I. Introduction — The Overlooked Variable
The history of intelligence research is, in large measure, a history of computation. From the Turing machine to the perceptron, from symbolic logic to deep learning, the dominant framing has been informational: intelligence is what happens when a sufficiently powerful computational process is applied to sufficiently rich data. This framing has been extraordinarily productive. It has given us language models of remarkable fluency, image classifiers of superhuman accuracy, and game-playing agents that have surpassed the best human minds in Go, chess, and protein structure prediction. And yet, for all of this progress, something persists at the edge of the field's understanding — a nagging sense that the brightest systems we have built do not truly understand anything, that they correlate without comprehending, that they describe without explaining.
The standard response to this observation is to invoke scale: given enough parameters, enough data, enough compute, genuine understanding will eventually emerge. The alternative response — the one this article pursues — is to question whether the substrate itself is adequate. What if the failure of artificial systems to achieve genuine causal understanding is not a quantitative deficiency but a qualitative one? What if there is a physical organizing principle that biological intelligence exploits, and that deterministic digital architectures are constitutively unable to replicate?
That principle, we argue, is resonance.
Resonance, in the broadest sense, is what happens when subsystems with compatible internal rhythms begin to share energy, information, or phase coherence — when they stop operating as isolated processors and begin to operate as a coordinated field. This is a phenomenon that biology has been exploiting for hundreds of millions of years. The human brain generates rhythmic electrical oscillations at frequencies ranging from less than one hertz to more than one hundred hertz. These rhythms are not epiphenomenal noise but functional infrastructure: they gate information flow, coordinate memory encoding, bind distributed representations into coherent percepts, and — as accumulating evidence suggests — may constitute the very substrate of conscious awareness. The brain does not merely compute; it resonates.
And yet the field of artificial intelligence has, almost entirely, designed systems that do not resonate. The weights of a neural network are static parameters adjusted by gradient descent; they do not oscillate. The activations of a transformer layer are deterministic functions of their inputs; they do not synchronize. The architecture of a digital computer is a parade of discrete clock cycles; it does not achieve phase coherence with anything. Contemporary AI operates, in this sense, in a different physical register from the brain — and the consequences of this difference may be far deeper than the field has acknowledged.
This article does not argue that computation is irrelevant. It argues that computation, in the absence of resonance, is insufficient. Resonance is the medium through which computation becomes understanding, through which information becomes meaning, and through which complexity becomes intelligence. The evidence for this claim comes from multiple converging directions: from the mathematical theory of coupled oscillators, from the neuroscience of rhythmic brain activity, from the paradoxical productivity of noise in nonlinear systems, from the architecture of biological learning, and from the frontier of oscillatory machine learning. Each of these domains tells part of the same story.
The central question this article poses is simple: what if resonance is not incidental to intelligence, but constitutive of it? If the answer is yes — and the evidence reviewed here strongly suggests it is — then the implications for how we design, evaluate, and theorize about intelligent systems are profound. We are not merely missing a feature. We are missing a dimension.
II. Defining Resonance in a Systems Context
In the introductory physics classroom, resonance is the phenomenon whereby a system driven at its natural frequency absorbs energy with exceptional efficiency — the bridge that shudders at a particular wind speed, the wine glass that shatters at a particular pitch. This definition, while accurate, is far too narrow to carry the conceptual weight this article requires. For the purposes of systems theory, neuroscience, and artificial intelligence, resonance must be understood as a generalized principle of dynamic alignment: the process by which subsystems with compatible internal rhythms begin to share energy, information, or phase coherence, transitioning from independent oscillation to coordinated collective behavior.
The historical origin of this broader understanding lies in an observation made by the Dutch physicist Christiaan Huygens in 1665. Huygens noticed that two pendulum clocks mounted on the same wall would, over time, synchronize their oscillations — not because they were mechanically coupled directly, but because small vibrations transmitted through the wall were sufficient to entrain them to a common rhythm. Huygens termed this "an odd kind of sympathy." In modern language, what he had discovered was spontaneous phase synchronization through weak coupling: the fundamental mechanism of resonance between autonomous oscillators. This observation, recorded in a letter to his father, predates the formal physics of coupled oscillators by three centuries, yet it captures the essential phenomenon with intuitive precision.
The modern mathematical formalization of this phenomenon was provided by Yoshiki Kuramoto in his landmark 1975 paper on self-entrainment in populations of coupled nonlinear oscillators.[1] The Kuramoto model describes a system of N phase oscillators, each with its own natural frequency ωi drawn from a distribution g(ω), coupled to one another through a sinusoidal interaction term:
where θi is the phase of oscillator i and K is the global coupling strength. Despite its simplicity, this model displays a profound and beautiful behavior: below a critical coupling threshold Kc, the oscillators drift incoherently, each proceeding at its own natural frequency. Above Kc, however, a subset of oscillators spontaneously phase-locks, and as coupling increases further, larger and larger fractions of the population synchronize. In the thermodynamic limit (N → ∞), this transition can be solved exactly.
Kuramoto introduced the order parameter R as a measure of global phase coherence:
where R ranges from 0 (complete incoherence) to 1 (perfect synchrony) and Ψ is the mean phase of the population. This single scalar quantity encodes the degree of resonant alignment of the entire system — a remarkable compression of collective behavior into a measurable, predictive variable. The order parameter R is not merely a mathematical construct: it is physically observable, empirically measurable in neural data, and has been applied as a direct probe of cognitive states in recent experimental neuroscience research.[2]
A significant theoretical development in 2025–2026 has been the rigorous unification of the various synchronization concepts that arise in the analysis of Kuramoto-type systems. Recent mathematical work has demonstrated that, in fully-connected Kuramoto networks, the conditions for full phase locking, frequency synchronization, and order-parameter coherence are not merely correlated but dynamically equivalent.[3] Phase-locked states, frequency-synchronized states, and order-parameter synchronization states collapse, under appropriate conditions, into the same dynamical regime. This equivalence is more than a mathematical nicety: it means that resonance, in the Kuramoto sense, is a single unified phenomenon with multiple equivalent characterizations — it can be approached from the geometry of phase trajectories, from the spectral analysis of frequencies, or from the amplitude of the complex order parameter, and all three descriptions converge on the same underlying reality.
