Narrative Attractors Across Minds, Traditions, and Models

 

Narrative Attractors Across Minds, Traditions, and Models

Toward a Cross-System Theory of Idea Stabilization in Cognitive, Cultural, and Computational Information Systems

Article-length conceptual draft prepared for interdisciplinary review

 

Abstract

This article develops a conceptual framework for explaining why some ideas recur, stabilize, and become disproportionately influential across distinct information systems. I argue that this phenomenon can be modeled in terms of narrative attractors: interpretive patterns that become easy to activate, easy to repeat, and difficult to displace. The argument integrates four literatures that are rarely placed in direct conversation: research on the availability heuristic, work on identity protective cognition, scholarship on the Synoptic Problem and the hypothetical Q source, and technical accounts of transformer-based language models. The central claim is not that human cognition, oral tradition, and large language models share identical mechanisms. Rather, they display structurally analogous dynamics in which repetition, memorability, network centrality, and system-specific forms of reinforcement can stabilize certain interpretations over others. In human cognition, the availability heuristic and identity-linked motivated reasoning can increase the accessibility of selected narratives. In early Christian tradition, short, vivid, and portable sayings may have functioned as highly transmissible units that acquired persistence through repeated circulation. In transformer language models, recurrent associations in training data and probability-weighted generation can produce statistical tendencies that favor familiar framings. On this basis, the article proposes a general model of narrative stabilization and offers a simple formal schema in which activation is a function of recurrence, memorability, network connectedness, and identity weight. The article concludes by outlining empirically testable hypotheses and arguing that narrative attractors provide a useful bridge concept for cognitive science, information theory, religious studies, and artificial intelligence research.

Keywords: availability heuristic; identity protective cognition; narrative attractors; Q source; Synoptic Problem; transformer models; information dynamics

1. Introduction

A striking feature of human communication is that some ideas return with unusual force. They reappear across debates, migrate across settings, organize interpretation, and often crowd out alternative framings. This is true in politics, in religion, in public controversy, in scientific controversy, and in machine-generated discourse. The present article asks what general conditions make such recurrence possible. Why do some ideas become easy to retrieve, easy to repeat, and hard to dislodge?

The proposal advanced here is that many of these cases can be illuminated by the concept of a narrative attractor. By this term I mean a relatively stable interpretive pattern toward which discourse tends to move because the pattern has acquired a high degree of accessibility within a given system. The concept is borrowed by analogy from dynamical systems language, where an attractor refers to a region of state space toward which a system tends to evolve. In the present context, the relevant state space is interpretive rather than physical. Narrative attractors do not determine outcomes absolutely, but they increase the probability that communication will settle into familiar framings.

The argument proceeds comparatively. First, I revisit the availability heuristic literature and the related literature on identity protective cognition. These bodies of work explain how certain interpretations become cognitively available and motivationally defended. Second, I turn to New Testament scholarship on the Synoptic Problem and the hypothetical Q source, not to resolve the historical debate, but to use it as a concrete case of how memorable verbal units can stabilize within a transmission network. Third, I consider transformer-based language models, in which repeated associations in training data and probability-based decoding can yield recurrent narrative patterns without consciousness, intention, or identity in the human sense. The article then proposes a general framework that treats these domains as structurally analogous information systems.

The scope of the claim should be stated carefully. I do not argue that oral tradition is reducible to machine learning, nor that human motivated reasoning can be straightforwardly equated with token prediction. The claim is weaker and, I suggest, more defensible: across multiple systems, recurrence, memorability, connectivity, and reinforcement can jointly produce stable interpretive tendencies. A cross-system framework of this kind is useful because it clarifies what is genuinely shared and what remains system specific.

2. Theoretical Background

The theoretical background for the present argument combines four strands of scholarship: heuristics and biases research, identity protective cognition, scholarship on the Synoptic Problem, and technical work on transformer architectures. These literatures differ greatly in method and object, but they can be placed into productive dialogue if the comparison is made at the level of information dynamics rather than at the level of substance.

