Abstract

Hypertext is usually understood as a technology and a paradigm for working with digital text. We propose a stronger reading: hypertext does not impose relational structure on the world but approximates a structure the world plausibly already has. We offer this as a regulative provocation rather than as a settled ontology. Our organising thesis is that, under a suitable explication, structural relation and information can be treated as two descriptions of one structural fact: every information-bearing relation presupposes a relation, and every structural relation carries information. We call this the Relation–Information Co-extensiveness thesis, or RIC. Drawing an edge does not conjure a relation that was not there; rather, under the proposed explication, the edge and the mutual information co-describe a single structural fact. The pay-off is methodological. If reality has anything like a native graph architecture, then Bush, Engelbart, and Nelson did not invent hypertext so much as build technologies that approximate a prior structure. It was, in a sense, always there to be found.

The paper's transferable contribution stands independently of this metaphysical provocation. We introduce H(G), a multidimensional structured profile of the hypertextuality of a system, defined as a vector of graph parameters: relation density, reverse-traversability, decentralisation, modularity, relation-type heterogeneity, transclusion, structural entropy, path compactness, robustness, and generative capacity. H(G) converts hypertextuality from a binary dichotomy into a measurable gradient, allowing comparison of the Web, Xanadu, literary hypertext, and other relational systems within one frame. A preliminary stress-test on a deliberately non-canonical relational graph - 1,535,585 tweets from the FiveThirtyEight Internet Research Agency corpus, yielding 561,588 nodes and 711,950 directed edges - suggests that the profile is not reducible to a single scalar. Principal component analysis indicates that three principal components account for roughly 94% of variance, suggesting that the empirical dimensionality of hypertextuality is itself measurable. In the light of H(G), Xanadu is reread not as a failed project but as an uncompromising attempt to make high-dimensional hypertextuality an explicit design goal.

We offer this as a provocation for conference discussion: what changes in hypertext theory and design if we treat hypertext not as an overlay on text, as the classical tradition of cognitive augmentation assumed, but as a candidate material approximation of the structure of reality itself?

1 Introduction: Why Does Hypertext Feel Natural?

Hypertext does not impose relational structure on the world. It approximates a structure the world plausibly already has. This is not offered as a metaphor, and it is not offered as a proof. It is a regulative claim: a hypothesis about the world's architecture that hypertextual technologies partially materialise, held up so that its consequences for hypertext theory can be examined. We are aware that this courts the charge of metaphysical overreach. We accept the charge and contain it. The paper is built so that its measurement apparatus, introduced in Section 4, survives intact even for a reader who rejects the metaphysics outright. The speculation and the instrument are deliberately separable.

A puzzle motivates the paper. Why does hypertext feel intuitively natural to so many of its users - and, just as tellingly, why do some readers and system designers, in Ted Nelson's phrase, simply “not get it”? The standard answer is that thinking is associative, so a medium that links matches the mind. We think this answer is weaker than it looks. We have very limited introspective access to the mechanisms of our own thought; the felt experience of associative thinking is not reliable evidence about how cognition actually works. If the naturalness of hypertext cannot be securely grounded in introspection about cognition, one is pushed to ask whether its source lies elsewhere: in the structure of what hypertext describes rather than in the structure of the act of describing.

This reframing also corrects a historical point. Engelbart [9] and Nelson [18], developing Bush's [6] memex, were not primarily theorists of cognition, at least not in their early work; their concern was with augmenting intellect and with structuring knowledge so that connections survive. The question “does hypertext feel natural because it matches the mind, or because it matches the world the text is about?” is therefore not idle. It is the fork on which this paper turns, and we take the second horn on purpose. If hypertext approximates the structure of reality, then the tools of Bush, Engelbart, and Nelson do more than extend human capacities: they make contact with an architecture that precedes the act of cognition. They did not invent hypertext. On this reading, they found it; it was always there to be discovered.