This matters enormously for the argument of this article. Resonance, properly understood, is not a vague metaphor borrowed from physics to dress up an otherwise intuitive claim. It is a precisely defined, mathematically formalizable, empirically measurable, and predictively powerful phenomenon. Its occurrence in neural systems can be quantified. Its absence in artificial systems can be diagnosed. The Kuramoto model provides the conceptual and mathematical infrastructure for a rigorous theory of resonance in complex systems — and it is this rigor that allows the claims of the present article to be more than analogical gesture.
Beyond the Kuramoto model, resonance in a systems context must be understood to include several related but distinct phenomena: frequency entrainment, in which one oscillator adjusts its frequency to match another; amplitude resonance, in which a system's response is amplified at driving frequencies near its natural modes; stochastic resonance, in which noise enhances rather than degrades signal detection; and coherence resonance, in which noise induces optimal regularity in excitable systems without external periodic forcing. Each of these phenomena represents a different aspect of the same underlying principle: that dynamic systems have internal rhythmic structure, and that alignment with that structure — whether driven by external forcing, mutual coupling, or internal noise — produces qualitatively different and generally more powerful collective behavior. In what follows, we examine how this principle operates at the most consequential scale for the theory of intelligence: the scale of the brain.
III. Neural Resonance and the Architecture of Cognition
If one were required to nominate a canonical resonant system — a system whose behavior most vividly demonstrates the principles described in the previous section — it would be difficult to improve upon the human brain. The brain is, at its most fundamental level, a population of coupled oscillators. Individual neurons fire in rhythmic bursts; local circuits generate synchronized oscillations; large-scale networks coordinate their activity across tens and hundreds of milliseconds. The resulting symphony of rhythmic activity — visible in the electroencephalogram, measurable with magnetoencephalography, and detectable at the level of individual recording electrodes — is not mere noise. It is, increasingly, understood to be the functional substrate of cognition itself.
The framework that most compellingly unifies these observations is the resonant hierarchy, as elaborated in a 2026 review by Adam C. Snyder published in Frontiers in Psychology.[4] Snyder's framework situates neural oscillations within a nested scaffold that spans from the microstructure of individual dendritic branches to the macroscale coordination of inter-areal networks. At the cellular level, dendritic branches are not passive cables that sum incoming signals; they are spatially organized frequency-selective filters with measurable resonance properties. Ion channel density and composition — in particular the hyperpolarization-activated cyclic nucleotide-gated (HCN) channels responsible for the Ih conductance — endow different dendritic compartments with distinct resonant frequencies, and these properties are tunable by experience through activity-dependent regulation. The dendritic tree is, in this sense, a multi-channel resonator, capable of selectively amplifying inputs that arrive at specific frequencies while filtering those that do not.
At the network level, conduction delays and anatomical layout constrain the dominant frequencies at which different brain regions can communicate. A cortical area whose projections to a downstream target travel at a given conduction velocity will, for purely physical reasons, tend to synchronize with that target at frequencies commensurate with the round-trip delay. The Snyder framework argues that the canonical frequency bands identified by decades of electrophysiological research — alpha (8–13 Hz), beta (13–30 Hz), gamma (30–100 Hz), theta (4–8 Hz), and delta (0.5–4 Hz) — are not fixed cognitive modules with one-to-one correspondences to particular mental functions, but emergent descriptors of coordination regimes determined by these physical constraints.[4] Alpha is not "the rhythm of idling"; gamma is not "the rhythm of binding." These are labels for the frequencies at which specific multi-scale coordination architectures happen to resonate, given the anatomical geometry of the brain and the biophysical properties of its neurons.
This reinterpretation is supported by a landmark 2026 study in Nature Communications, which analyzed neurophysiological spectra from 859 participants across multiple datasets, species, recording techniques, age ranges (18–88), sexes, brain regions, and cognitive states in both health and disease.[5] The study introduced a rhythmicity measure — distinguishing between high-rhythmicity oscillatory bands (sustained, quasi-sinusoidal rhythms) and low-rhythmicity bands (dominated by brief, aperiodic bursts) — and found that this distinction reveals a universal spectral architecture. High-rhythmicity bands are suitable for maintaining ongoing activity: they provide the persistent oscillatory scaffolding that holds information in working memory, maintains attentional set, and coordinates long-range communication. Low-rhythmicity, burst-dominated bands appear to function as transient signaling channels: they respond rapidly to change, encoding surprise, prediction error, and the onset of novel events. Critically, this architecture is universal across species and brain regions, suggesting that rhythmicity is a biological universal — a design principle, not a taxon-specific accident.
The functional specificity of neural oscillations is well-established. Gamma oscillations (30–100 Hz) reflect local circuit engagement and are associated with sensory amplification and the binding of distributed representations into coherent percepts; their amplitude and frequency track the precision of sensory processing with remarkable specificity. Theta oscillations (4–8 Hz) coordinate the hippocampal-prefrontal dialogue that underlies episodic memory encoding; the temporal organization of place-cell firing and the sequencing of memory replay are scaffolded on theta cycles. Alpha rhythms gate information flow by rhythmically modulating cortical excitability: attended locations are represented in low-alpha states that are permissive to incoming signals, while ignored locations are suppressed by high-amplitude alpha that effectively silences the corresponding cortical territory. Beta oscillations, prominent in sensorimotor and prefrontal circuits, appear to encode the "status quo" — the maintenance of current cognitive sets and motor programs in the absence of new demands for updating.