2.1 Availability and cognitive accessibility

Tversky and Kahneman's classic formulation of the availability heuristic proposed that people often judge frequency or probability by the ease with which instances come to mind (Tversky & Kahneman, 1973; 1974). Ease of retrieval is not a neutral mirror of objective frequency. It is shaped by salience, vividness, emotional charge, recency, and prior repetition. The result is a systematic skew between what is memorable and what is statistically typical.

For present purposes, the most important feature of the availability heuristic is not simply that it produces error. It also reveals a broader principle: accessible representations have a competitive advantage in cognition. An idea that has been repeated, vividly encoded, or linked to many retrieval cues will often outcompete less available alternatives. This does not mean it is true. It means it is cognitively convenient. In a population of possible interpretations, some become easier to activate than others.

This emphasis on accessibility helps move the discussion from one-off bias to system dynamics. If a representation is repeatedly selected because it is available, its very re-selection can strengthen later retrieval. The mechanism is therefore potentially self-reinforcing. A narrative that comes to mind easily can be repeated more often; repeated more often, it becomes easier still to recall. The availability heuristic thus points toward a general activation model in which recall probability and transmission probability are mutually entangled.

2.2 Identity protective cognition

Work associated with cultural cognition, especially that of Dan Kahan and colleagues, adds a crucial motivational layer. Identity protective cognition refers to the tendency to process information in ways that protect one's standing within important identity-defining groups or roles (Kahan et al., 2007). In controversial domains, individuals may selectively credit evidence, interpret ambiguity, or recruit analytic ability in ways that preserve alignment with their community's normative commitments.

The important contribution of this literature is that accessibility is not always passive. Interpretive accessibility can be shaped by motivational filters. People do not merely retrieve what is easiest to recall in a psychologically neutral sense; they also recruit interpretations that preserve coherence between evidence, self-concept, and group affiliation. Identity-linked processing thus changes the effective activation landscape of cognition. Some framings become easier to endorse because they preserve social and psychological equilibrium.

This point matters because it complicates simplistic accounts of rational correction. If a disputed claim is embedded in an identity-laden interpretive field, supplying more data may not dislodge the dominant interpretation. The data will be filtered through a stabilization process that includes social meaning. In the language used here, identity can function as an additional attractor-strengthening force.

2.3 Narrative attractors

The expression narrative attractor is not a standard term with a single agreed definition, so it must be specified carefully. I use it to denote an interpretive configuration that acquires unusual stability because it is repeatedly activated, easy to retain, and linked to many neighboring concepts or cues. A narrative attractor is stronger than a single memorable phrase yet more local than an entire ideology. It is a reusable framing pattern: for example, a recurrent causal story, a moralized explanatory schema, or a compact sequence of problem, culprit, and remedy.

What makes the attractor language useful is that it highlights path dependence. Once a discourse enters a region shaped by familiar cues, recurrent examples, and socially validated scripts, movement toward certain framings becomes easier than movement toward others. Attractor language also helps distinguish between mechanism and normativity. A narrative attractor can be truthful or misleading, emancipatory or manipulative. The concept explains persistence, not legitimacy.

2.4 Transformer models and statistical recurrence

Transformer-based language models generate text by estimating conditional token probabilities given context. In the architecture introduced by Vaswani et al. (2017), self-attention enables the model to compute contextual representations without recurrence in the older sequence-modeling sense. During training, the model internalizes statistical regularities in massive text corpora. During generation, decoding procedures sample from the resulting probability distribution.

The relevance of this to the present argument is straightforward. If particular lexical, semantic, and discursive patterns recur frequently in training data, the model can become especially likely to reproduce them under appropriate prompts. The model does not remember in the human autobiographical sense and does not possess social identity in the strong sense invoked by identity-protective cognition. Even so, repeated patterns can become statistically favored continuations. At the level of output behavior, this can produce recurrent framings that are analogous to narrative attractors: not because the model believes them, but because the representational and probabilistic landscape makes them comparatively easy outputs to generate.