Relational accounts of structure are not new. They run from Leibniz's relational conception of space, through the structural realism of French and Ladyman [12], to the logical-empiricist project of the Vienna Circle, which gave us both a deflationary account of structure and much of the conceptual groundwork of computation. We engage these traditions directly in Section 2 because the strongest objection to our thesis - that we are merely stipulating a relation into existence by drawing an edge - belongs to this philosophical lineage. The formalism should clarify commitments rather than manufacture them. Structuralism in the philosophy of science, and in particular the ontic structural realism of [12], already holds that relations are prior to relata; RIC does not re-argue that thesis but operationalises one corner of it, asking how the relational structure of a system can be measured rather than whether relations are fundamental.

This philosophical lineage also has a specifically hypertextual counterpart. Bernstein, Hooper, and Anderson's recent return to the Information City treats hypertext not merely as traversal but as a way of making information structure perceptible [3]. That spatial and structural emphasis is close to our concern here: H(G) treats hypertextuality as a profile of relations, centres, paths, modules, and contextual re-use, rather than as the mere presence of links.

The rest of the paper proceeds as follows. Section 2 states three conceptual clarifications and frames RIC explicitly as an explication rather than a discovery. Section 3 gives a graph formalism, situates it against existing hypertext reference models, and treats classical intuitions from physics as convergences of vocabulary on which the thesis does not depend. Section 4 develops H(G), the paper's transferable contribution. Section 5 draws out implications for hypertext theory and design. Section 6 opens questions for discussion.

2 Three Clarifications and RIC as an Explication

Three concepts require disambiguation before the thesis can be stated precisely.

Information. We distinguish three levels: mathematical, physical, and semantic. Mathematical information includes Shannon mutual information I(A: B), a measure of statistical dependence [8]. Physical information is inseparable from its material carrier, from Landauer's principle to the holographic principle [25]. Semantic information requires meaning, a reference context, and an interpreting agent [11]. RIC operates at the mathematical-structural level only. Convergences with the physical and semantic levels are treated as supporting vocabulary, never as proof, and the thesis does not lean on them. Unless otherwise qualified, “information” below means structural - that is, relational - information.

Relation. We distinguish structural relations, correlations, interactions, and causal relations. Structural relations are represented as edges in a graph: obtaining states of affairs that either hold or do not. Correlations are statistical co-dependencies and do not by themselves imply interaction. Interactions are dynamical couplings. Causal relations involve asymmetric temporal ordering. Physical interaction involves exchange of energy or momentum, but ordinary language also calls belief updates, observations, and social encounters “interactions.” RIC concerns structural relations, not all such semantic or pragmatic events. A hyperlink is a structural relation; it may or may not be typed, directed, reverse-traversable, or dynamically generated, and each of these dimensions enters the H(G) profile.

Explication. We use “explication” in Carnap's sense: the replacement of a vague pre-theoretic concept by a more exact one, to be judged by similarity to the pre-theoretic notion, exactness, fruitfulness, and simplicity [7]. RIC is not presented as a discovery that relation and information are metaphysically identical. It is a proposed regimentation under which relational and informational descriptions can be compared within a shared formal vocabulary.

2.1 RIC as Explication, Not Discovery

The natural objection to any relation–information thesis is sharp and correct: drawing a link from A to B does not bring into being a relation between A and B that did not already exist. We agree. We are not claiming that notation creates ontology. The anti-inflationary lesson of logical empiricism is that formal language should clarify commitments rather than multiply entities.

Accordingly, we frame RIC as a Carnapian explication [7]: a deliberate regimentation that replaces a vague pre-theoretic pairing of “relation” and “information” with a precise one. Under this regimentation, “there is a structural relation between A and B” and “there is positive mutual information between the states of A and B” are stipulated as two descriptions of a single structural fact. The edge and the mutual information do not create one another, and neither is ontologically prior. They co-describe. This formulation avoids the stronger and misleading claim that either relation or information ontologically generates the other.

Let R(A, B) denote the structural relation between objects A and B in a graph G, and I(A: B) the Shannon mutual information between the distributions of their states.