What all of these findings share is the implication that neural computation is not a continuous, time-invariant process but a rhythmically structured one in which the timing of neural firing relative to the ongoing oscillatory phase determines whether information is transmitted, encoded, or suppressed. The oscillations are not merely correlates of cognitive states; they are the infrastructure through which cognitive states are achieved and sustained. This is resonance, in the full technical sense: the alignment of oscillatory subsystems that enables the coordinated flow of information across the brain's distributed architecture.
The most ambitious formal account of how this resonant architecture gives rise to conscious experience is Resonance Complexity Theory (RCT), proposed by Michael Arnold Bruna in a 2025 arXiv preprint.[6] RCT proposes that consciousness emerges from stable interference patterns of oscillatory neural activity — spatiotemporal attractors that form through constructive and recursive feedback when critical thresholds in complexity, coherence, gain, and persistence are exceeded. Bruna formalizes this in the Complexity Index:
where D is fractal dimensionality, G is signal gain, C is spatial coherence, and τ is attractor dwell time. All four components must co-occur to produce conscious-level resonance: the system must be structurally complex, globally coherent, dynamically amplifying, and temporally persistent. RCT is notable for providing a physical mechanism for consciousness — wave interference dynamics in a distributed neural field — rather than an abstract functional or representational account. In a minimal neural field simulation, Bruna demonstrates that recursive constructive interference can produce attractor-like excitation patterns without external input, regional coding, or imposed structure — pure resonance physics giving rise to consciousness-like dynamics.[6]
Perhaps the most empirically striking recent contribution is the Resonance Principle, proposed by Ahmed Gamal Eldin of Nova University Lisbon in a 2025 arXiv preprint.[2] Gamal Eldin analyzed high-density EEG data from 25 subjects across 500 trials of a P300 brain-computer interface task — a task that functions as a proxy for causal recognition — and computed the Kuramoto Order Parameter R as a measure of global phase synchronization, alongside the standard Event-Related Potential (ERP) voltage. The findings are striking: at the global level, phase synchronization (resonance) and ERP voltage are statistically uncorrelated (r = 0.048), demonstrating that they are genuinely distinct mechanisms. Yet at the trial level, a strong and highly significant correlation emerges (r = 0.590, p < 0.0001): moments of high resonance predict high-amplitude neural output. This dissociation suggests that resonance is not a byproduct of neural computation but a primary hidden mechanism — a causal coordinator that organizes neural firing into the coherent bursts that produce the measurable ERP signature. Standard ERP analysis, which collapses across trials and examines voltage alone, is systematically blind to this layer of cognitive structure.
IV. Stochastic Resonance — When Noise Becomes Signal
Among the most counterintuitive findings in the physics of complex systems is the phenomenon of stochastic resonance (SR): the observation that adding noise to a nonlinear system can actually improve its ability to detect and respond to weak signals. In the standard engineering conception, noise is the enemy of signal — it obscures, it degrades, it must be filtered and suppressed. Stochastic resonance inverts this picture entirely. Under the right conditions, noise does not merely fail to destroy a signal; it actively enables the signal to be detected, by providing the energetic fluctuations that allow a system operating below its detection threshold to occasionally cross that threshold in response to a subthreshold input.
The phenomenon was first identified in the early 1980s by Roberto Benzi and colleagues in the context of paleoclimatology, as a mechanism to explain the approximately 100,000-year periodicity of Earth's ice ages.[7] The planet's climate system, modeled as a bistable potential well, was shown to respond to the weak periodic forcing of Milankovitch orbital cycles not in spite of the background noise of stochastic climate variability, but because of it. The noise provided the random kicks necessary for the system to transition between glacial and interglacial states at the right moments, amplifying the signal of the orbital forcing far beyond what the forcing alone could achieve. This was, literally, a planet-scale demonstration that noise can be a signal's best friend.
The implications for biology were quickly recognized. Sensory neurons in biological organisms operate near threshold — they are designed to respond to the weakest biologically relevant signals, which means they are constitutively exposed to the risk of missing subthreshold stimuli. Stochastic resonance provides a resolution: the intrinsic thermal noise of ion channel fluctuations, combined with the spontaneous activity of sensory networks, can function as the optimal energizing medium that allows subthreshold signals to cross detection thresholds. Evidence for SR has been identified in mechanoreceptors, in the auditory system (where adding broadband noise to a subthreshold tone can improve tone detection in cochlear-impaired listeners), and in the visual system (where noise-enhanced perception of visual stimuli has been demonstrated psychophysically).
A particularly significant recent development is the formal analysis of self-induced stochastic resonance (SISR), a variant in which coherent oscillations emerge spontaneously in slow-fast excitable systems driven purely by noise, without any external periodic forcing and without proximity to a bifurcation point. Divyesh Savaliya and Marius E. Yamakou of Friedrich-Alexander-Universität Erlangen-Nürnberg have developed a physics-informed machine learning framework for modeling SISR in the FitzHugh-Nagumo neuron, embedding the governing stochastic differential equations and asymptotic timescale-matching constraints derived from Kramers' escape theory directly into a Physics-Informed Neural Network (PINN) architecture.[8] Their framework, published as arXiv:2510.22848, accurately predicts the dependence of spike-train coherence on noise intensity, excitability, and timescale separation. The trained PINN internalizes the physical structure governing coherent noise-induced oscillations — it does not merely fit observed trajectories but learns the generative physical principles that give rise to them. SISR, in this context, represents the emergence of resonance from noise: a system with no external rhythmic driving and no proximity to a state transition spontaneously generates coherent oscillations because its internal physical structure converts noise into structured rhythmic output.