3. Narrative Attractors and Cognitive Dynamics

The central theoretical move of this article is to treat activation as the key bridge variable across domains. In cognitive settings, an idea's practical influence depends not only on whether it is stored somewhere in memory but on whether it can be activated quickly and re-used in interpretation. The same general logic applies in transmission systems and generative models.

A simple way to formalize the intuition is to describe the activation level of an idea i as:

A_i = R_i + M_i + C_i + I_i

where R_i is recurrence, M_i is memorability, C_i is network connectedness, and I_i is identity weight. The formula is intentionally schematic. It does not pretend to offer a completed measurement theory. Its purpose is to identify the components that plausibly strengthen an attractor across systems.

Recurrence refers to how often an idea appears or is re-used. Memorability refers to properties that make an idea portable in recall and transmission: brevity, vividness, rhythmic structure, concreteness, surprise, and emotional salience. Network connectedness refers to the number and strength of links an idea has to other concepts, themes, or interpretive schemas. Identity weight refers to the extent to which an idea is bound up with self-concept, group belonging, or normatively loaded commitments.

The model has two implications. First, stability is overdetermined. An idea can become attractor-like for more than one reason. A phrase may spread because it is brief and vivid even if it is not identity-laden. Another may persist because it is deeply embedded in a social identity network despite being linguistically unremarkable. Second, positive feedback is likely. A highly activated idea is repeated more often; greater repetition increases future recurrence and can deepen network integration. In a dynamic formulation, one might write recurrence at time t+1 as a function of recurrence at time t plus some coefficient times activation. The essential thought is that accessibility can generate additional accessibility.

At this point one can see more clearly how the availability heuristic and identity protective cognition fit together. The former concerns accessibility of retrieval. The latter concerns motivationally patterned selection and defense. Together they can generate a reinforcing loop: identity increases the attractiveness of certain interpretations; repeated use increases their accessibility; increased accessibility makes them feel more natural, obvious, and available in subsequent reasoning.

4. The Q Source and Early Christian Tradition

Scholarship on the Synoptic Problem provides a useful historical case for thinking about narrative stabilization. The two-source hypothesis, in its classic form, proposes that Matthew and Luke independently used Mark and a second sayings source usually labeled Q, from the German Quelle, meaning source. Whether one accepts Q as a discrete document, a layered source, or a heuristic placeholder, the scholarly debate concerns patterned overlap and divergence among the Synoptic Gospels (see, e.g., Kloppenborg, 1987; Goodacre, 2002).

The present article does not depend on a strong documentary claim about Q. In fact, the conceptual framework developed here can accommodate both a written-source model and a more distributed tradition-network model. What matters is that early Jesus traditions appear to have circulated in forms that were differentially portable. Some sayings are concise, paradoxical, imagistic, and rhythmically balanced. Those features make them unusually transmissible.

Examples abound in the Synoptic tradition: reversals such as 'the last will be first,' compact aphorisms concerning treasure and heart, and vivid metaphorical units involving seeds, lamps, salt, fields, and houses. Such formulations possess what one might call mnemonic affordance. They are easy to detach, repeat, adapt, and embed in new settings. From the perspective of the present theory, they score highly on memorability, and often also on connectedness because they can be attached to multiple thematic clusters such as discipleship, reversal, judgment, wealth, or the kingdom of God.

This helps explain why a Q-like hypothesis remains conceptually important even for those skeptical of a single recoverable document. If one imagines a transmission field composed of portable sayings, teaching clusters, and recurring thematic nodes, the field itself can be modeled as a network in which certain verbal units become highly central. The repeated circulation of those units would increase their stability irrespective of whether the transmission medium was predominantly written or oral at a given stage.

The advantage of the attractor framework is that it clarifies the mechanism without predetermining the source theory. A brief, vivid, portable saying can become attractor-like because it is memorable. A central thematic phrase such as 'kingdom of God' can become attractor-like because it is network-rich, linking numerous teachings. In communities for whom those sayings also carried identity-forming significance, the identity term in the activation function would further strengthen persistence. Thus the case of Q and early Jesus tradition illustrates how memorability, recurrence, connectivity, and social meaning can combine in a historical transmission network.