RIC, strong form, as explication:

An edge joining two statistically independent nodes, where I(A: B) = 0, is, under this regimentation, not a structural relation in the sense of RIC but an artefact of graph notation. This clause makes the strong form analytic: it is true by the design of the explication, not by empirical discovery. That is not a concealed weakness but a feature of an explication, whose interest lies in fixing a shared measurement language in which relational and informational descriptions can be compared.

RIC, weak form:

The weak form is logically uncontroversial and suffices to establish that informational structure implies relational architecture. H(G) is independent of the choice between the strong and weak forms.

We therefore do not assert that the world is a Shannonian network, that all information is conscious, that consciousness is information, or that RIC resolves the mind–body problem. RIC is a precision tool for talking about structure, plus a regulative bet that the tool fits the world. Its standing is to be tested by internal coherence, by fit with research programmes in the foundations of physics, and above all by heuristic fertility in the domain of hypertext.

2.2 Convergences We Do Not Lean On

Several research programmes use a vocabulary in which the relation/information distinction is already thin. We list them as convergences and explicitly state that the thesis does not rest on any of them.

In the foundations of physics, work deriving spacetime structure from entanglement structure [24] blurs, at the fundamental level, the line between “relation” and “information.” The claim that entanglement shows structural relations can be non-local is one interpretation of the Einstein–Podolsky–Rosen situation, not the only one: on an epistemic reading, particle B had its relevant property all along and the only change on measuring A is that an observer comes to know it. Moreover, non-locality of structural relation is not peculiar to entanglement. Gravitation, electric charge, and magnetism furnish familiar macroscopic examples of action at a distance. We therefore do not argue from physics to RIC. Whether EPR is ontic or epistemic is irrelevant to an explication, which is a framework rather than a physical hypothesis. The convergence is one of vocabulary, and we present it as nothing more.

From logic, Bateson's “a difference that makes a difference” presupposes at least two distinguishable elements, the minimal structural relation. RIC takes from Bateson only this structural layer.

From structural realism, Heller [16] argues that the universe is a structure and identifies that structure with information; French and Ladyman [12] treat structure as a pattern of relations. If structure is information and structure is a pattern of relations, then within the explication relation and information are co-extensive by transitivity. This is the tradition our thesis belongs to, and we present RIC as a sharpening of it rather than as a novelty without ancestry.

3 Hypertext as Relational-Informational Architecture

3.1 Formalisation

We model hypertext as a labelled graph G = (V, E, σ, τ), where V is the set of nodes, E⊆V × V the set of edges, σ: V → Σ the node-content function, and τ: E → T the relation-type function. T includes semantic, hierarchical, associative, temporal, and formal-causal types. The graph may be directed, undirected, or mixed; simple or a multigraph; it may have transclusion, where σ(v) depends on σ(u) for some u ≠ v.

This is not a new hypertext reference model, and we do not present it as one. It is a deliberately minimal relabelling whose only purpose is to foreground the relation/information pairing for the purposes of H(G). Serious formal modelling of hypertext has been done elsewhere, and a paper that puts a graph on the page owes that work a gesture: the Dexter Hypertext Reference Model of Halasz and Schwartz [15], the Petri-net-based document semantics of Stotts and Furuta [23], the design-model tradition of Garzotto, Paolini, and Schwabe [13], and open-hypermedia integration work associated with Whitehead [26]. Our contribution is not a competitor to these; it is an analytical lens that can be applied on top of any of them.

If the regulative bet of Section 2 holds, the architecture of reality takes the form of a graph G in the same formal category, and the Engelbart–Nelson–Berners-Lee tradition technologically approximates it. The methodological question then becomes: how do we measure the degree of approximation? A single number will not do. A structured profile is needed.

3.2 Four Structural Intuitions, Held Loosely

There is a temptation to identify entanglement with the hyperlink, holography with transclusion, and so on. We reject these identifications as too strong. Instead, we identify four structural intuitions that hypertext and certain physical descriptions express independently.