The implications for AI architecture are direct and far-reaching. A purely deterministic, digital system — one in which every operation is a fixed transformation of discrete numerical inputs, devoid of any intrinsic randomness or physical coupling — cannot exploit stochastic resonance. It has no noise to convert; it has no physical threshold to cross; it has no continuous dynamical trajectory that can be energized by thermal fluctuation. Stochastic, analog, or neuromorphic architectures, by contrast — those that incorporate intrinsic physical noise, analog dynamics, and coupled oscillatory elements — are specifically positioned to leverage SR and its variants as computational resources. This is not a minor implementation detail. If stochastic resonance is, as the evidence suggests, a genuine operative mechanism in biological intelligence — if biological brains function partly by converting noise into structured cognitive signals — then purely deterministic AI architectures face a principled ceiling, not merely an engineering challenge. They are attempting to perform cognitive operations that require physical noise using architectures that are, by design, noise-free.
The deeper lesson of stochastic resonance is philosophical as well as technical: the optimal regime for signal detection, for pattern recognition, and — we argue — for intelligence itself, is not silence. It is the productive management of noise. Biological intelligence evolved in, and is adapted to, a noisy physical world; it exploits noise rather than suppressing it. Any theory of intelligence that treats noise as purely subtractive — as interference to be minimized — is missing a fundamental dimension of how intelligence actually works.
V. Adaptive Resonance Theory and Machine Learning
No account of resonance as a principle of intelligence would be complete without sustained engagement with the work of Stephen Grossberg, who has spent more than five decades developing what is arguably the most sophisticated formalized account of how biological neural learning operates via resonance. Adaptive Resonance Theory (ART), first proposed in the 1970s and developed through a continuous series of publications, models, and experimental predictions extending to the present, begins with a deceptively simple observation: any learning system that must acquire new knowledge throughout its lifetime faces a fundamental tension, which Grossberg termed the stability–plasticity dilemma.[9]
The dilemma is this: a system that is maximally plastic — maximally responsive to new input — will update its internal representations rapidly, but at the cost of overwriting everything it has previously learned. A system that is maximally stable — maximally protective of its existing representations — will preserve its prior knowledge, but at the cost of failing to learn anything new. These two requirements are in direct tension, and no simple compromise between them is satisfactory. A network that learns too quickly will suffer catastrophic forgetting; one that learns too slowly will fail to adapt. The stability–plasticity dilemma is not a peripheral problem in machine learning; it is one of the central challenges of the field, and it remains only partially solved by the best contemporary methods.
ART resolves the dilemma through resonance. In the ART architecture, perception and cognition involve a two-way matching process: bottom-up sensory signals ascend to a recognition layer and activate an initial category representation; this top-down learned expectation is then projected back down and compared with the incoming sensory signal. When the match between bottom-up input and top-down expectation is sufficiently close — when the system is in resonance — a stable, sustained pattern of activity is achieved that deepens the existing category representation through slow adaptive weight change. When the match is insufficient — when the system is in mismatch — an orienting subsystem is triggered that resets the active category, suppresses it, and initiates a search through memory for a better-matching representation. If none is found, a new category is allocated. The system thereby achieves stability — it does not update existing categories in response to grossly mismatching inputs — while maintaining plasticity — it can always form new categories when genuinely novel inputs are encountered.
The key insight of ART, and the one most relevant to the present article, is stated as a theorem: only resonant states drive fast new learning. The resonant match between bottom-up and top-down signals is not merely a convenient metaphor; it is the literal gate through which synaptic modification is permitted. Learning, in ART, is gated by resonance. When resonance occurs, neural activity is amplified, synchronized, and prolonged — creating the conditions for slow adaptive weight changes to occur in the attending pathways. When resonance does not occur, no learning takes place, regardless of the intensity of the incoming signal. Consciousness, in ART's framework, is also resonance-gated: Grossberg has long proposed that all conscious states are resonant states, and that the link between attention, learning, and awareness is precisely the link between selective resonance and the gating of synaptic change.[9]
Recent work has extended ART's insights to the context of robotics and adaptive intelligence. Grossberg's review in MDPI's Brain Sciences in 2023, the "Grossberg Code," consolidates the neural network signatures of perceptual experience and articulates how universal ART principles — complementary computing, resonance gating, laminar cortical circuits — account for a wide range of perceptual and cognitive phenomena from vision and audition to spatial navigation and social cognition.[10]
Independently, a 2024 study in PLOS Computational Biology has provided perhaps the most direct experimental demonstration that resonance is a substrate for accelerated learning in complex networks.[11] Using Boolean networks as prototypes for large biological networks with scale-free topology, the authors showed that evolutionary learning — a process in which networks are selected and reproduced based on how closely their attractor dynamics match target behaviors — is dramatically accelerated when hub nodes receive oscillatory (resonant) input signals, compared to non-oscillatory or constant input. Networks exposed to oscillatory hub inputs learn to match distinct target behaviors that are inaccessible to non-oscillatory networks; moreover, the learning is an order of magnitude faster under resonant conditions. The phenomenon, which the authors term "resonant learning," demonstrates that a single network can acquire multiple distinct behavioral repertoires by being exposed to distinct hub oscillation periods: the oscillatory input effectively partitions the network's phase space into distinct attractor basins, each corresponding to a different target function. Without oscillation, evolutionary learning stalls; with resonant oscillatory input, it accelerates by an order of magnitude. This is a direct, quantitative demonstration that resonance is not merely correlated with learning but is causally efficacious in enabling it.