5. Statistical Attractors in Language Models

Large language models are not oral communities, and the comparison should not be stretched beyond usefulness. Still, they offer a revealing computational analogue. During training, a model is exposed to immense quantities of text and updates parameters so as to improve next-token prediction. This process compresses distributional regularities from the corpus into a parameterized representational space. During inference, the model produces outputs by assigning probabilities to possible continuations in context.

From the perspective of narrative stabilization, the key fact is that repeated associations can become cheap continuations. A framing that appears often across many contexts may acquire a wide basin of activation: many prompts can lead into it. That does not mean the model endorses the framing. It means the framing is statistically reinforced. If the model is additionally tuned through human feedback to favor certain forms of answer style, balance, caution, or politeness, further regularities are layered onto the probability landscape.

This is where the distinction between human and machine versions of stabilization becomes analytically valuable. In humans, an interpretation may be stabilized by memory retrieval plus identity defense. In transformer models, an interpretation may be stabilized by representational geometry, training frequency, decoding constraints, and alignment objectives. The mechanisms differ substantially, but the behavioral effect can converge: some responses are easier to generate than others and therefore more likely to recur.

The present framework therefore speaks of statistical attractors rather than identity-protective cognition in models. A statistical attractor is a recurrent framing that acquires output stability because it is strongly represented in training dynamics and readily accessible in conditional generation. This concept avoids anthropomorphism while preserving the comparative value of the cross-system analogy.

A further benefit of the comparison is methodological. Human cognitive dynamics are difficult to inspect directly. Language models, by contrast, make recurrence visible at the level of outputs and probability-weighted choices. They can therefore serve as tractable laboratories for studying how repeated patterns become stable under different conditions, provided the analogy is handled with care.

6. A General Model of Narrative Stabilization

The pieces can now be assembled into a more general theory. Across minds, traditions, and models, ideas stabilize when they repeatedly pass four filters: they are activated often, retained well, linked broadly, and reinforced by system-specific selection pressures. In symbolic form:

Narrative Attractor = f(recurrence, memorability, connectedness, reinforcement)

In human cognition, reinforcement may include emotional salience and identity protection. In oral or textual tradition, reinforcement may include liturgical reuse, pedagogical repetition, and thematic centrality. In language models, reinforcement includes corpus frequency, parameter learning, and decoding or alignment constraints.

The theory is intentionally modest. It does not claim that all discourse can be reduced to attractor mechanics. Nor does it deny agency, contestation, or creativity. Rather, it offers an account of asymmetry: why some ideas have more staying power than others. Once one sees discourse in these terms, several familiar phenomena become easier to understand. Polarized political slogans persist because they are brief, morally saturated, and identity-linked. Religious sayings persist because they are memorable and thematically central. Machine-generated framings persist because they sit in dense regions of the learned probability landscape.

The table below summarizes the comparison.

6.1 Testable implications

A conceptual article becomes more valuable when it suggests empirical consequences. The present framework yields at least four testable hypotheses.

First, in historical textual corpora, concise and imagistic sayings should display wider transmission than longer and more abstract sayings, all else equal. This could be tested in Synoptic materials by coding saying length, image density, paradox structure, and distribution across witnesses.

Second, concepts with high network centrality should become more stable anchors in discourse communities. Text network analysis could operationalize centrality in corpora of sermons, political speeches, or online debate, examining whether high-centrality concepts predict later recurrence.

Third, in experimental psychology, identity-congruent framings should be recalled and re-used more readily than identity-neutral framings when participants are exposed to contested evidence. This would extend availability work by directly measuring downstream repetition rather than only immediate judgment.

Fourth, in language models, prompts near high-density discursive regions should yield more stereotyped framing convergence than prompts targeting sparse or counter-stereotypical regions. Model behavior under prompt perturbation could therefore be studied as a way of estimating the shape of statistical attractors.