Non-locality. A structural relation can connect distant elements. Hypertext is non-sequential: a link can lead from any node to any other regardless of position. Entanglement is one illustration of non-local structural relation in physics, but it is neither the only such illustration nor the only reading of the phenomenon. Classical action at a distance makes the same point without relying on the quantum case. We rely on the general intuition, not on a specific interpretation of quantum mechanics.

Reverse-traversability. Nelson's links are not symmetric; they are asymmetric but reverse-traversable. Given a node N, one can always ask “what links to N?” [19]. The relevant H(G) coordinate is therefore not link symmetry but the degree to which a system supports reverse traversal, which is the property Xanadu actually sought to maximise.

Contextual participation, or transclusion. In Xanadu, the same content is present in many documents by reference. In holography, the same volume information is encoded on boundary fragments. The shared intuition is that a unit of content can participate in many contexts without losing its source identity. This is the strongest of the four analogies and the one most directly relevant to hypertext design.

Absence of a privileged centre. Many hypertexts have authored entry points: afternoon has a cover page and Google has a home page. The claim is not that hypertexts lack entry points. The claim is structural: the formalism requires no canonical root, and there is no node whose removal collapses the meaning of the whole. Rovelli's relational quantum mechanics likewise admits no structurally privileged observer [22]. Authored entry points and structural decentralisation are different things, and only the latter is at issue. At the level of network analysis, structural decentralisation does not imply the absence of empirical centrality [17]. A system can be decentralised in principle and hub-dominated in fact.

4 H(G): A Multidimensional Hypertextuality Profile

The formalisation raises a methodological question that is independent of the metaphysics. If hypertext is a labelled relational-informational graph, how do we measure the degree to which a system realises hypertextual properties? The question should not be binary. The mere presence of links is insufficient: a system can be link-rich yet centralised, semantically homogeneous, poorly reverse-traversable, and structurally brittle. Hypertextuality is better understood as a multidimensional profile than as a scalar.

This continues earlier work. Botafogo, Rivlin, and Shneiderman analysed hypertext through structural metrics that identify hierarchies and navigational complexity [5]. Bernstein argued that linking patterns co-constitute the cognitive, functional, and compositional structure of a system [1, 2]. Research on link topology analyses relations through communities, hubs, and local patterns [14, 17, 21]. Bernstein, Hooper, and Anderson's return to the Information City further emphasises the spatial and perceptual intelligibility of hypertextual structure [3].

Given G = (V, E, σ, τ), the hypertextuality profile is an ordered vector:

The ten coordinates are provisionally defined as follows:

d - relation density: edges as a fraction of possible edges.\

ρrt - reverse-traversability: the degree to which “what links here?” is answerable, generalising Nelson's link reversibility.\

D - decentralisation: the extent to which the structure lacks a dominating centre, distinct from authored entry points.\

Q - modularity: the strength of community structure under a standard partition [4, 20].\

η - relation-type heterogeneity: the diversity of τ over E.\

θ - transclusion or multi-contextuality: the extent to which content participates in multiple contexts by reference.\

S - structural entropy: the disorder of the connection pattern.\

P - path compactness: typical reachability or path length, normalised.\

κ - structural robustness: resistance of the giant component to targeted hub removal.\

γ - generative capacity: the degree to which structure is produced dynamically rather than fixed.\

These coordinates were selected to span five recurrent design dimensions in the hypertext literature: connectivity, captured by d and P; reversibility and access, captured by ρrt; organisation, captured by D, Q, and S; semantic or contextual re-use, captured by η and θ ; and temporal or systemic behaviour, captured by κ and γ.

We make two limitations explicit. First, we do not claim the ten coordinates are necessary and sufficient. We claim they are a candidate profile, chosen to span dimensions the hypertext literature has found load-bearing, and we put it forward so that the community can prune, merge, and extend it. Second, the coordinates are not independent: one expects, for instance, modularity Q and robustness κ to be related. We treat that not as an embarrassment but as an empirical question the profile is built to answer. The point of keeping a vector rather than collapsing to a scalar is that two systems can reach similar aggregate scores while differing structurally - one dense and centralised, the other sparse but modular and robust - and the vector preserves the difference.