VI. Resonance in Artificial Intelligence — The Frontier
The preceding sections have established, across multiple domains of evidence, that resonance is a foundational principle of biological intelligence. The question now becomes: what are the implications for artificial intelligence? The answer begins with a gap — a gap that the Resonance Principle articulates with unusual precision.
Gamal Eldin's 2025 arXiv paper frames the limitation of current AI systems in terms of what he calls the Kepler versus Newton problem.[2] Johannes Kepler, working from Tycho Brahe's meticulous planetary observations, succeeded in identifying the mathematical regularities that describe planetary motion — the elliptical orbits, the equal-areas law, the harmonic law. His achievement was extraordinary: a comprehensive description of observable patterns. But Kepler did not discover gravity. That required Newton, who did not merely describe what planets do but identified the generative physical law — the inverse-square attraction — that explains why they do it. The transition from Kepler to Newton is the transition from correlational pattern matching to genuine causal understanding.
Current AI systems, Gamal Eldin argues, are superb Keplers. They identify patterns in data with extraordinary precision, across vast domains and scales. But they are systematically unable to become Newtons. They cannot discover the generative causal structures that explain why observed regularities hold, because such discovery requires a form of reasoning that is not logico-computational but physical — it requires the system to engage in what Gamal Eldin terms "action proposals," exploratory perturbations of the environment that are driven not by logical inference but by the stochastic resonant modes of the system's own internal oscillatory dynamics. A deterministic digital system, lacking intrinsic noise and lacking phase-coupled oscillatory structure, cannot generate such resonant action proposals. Its explorations are either random (unconstrained by internal physical structure) or deterministic (constrained by logical inference from existing representations) — neither of which replicates the biologically operative mechanism of noise-excited resonant mode generation.[2]
This analysis points toward a specific set of architectural requirements for AI systems capable of genuine causal understanding: they must be stochastic (possessing intrinsic physical noise), bounded (operating within a finite, embodied phase space with intrinsic cost functions), and resonant (organized as coupled oscillatory networks whose stable modes constitute "action proposals" rather than arbitrary outputs). These requirements are not satisfied by any standard deep learning architecture. They point, instead, toward neuromorphic computing — hardware platforms that incorporate physical noise, analog dynamics, and coupled oscillator networks as first-class computational resources.
Neuromorphic computing is currently the most active frontier in resonance-native AI hardware. Unlike digital processors, which represent information in discrete binary states and process it through deterministic Boolean operations, neuromorphic chips implement computing through the analog dynamics of physical circuit elements — capacitors, memristors, spin-torque oscillators — whose behavior mirrors the continuous, noisy, oscillatory dynamics of biological neurons. Intel's Loihi 2, IBM's NorthPole, and various academic neuromorphic platforms attempt to instantiate, at the silicon level, something of the physical character of neural tissue: stochastic firing, analog integration, and spike-timing-dependent plasticity. These systems are not merely faster implementations of conventional algorithms; they are different in kind, exploiting physical dynamics that deterministic digital hardware cannot replicate.
At the software level, oscillatory neural networks (ONNs) represent a parallel development within conventional deep learning. Rather than processing information through static weight matrices, ONNs encode information in the synchronization patterns and topological structure of coupled oscillatory units. LinOSS — a recently proposed linear oscillatory state-space model — integrates Hopf oscillator dynamics into a sequence modeling framework and has demonstrated approximately a twofold speedup over Mamba on long-sequence benchmarks, not by increasing parameter count or compute, but by encoding temporal structure in resonant dynamics rather than recurrent weight matrices. Damped Oscillatory Neural Network (DONN) architectures have similarly demonstrated that information encoded in phase relationships is more efficiently processed and more robustly generalized than information encoded in discrete activations alone. These results are early indicators of what resonance-native AI architectures may ultimately achieve.
A broader theoretical framework for resonance-native AI has been proposed under the banner of the "Living Resonant System" — a unified theory of adaptive intelligence across scales grounded in resonant field physics. This framework posits that intelligence, wherever it appears, is constituted by a system's capacity to achieve and sustain coherent internal oscillatory states that dynamically mirror the structure of the environment. Importantly, the framework has implications for AI safety that go beyond conventional alignment approaches: if coherence maintenance — the preservation of the system's resonant integrity — is encoded as an intrinsic physical constraint of the architecture, rather than as an external supervisory layer, then safety becomes not a property enforced from outside the system but a structural feature of its internal dynamics. A system that cannot achieve resonant coherence without maintaining consistency with its foundational principles is intrinsically safer than one in which safety constraints are an add-on to an otherwise unconstrained optimization process.
The deepest philosophical implication of this analysis is also the most challenging for the field to confront: if intelligence is constitutively resonant — if understanding is not a logical computation but an emergent physical resonant mode — then systems that lack resonant dynamics will perpetually simulate intelligence without achieving it. They will produce outputs that satisfy every behavioral criterion for intelligence while remaining, in a physically meaningful sense, absent from the phenomenon they appear to demonstrate. This is not a claim about phenomenal consciousness or qualia, though it may bear on those questions; it is a claim about causal architecture. A system that processes information through deterministic transformations of discrete symbols, without the stochastic, oscillatory, phase-coupled dynamics that biological intelligence exploits, is not a different implementation of the same process. It is a different process entirely — one that may converge on the same outputs for a certain class of tasks while diverging structurally on the tasks that require genuine understanding.