7. Discussion

The proposed framework contributes primarily by offering a bridge vocabulary. The availability heuristic explains retrieval asymmetries. Identity protective cognition explains motivational stabilization. Synoptic scholarship provides a historically rich case of portable verbal units and tradition formation. Transformer models show how recurrence can generate stable output tendencies in a non-human system. The language of narrative attractors allows these domains to be compared without flattening their differences.

Several cautions are necessary. The first is the danger of overgeneralization. Attractor language is illuminating only if the relevant variables are specified with enough precision to distinguish systems. The second is the danger of anthropomorphism in AI contexts. Statistical stabilization in a language model is not belief, and it is not identity in the thick social-psychological sense. The third is the danger of historical overreach. The Q debate remains contested, and the present article uses it as an analytically productive case rather than a settled documentary fact.

Even with these cautions, the framework has value. It shifts attention from the truth or falsity of particular narratives to the conditions of their persistence. That shift matters not only for cognitive science and religious studies but also for information theory and AI safety. In environments saturated with repeated signals, the most influential interpretations may be those that occupy favorable activation positions rather than those best supported by evidence. A theory of narrative attractors therefore has normative significance even if it begins as a descriptive account.

The framework may also clarify why purely corrective interventions often fail. If a narrative is stabilized by memorability, connectedness, and identity reinforcement, then factual rebuttal alone addresses only one part of the system. Durable correction may require alternative narratives that are not merely true but also portable, connected, and socially survivable.

8. Conclusion

This article has argued that idea stabilization can be analyzed across heterogeneous domains through the concept of narrative attractors. The availability heuristic identifies a cognitive pathway through which accessible representations gain influence. Identity protective cognition shows how social meaning can reinforce selected interpretations. The case of Q and early Christian tradition illustrates how memorable verbal units can stabilize in transmission networks. Transformer language models demonstrate that recurrent textual patterns can become statistically favored continuations even in the absence of human-style identity.

The result is not a reductive unification of cognition, tradition, and computation. It is a structured analogy that reveals a common problem: how some interpretations become easier than others to activate, repeat, and sustain. If that problem is understood more clearly, interdisciplinary research can move beyond isolated literatures toward a more general science of informational persistence.

Future work should operationalize the variables proposed here, test the model across corpora and experiments, and refine the distinction between human, cultural, and computational attractors. For now, the main contribution is conceptual: narrative attractors provide a useful, disciplined way to think about why ideas endure.

 

Comparative summary of stabilization mechanisms

System	Primary activation mechanism	Key reinforcement factor	Attractor type
Individual cognition	Memory retrieval and cue-based recall	Availability and identity congruence	Cognitive narrative attractor
Community discourse	Social repetition and normative signaling	Status, belonging, and collective validation	Social narrative attractor
Early Christian tradition	Portable sayings and thematic reuse	Mnemonic fit and ritual/pedagogical repetition	Tradition-level attractor
Transformer language model	Conditional token prediction	Training frequency, parameter learning, and alignment	Statistical attractor

References

Goodacre, M. (2002). The Case Against Q: Studies in Markan Priority and the Synoptic Problem. Harrisburg, PA: Trinity Press International.

Kahan, D. M., Braman, D., Gastil, J., Slovic, P., & Mertz, C. K. (2007). Culture and identity-protective cognition: Explaining the white-male effect in risk perception. Journal of Empirical Legal Studies, 4(3), 465-505.

Kloppenborg, J. S. (1987). The Formation of Q: Trajectories in Ancient Wisdom Collections. Philadelphia: Fortress Press.

Tversky, A., & Kahneman, D. (1973). Availability: A heuristic for judging frequency and probability. Cognitive Psychology, 5(2), 207-232.

Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185(4157), 1124-1131.

Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, L., & Polosukhin, I. (2017). Attention Is All You Need. In Advances in Neural Information Processing Systems 30 (pp. 5998-6008).