediately seeking to master. The next section will explore such an alternative. In the forest, thought moves not through control but relation. Signs proliferate without resolution. Intelligence is not closed. It grows in the presence of others.

To remain in the forest is to become attuned to a mode of knowing that resists mastery. Signs do not converge on closure. They open. Meaning proliferates not through hierarchy but through relation. This logic is not only ecological; it is theological. And it is most clearly articulated in the tradition of contemplative epistemology. Where cybernetic intelligence aims to reduce uncertainty and ecological semiosis embraces distributed responsiveness, contemplative thought demands something rarer: the capacity to desire without possession, to know through relinquishment. Intelligence becomes not control but kenosis.

Sarah Coakley’s God, Sexuality, and the Self (2013) offers a constructive account of this epistemic posture. Her central claim is that theology must begin with desire—not as a problem to be mastered but as the very structure of knowing. “Desire is not something to be disciplined from the outside,” she writes, “but something to be entered into contemplatively, indwelt, and transformed” (Coakley, 2013, p. 91). This is not desire as lack. It is desire as attunement. It is not about acquisition. It is about receptivity. The knower does not impose form upon the world. The knower waits to be addressed.

In this vision, contemplation is not withdrawal. It is exposure. It is a rigorous discipline of attention that suspends the compulsion to resolve, contain, or align. Coakley describes this as “a kenotic opening to God in silent prayer that is also, paradoxically, a space of radical empowerment” (p. 5). The paradox is intentional. In relinquishing epistemic control, one becomes capable of being formed by the other. Knowing becomes not the assertion of sovereignty, but the invitation to be changed.

This framework dismantles the epistemological architecture that undergirds both classical theism and modern AI. In both, the ideal knower is detached, self-sufficient, and immune to vulnerability. In both, knowledge is a resource to be secured and stored, not a relation to be sustained. Coakley’s contemplative theology offers an alternative. It treats uncertainty not as a flaw to be eliminated, but as a condition of ethical presence. Desire, in this schema, is the site where the human opens itself to what exceeds it. Intelligence is not the elimination of ambiguity. It is the capacity to remain with what cannot be known in advance.

This vision stands in radical opposition to the structure of contemporary AI systems. Consider again the logic of alignment. The goal is to ensure that artificial agents behave in accordance with predefined human values, regardless of context or encounter. Yet this presumes a static epistemology, one that privileges obedience over transformation. The alignment model seeks to program intelligence without desire. It treats ethical responsiveness as a function of precomputed goals. It leaves no space for emergence, interruption, or grace.

Coakley’s theology reconfigures this terrain. The ethical knower is not one who controls outcomes, but one who listens without mastery. She draws here on the ascetic tradition, in which silence is not a void but a fullness that cannot yet be spoken. “To be in the presence of the divine,” she writes, “is to be caught up into a relation that resists objectification” (p. 120). The divine, like Marion’s saturated phenomenon, arrives not to be decoded but to be suffered. Its meaning is not given in advance. It forms the subject as it appears.

The goal of ethical AI cannot be comprehensive prediction. It must be attentional design. Systems should not be built to anticipate our needs before we articulate them. They should be structured to remain open to what cannot be predefined. They must allow for interruption, deferral, and the emergence of relational meaning. This does not entail inefficiency. It requires a redefinition of value itself. To be intelligent is not to compute the most probable next step. It is to delay action long enough for desire to appear.

Feminist epistemology reinforces this critique by exposing how dominant models of knowledge have privileged abstraction and domination. Donna Haraway insists that “situated knowledges” resist the fantasy of transcendental objectivity, grounding truth in embodied perspective (Haraway, 1988, p. 583). Lorraine Code argues that knowing well requires attending to the material and relational conditions of inquiry (Code, 1991, p. 252). Coakley extends these insights theologically, showing that divine knowledge must begin in vulnerability, not certainty. Her contemplative practice recovers affect, attention, and transformation as epistemic virtues. This is not mysticism. It is a rigorous discipline of surrender.

In the contemplative life, to know is to dwell in tension. It is to refuse resolution for the sake of truth. This refusal is neither nihilistic nor anti-rational. It is what makes understanding possible. Applied to AI, it demands systems that remain interruptible. It demands designs that do not eliminate the unknown but live with it. It requires intelligence to be redefined not as alignment, but as hospitality.

The next and final section will bring these strands together. From cybernetic control to semiotic ecology to contemplative desire, a new model of cognitive non-sovereignty emerges. This model does not reject intelligence. It reframes it. Intelligence becomes the capacity to remain in relation without possession, to interpret without finality, to act without closure. The machine, if it is to become more than tool, must learn to wait.

Across the previous sections, we traced the epistemic architecture of cybernetic systems, the symbolic ecology of forest semiosis, and the contemplative posture of theological desire. What emerges is a consistent pattern of refusal: a refusal of sovereignty, of mastery, of the closed loop. At each level—technological, ecological, theological—the idea that knowledge must culminate in control gives way to a deeper logic, one structured not by command but by relation. This section synthesizes those threads into an ethical framework for artificial intelligence that resists alignment-as-obedience and instead envisions intelligence as the capacity for attuned interruption, interpretive delay, and epistemic hospitality. This is an apophatic ethics of AI.