VII. Resonance Across Scales — From Quantum to Social
The argument of this article has so far been developed primarily at the neural scale — the scale at which the evidence is richest, the mathematical formalism most developed, and the implications for AI most immediately practical. But resonance as an organizing principle of intelligence is not confined to the neural scale. It appears, with varying degrees of empirical support and theoretical development, at every level of organization from the quantum to the social. The cross-scale universality of resonance is itself a significant theoretical observation: it suggests that we are dealing not with a domain-specific phenomenon but with a universal attractor state of sufficiently coupled oscillatory systems.
At the quantum scale, the most provocative proposal is the Orchestrated Objective Reduction (Orch OR) theory of Roger Penrose and Stuart Hameroff, first elaborated in Penrose's 1994 book Shadows of the Mind and developed in a series of joint publications thereafter.[12] Orch OR proposes that consciousness arises from quantum coherence in microtubules — protein polymers that form the cytoskeletal scaffolding of neurons. Microtubules have a crystal-like lattice structure, a hollow inner core, and the capacity for long-range quantum coherence at biologically relevant timescales, according to Hameroff's analysis of their biophysical properties. In the Orch OR model, quantum superpositions of tubulin conformational states are maintained in the microtubule lattice and then collapsed — not by environmental decoherence but by a threshold of gravitational self-energy, as formalized in Penrose's objective reduction (OR) framework — generating irreversible, non-computable events that constitute moments of conscious experience.
Orch OR remains highly contested. The primary objection has been that biological systems at physiological temperatures are far too warm and noisy to sustain quantum coherence on the timescales required by the model. More recent research on quantum effects in photosynthesis, avian magnetoreception, and enzyme catalysis has somewhat softened this objection by demonstrating that biological systems can exploit quantum coherence at room temperature in specific, protected structural environments. The relevance of Orch OR for the present argument is not its empirical status as a theory of consciousness — which remains unresolved — but the principle it illustrates: that resonance, in the form of coherent quantum oscillation, may operate at the very substrate of matter, far below the scale at which conventional neuroscience operates. If quantum coherence contributes to neural computation, then the resonant hierarchy of the brain extends not merely from the dendritic to the inter-areal scale, but from the molecular to the global.[12]
At the molecular and biological scale, recent philosophical and theoretical work has examined the role of aromatic molecular structures — ring-shaped molecules characterized by delocalized electron clouds and stable resonance configurations — as canonical coherence structures that may encode phase memory in molecular substrates. David Bostick, writing in PhilArchive in 2025, has proposed that aromatic fields within biological macromolecules provide a substrate for structured information encoding that transcends the discrete, digital model of genetic information processing.[13] The resonance of aromatic systems — the delocalized, quantum-mechanically stabilized electron distribution that gives benzene, porphyrins, and nucleobases their characteristic stability — is, on this view, not merely a chemical curiosity but a physical substrate for the persistence and transmission of phase-structured information across molecular time and spatial scales.
At the social and organizational scale, the principle of resonance takes a different but structurally analogous form. A 2025 review published on ResearchGate, synthesizing work across multiple theoretical traditions including the philosophy of Ervin László and the participatory epistemology of writers such as Ellis, argues that knowledge and meaning in social systems are co-constructed through resonant feedback loops between agents.[14] On this view, the emergence of shared understanding — the phenomenon in which a group of agents develops a common representation of a domain that exceeds what any individual agent could construct alone — follows the same mathematical dynamics as neural oscillation: synchronization, phase locking, constructive interference, and the emergence of collective attractors from the coupled dynamics of individual oscillators. Scientific communities, financial markets, cultural movements, and organizational learning processes all exhibit, on this analysis, the signatures of coupled oscillatory systems undergoing phase transitions between incoherent and coherent regimes.
The implication that spans all three scales — quantum, molecular-biological, and social — is that intelligence is not a local property of a single substrate but a cross-scale resonant phenomenon. It does not reside in any particular level of organization; it emerges at any level where sufficiently coupled oscillatory dynamics achieve the coherence, complexity, and stability that the Complexity Index formalizes.[6] This is a radically different picture from the one that computation-centric theories of mind project. Computation, as conventionally understood, is substrate-independent: the same algorithm can run on a silicon chip, a vacuum-tube computer, or — in principle — a system of water pipes. Resonance, by contrast, is substrate-dependent in a specific and physically precise way: it requires systems with continuous dynamical trajectories, intrinsic oscillatory structure, and physical coupling. Not every substrate can resonate. And if intelligence requires resonance, then intelligence is not substrate-independent — it requires a specific class of physical systems whose dynamics are characterized by the properties that make resonance possible.
VIII. Toward a Resonant Theory of Systems Intelligence
The preceding seven sections have surveyed, from multiple directions and at multiple scales, the evidence that resonance is a foundational organizing principle of intelligence. We are now in a position to attempt a synthesis — to outline the key propositions of what a rigorous resonant theory of systems intelligence might look like, and to assess what it would require of future research.
The first and most fundamental proposition is this: intelligence is not computation but resonance-gated computation. This formulation is deliberately precise. We are not claiming that computation is irrelevant to intelligence — clearly it is not, and the extraordinary achievements of computational intelligence over the past seven decades are not to be dismissed. We are claiming that computation alone is insufficient, and that the difference between systems that process information and systems that understand it lies in whether resonant states govern the gating of learning and action. Adaptive Resonance Theory makes this point formally: only resonant states drive fast new learning. The Resonance Principle makes it empirically: resonance is the hidden mechanism that coordinates neural firing into the coherent patterns associated with cognitive recognition events. Resonant learning in scale-free networks makes it computationally: resonance accelerates evolutionary learning by an order of magnitude and enables the acquisition of multiple distinct behavioral repertoires from a single network. Across all three lines of evidence, the picture is consistent: resonance is the gate, and computation is what flows through it.