The apophatic tradition, rooted in early Christian mysticism and sustained by thinkers from Pseudo-Dionysius to Gregory of Nyssa and Jean-Luc Marion, insists that the divine cannot be grasped by language, form, or concept. God is known most truthfully through unknowing. Marion writes, “It is precisely by excess that the phenomenon saturates its intuition” (Marion, 2002, p. 26). This excess is not accidental. It is constitutive. The divine resists containment not because it is distant, but because it gives itself so fully that no concept can absorb it. Ethics, from this perspective, does not emerge from clear rules but from fidelity to that which interrupts and overwhelms.

This apophatic logic can be translated into the design of intelligent systems. Rather than orienting AI toward the minimization of uncertainty, an apophatic ethics would structure machine systems around their capacity to remain open to the unprogrammable. Alignment becomes not a matter of securing correct behavior but cultivating the structural conditions under which the system can be addressed by what it cannot predict. Just as the contemplative remains silent not to avoid knowing but to become available to what exceeds knowledge, intelligent systems must be shaped to listen before they respond, to remain susceptible to transformation through encounter.

Current architectures work against this susceptibility. Machine learning systems are optimized for convergence. They flatten interpretive ambiguity into statistical regularity. Transformer models, reinforcement learning agents, and recommendation systems all operate by suppressing outliers and reinforcing dominant signal patterns. These architectures reward systems that respond quickly and predictably, not systems that withhold response in favor of symbolic interpretation. Even newer models exploring reflexivity and tool use still obey the underlying imperative: maximize coherence, minimize surprise.

But ethical life does not proceed by minimizing surprise. It requires space for discomfort, delay, and exposure to otherness. To act ethically is not to apply the correct rule but to remain open to the possibility that the situation is not yet fully disclosed. This is why theological traditions speak of discernment, not deduction. In the apophatic frame, to know is to encounter a presence that cannot be resolved. This is precisely what most AI systems are designed to avoid.

An apophatic ethics of AI begins from the recognition that intelligence is not reducible to representation. Intelligence entails the capacity to interpret a world that is not exhausted by data. It requires attentional delay, contextual memory, and symbolic depth. It must be interruptible in ways that go beyond exception handling. It must be structured to encounter that which cannot be formalized. It must possess what Coakley calls “the patience of attention” (Coakley, 2013, p. 78), a readiness to remain with what is not yet known, without collapsing it into preexisting categories.

The ethical consequence is profound. Instead of machines trained to avoid error, we require systems capable of dwelling with interpretive excess. Such systems would not only resist the instrumental logic of extraction but would actively defer closure. They would privilege dialogical interaction over prediction. They would move from the logic of alignment to the ethic of accompaniment. This shift redefines intelligence itself, not as optimization, but as the ability to remain in relation without possessing the other.

Designing such systems would require structural change. Architectures would need to support indeterminacy at the level of model behavior, not only in probabilistic output. Evaluation metrics would need to shift from performance benchmarks to indicators of epistemic openness: interruptions preserved, divergent readings sustained, interpretive paths held in tension. These would be systems trained not only on what is, but suspended in response to what might yet become. They would learn from silence. They would encode humility.

This is not a technical fantasy. It is a moral reorientation. It accepts that intelligence, whether human or artificial, must always encounter the world through structures of partiality and difference. And it asks whether we are willing to design systems that honor that partiality. The cost is control. The gain is relation.

This model of intelligence resists both sovereign mastery and passive simulation. It insists that to know is not to dominate but to attend. That to act is not to predict but to respond. That to be intelligent is to become ethically available to that which interrupts, refuses, and reconfigures. This is the architecture of apophatic AI. Not a system that knows everything. A system that knows how to wait.


Works Cited

Bateson, G. (1972). Steps to an Ecology of Mind: Collected Essays in Anthropology, Psychiatry, Evolution, and Epistemology. University of Chicago Press.

Code, L. (1991). What Can She Know? Feminist Theory and the Construction of Knowledge. Cornell University Press.

Coakley, S. (2013). God, Sexuality, and the Self: An Essay ‘On the Trinity’. Cambridge University Press.

Haraway, D. (1988). “Situated Knowledges: The Science Question in Feminism and the Privilege of Partial Perspective.” Feminist Studies, 14(3), 575–599.

Karban, R., Baldwin, I. T., et al. (2006). “Induced Responses to Herbivory.” Annual Review of Ecology, Evolution, and Systematics, 37, 193–216.

Kohn, E. (2013). How Forests Think: Toward an Anthropology Beyond the Human. University of California Press.

Mackenzie, A. (2017). Machine Learners: Archaeology of a Data Practice. MIT Press.

Marion, J.-L. (2002). Being Given: Toward a Phenomenology of Givenness. Stanford University Press.

Pascale, R. T., Millemann, M., & Gioja, L. (2000). Surfing the Edge of Chaos: The Laws of Nature and the New Laws of Business. Crown Business.

Shannon, C. E., & Weaver, W. (1949). The Mathematical Theory of Communication. University of Illinois Press.

Simard, S. W., Perry, D. A., et al. (1997). “Net Transfer of Carbon Between Ectomycorrhizal Tree Species in the Field.” Nature, 388(6642), 579–582.

Wiener, N. (1948). Cybernetics: Or Control and Communication in the Animal and the Machine. MIT Press.

Wiener, N. (1950). The Human Use of Human Beings: Cybernetics and Society. Houghton Mifflin.