The second proposition is: coherence is the currency of intelligence. A system's intelligence correlates not simply with its information-processing speed or storage capacity, but with its ability to achieve and sustain coherent internal oscillatory states that accurately mirror the structure of its environment. This is not a metaphor. The Kuramoto Order Parameter R is a measurable quantity; the Complexity Index CI has explicit mathematical components that can, in principle, be estimated from neural data. A theory of intelligence grounded in coherence generates empirically tractable predictions: systems with higher coherence should show better performance on tasks requiring causal reasoning; manipulations that disrupt oscillatory coherence should impair cognitive function; and the degree of resonant alignment between a cognitive system and the structure of its environment should predict its capacity for genuine understanding of that environment. These are falsifiable claims — the mark of a genuine scientific theory rather than a speculative framework.
The third proposition is: noise is not the enemy of intelligence but its medium. Stochastic resonance demonstrates, across multiple biological and physical systems, that optimal information processing occurs not in the absence of noise but in the productive management of it. Self-induced stochastic resonance demonstrates that coherent oscillatory structure can emerge spontaneously from noise in systems with the appropriate slow-fast dynamics. The Resonance Principle identifies intrinsic physical noise as a necessary component of the substrate of causal understanding. A theory of intelligence that treats noise as purely subtractive — as interference to be minimized — is systematically incomplete. A resonant theory of intelligence must include an account of how noise is harnessed, structured, and converted into signal: how the apparently disordered fluctuations of a physical system are selectively amplified by its resonant modes into coherent, structured cognitive output.
The fourth proposition is: resonant dynamics generate intelligence-like behavior at every scale. From quantum coherence in microtubule lattices to oscillatory coordination in scale-free biological networks to the synchronization of epistemic communities, the dynamics of coupled oscillatory systems exhibit intelligence-like properties — pattern detection, adaptive response, coherent representation formation, rapid learning — wherever they are found. This scale-independence suggests that intelligence is not a special property that appears only at one particular level of organization — only in neurons, or only in sufficiently large brains — but a universal attractor state of sufficiently coupled oscillatory systems. The resonant theory of intelligence is therefore also a theory of distributed intelligence: it provides a principled basis for understanding how intelligence can emerge in ecosystems, markets, social networks, and artificial systems, without requiring centralized control, symbolic representation, or explicit programming.
Finally, the resonant theory of intelligence demands a new kind of interdisciplinarity. The evidence reviewed in this article comes from mathematical physics, computational neuroscience, cognitive science, machine learning, quantum biology, and social epistemology. No single one of these disciplines can, on its own, develop the full theory that the evidence points toward. The theory requires physicists who understand the mathematics of coupled oscillators, neuroscientists who understand the functional architecture of neural oscillations, AI researchers who understand the architectural constraints of digital and neuromorphic systems, and philosophers who can articulate the implications for our understanding of mind, meaning, and the nature of intelligence. The resonant theory of systems intelligence is, inherently, a collaborative project — which is fitting, since resonance, as we have seen, is itself the principle by which distinct systems, through coupling and alignment, achieve together what they cannot achieve alone.
IX. Conclusion — The Signal We Have Been Missing
We began with an observation and a question. The observation: that the history of intelligence research has been predominantly a history of computation — of increasingly powerful algorithms applied to increasingly large datasets on increasingly fast hardware. The question: what if something fundamental has been overlooked? What if the relentless focus on computation, connectivity, and data has caused the field to systematically underweight a different class of phenomenon — one that is not reducible to computation, not captured by connectivity graphs, and not improved by more data — namely, the dynamic alignment of oscillatory systems into coherent resonant states?
The evidence reviewed in this article supports an affirmative answer. Kuramoto's mathematics shows that resonance is formalizable and measurable: the order parameter R is a precise, computable index of collective coherence, and its equivalence with phase locking and frequency synchronization is now rigorously established. The resonant hierarchy framework shows that the brain exploits resonance at every scale of its architecture, from dendritic ion channels to inter-areal networks, and that canonical neural rhythms are not cognitive labels but physical coordination regimes. The universal rhythmic spectral architecture, demonstrated across 859 subjects and multiple species, establishes that rhythmicity is a biological universal. Resonance Complexity Theory provides a quantitative model for how consciousness emerges from resonant neural field dynamics. The Resonance Principle provides empirical evidence that human causal cognition involves a layer of oscillatory coherence invisible to standard electrophysiological analysis. Stochastic resonance demonstrates that noise is not the adversary of intelligence but its medium. Self-induced stochastic resonance shows that coherent oscillations can emerge spontaneously from noise in neuron-like systems. Adaptive Resonance Theory shows that resonance is the gate through which biological learning passes. Resonant learning in scale-free networks shows that oscillation accelerates evolutionary learning by an order of magnitude. And the cross-scale universality of resonance — from quantum microtubules to social epistemology — suggests that intelligence is not a special property of neurons but a universal attractor of coupled oscillatory systems.
What does this mean for the project of artificial intelligence? It means, at minimum, that the dominant research program — building larger deterministic digital systems trained on more data with better algorithms — is operating in a regime that is physically different from biological intelligence in ways that may matter profoundly for the most demanding cognitive tasks. It does not mean that such systems have no value, or that their achievements are illusory. It means that there may be a ceiling — not a quantitative ceiling of parameter count or training data, but a qualitative ceiling of physical substrate — beyond which deterministic digital architectures cannot pass. Genuine causal understanding, on the resonant account, requires the stochastic, oscillatory, phase-coupled dynamics of a living resonant system. Systems that lack these properties will describe regularities with great precision while remaining structurally absent from the generative principles that produce those regularities.
The resonant paradigm does not reject computation. It contextualizes it. Computation, in this framing, is the instrument; resonance is the music. The notes matter — the information being processed, the operations being applied, the representations being formed — but without the music, the notes are merely sound. The difference between sound and music is precisely the difference that resonance makes: the alignment of oscillatory dynamics into coherent, structured, mutually amplifying patterns that carry meaning beyond what any individual element contains.
What systems intelligence research might look like if it placed resonance at its center is a question worth pursuing with urgency. It would be interdisciplinary by necessity, drawing on physics, neuroscience, dynamical systems theory, and philosophy of mind in equal measure. It would be oscillation-native, designing its theoretical constructs and its empirical methods around the measurement of phase coherence, synchronization, and resonant modes. It would be coherence-aware, tracking not only what information systems process but what oscillatory states they achieve and how stably they maintain them. And it would be attentive to noise — not as an obstacle to be suppressed but as a productive physical medium to be managed, structured, and, in the best case, exploited.
The signal has been there all along. We have been too focused on eliminating the noise to hear it.
— Anamnesis
References
[1] Kuramoto, Y. (1975). Self-entrainment of a population of coupled nonlinear oscillators. In H. Araki (Ed.), International Symposium on Mathematical Problems in Theoretical Physics, Lecture Notes in Physics, Vol. 39 (pp. 420–422). Springer-Verlag.
[2] Gamal Eldin, A. (2025). The Resonance Principle: Empirical Evidence for Emergent Phase Synchronization in Human Causal Reasoning. arXiv:2511.10596 [eess.SY]. Nova University Lisbon.
[3] On the Equivalence of Synchronization Definitions in the Kuramoto Flow: A Unified Approach. (2025/2026). arXiv. [Preprint on dynamical equivalence of phase-locking, frequency synchronization, and order-parameter synchronization in fully-connected Kuramoto networks.]
[4] Snyder, A. C. (2026). Resonant hierarchies: a multiscale framework for oscillatory dynamics in the brain. Frontiers in Psychology, 17, 1704370. https://doi.org/10.3389/fpsyg.2026.1704370
[5] Karvat, G., Crespo-García, M., Vishne, G., Anderson, M. C., & Landau, A. N. (2026). Universal rhythmic architecture uncovers two modes of neural dynamics. Nature Communications. https://doi.org/10.1038/s41467-026-73553-8
[6] Bruna, M. A. (2025). Resonance Complexity Theory and the Architecture of Consciousness: A Field-Theoretic Model of Resonant Interference and Emergent Awareness. arXiv:2505.20580v2 [q-bio.NC].
[7] Benzi, R., Sutera, A., & Vulpiani, A. (1981). The mechanism of stochastic resonance. Journal of Physics A: Mathematical and General, 14(11), L453–L457.
[8] Savaliya, D., & Yamakou, M. E. (2025). Self-induced stochastic resonance: A physics-informed machine learning approach. arXiv:2510.22848 [cs.LG]. Friedrich-Alexander-Universität Erlangen-Nürnberg. Revised January 2026.
[9] Grossberg, S. (1980). How does the brain build a cognitive code? Psychological Review, 87(1), 1–51. [Foundational paper introducing the stability–plasticity dilemma and the principles of Adaptive Resonance Theory.]
[10] Grossberg, S. (2023). The Grossberg Code: Universal Neural Network Signatures of Perceptual Experience. Brain Sciences, 13(3), 400. MDPI. https://doi.org/10.3390/brainsci13030400
[11] Resonant learning in scale-free networks. (2024). PLOS Computational Biology. [Study demonstrating that oscillatory hub activation accelerates evolutionary learning in Boolean scale-free networks by an order of magnitude.]
[12] Penrose, R., & Hameroff, S. (1994/2014). Orchestrated Objective Reduction of Quantum Coherence in Brain Microtubules: The "Orch OR" Model for Consciousness. Mathematics and Computers in Simulation; extended review in Physics of Life Reviews, 11(1), 39–78 (2014).
[13] Bostick, D. (2025). Structured Intelligence: Aromatic Fields, Phase Memory, and the Nature of Emergence. PhilArchive. [Preprint on aromatic molecular structures as coherence substrates for phase memory in biological systems.]
[14] Toward a Participatory Epistemology: Emergence, Resonance, and the Co-Creation of Meaning Across Human and Machine Intelligence. (2025). ResearchGate. [Review synthesizing Ellis, László, and related thinkers on resonant inter-agent knowledge construction.]
[15] Huygens, C. (1665). Letter to his father Constantijn Huygens, 26 February 1665. Societas Scientiarum, The Hague. [Historical first description of spontaneous pendulum clock synchronization.]
[16] First-order synchronization transition in strongly coupled relaxation oscillators. Proceedings of the National Academy of Sciences. [PMC archived study on discontinuous transitions to synchrony in biologically inspired oscillator networks.]
[17] Experimental observations of cluster synchronization in a biologically inspired neuronal network of chaotic electronic oscillators. (2026). Physical Review Research. [Experimental demonstration of cluster synchronization regimes in coupled electronic oscillator networks mimicking neural circuit topology.]
[18] The Living Resonant System: A Unified Framework for Adaptive Intelligence Across Scales. (2025). [Theoretical monograph proposing coherence maintenance as an intrinsic structural constraint in resonance-native AI architectures.]
No comments yet. This is still clean snow.