We retain the symbol H(G) for the hypertextuality profile. It is not a Hamiltonian, nor is it Boltzmann or Shannon entropy. Structural entropy enters the profile only as one coordinate, S.

4.1 Stress-Testing the Apparatus on a Non-Canonical System

As a stress-test, we applied H(G) to a deliberately borderline system: a directed interaction graph built from the FiveThirtyEight Internet Research Agency dataset [10], with 561,588 nodes and 711,950 directed edges. The point is not to classify the IRA graph as hypertext, but to test whether H(G) can describe relational systems beyond canonical document-hypertexts without collapsing them into a scalar.

Table 1: H(G) profile for the IRA interaction graph.

Coordinate	Symbol	Value
Relation density	d	2.26 × 10− 6
Reverse-traversability	ρrt	8.99 × 10− 5
Decentralisation	D	0.543
Modularity	Q	0.567
Relation-type heterogeneity	η	0.778
Transclusion / multi-contextuality	θ	0.119
Structural entropy	S	0.093
Path compactness	P	0.500
Structural robustness	κ	2.32 × 10− 5
Generative capacity	γ	0.928

The graph is simultaneously extremely sparse, almost entirely non-reverse-traversable, moderately decentralised, distinctly modular, relatively heterogeneous, weak in multi-contextuality, and highly vulnerable to targeted hub removal. Removing the most central 1% of nodes collapses the giant component from 561K nodes to a small residual. Yet γ ≈ 0.928 indicates strong generative dynamics: this is not a stable document-hypertext but a temporally evolving relational system.

In a principal component analysis of this corpus, three principal components accounted for roughly 94% of variance. The result is suggestive only, since confirmation requires multiple corpora and a fully specified sampling protocol. We read the result in two ways. It vindicates the decision to keep a vector, since the coordinates are not interchangeable; but it also concedes that the ten coordinates are partly reducible and should not be mistaken for a basis. The empirical dimensionality of hypertextuality is itself a measurable quantity. These results are preliminary, on a single corpus, and we offer them not as validation but as an invitation to stress-test, extend, and contest the profile.

4.2 Comparative Application

H(G) gains its value comparatively. The standard Web shows large scale, high reachability, and strong hubs, but low reverse-traversability and limited transclusion [14, 17]. Xanadu is a regulative model in a methodological sense: it guides comparison by making several design dimensions explicit, without requiring that any implemented system fully realise them. It foregrounds reverse-traversability, transclusion, source traceability, and durable re-use [19]. Literary hypertext in environments such as Storyspace operates at smaller scale but can show higher local semantic motivation. AI-augmented systems are best analysed through generative capacity, relation heterogeneity, multi-contextuality, and structural entropy. H(G) translates the relational-informational intuition into the language of structural measures.

5 Implications for Hypertext Theory and Design

5.1 Hypertext as Ontological Laboratory

If the regulative bet holds, hypertext systems are not only knowledge-management tools but experiments in describing the architecture of reality. Poems and novels are also experiments in describing reality; literary hypertext is continuous with this broader project rather than an exception to it. Storyspace, Tinderbox, and the work of Joyce, Moulthrop, and Jackson belong in the argument rather than at its margins. When Storyspace introduces conditional links, or Tinderbox grows structure from prototypes, the system probes the limits within which information can be organised into relations. Collaboration between hypertext theory and the metaphysics of information [11, 25] becomes natural rather than extravagant.

5.2 AI and the Question of Participation

The HT 2026 call frames the field against the prospect that an estimated 98% of textual production may move toward AI generation. This makes the question of machine linking unavoidable. Does machine linking participate in RIC? If an AI-generated link is only a statistical projection from a training set, machine nodes are phantom nodes: σ without genuine τ, relations syntactically present but structurally illusory. If AI linking reflects genuine semantic dependencies, AI hypertexts participate in the same architecture as human texts, with a different kind of agency. H(G) makes this measurable: does AI-generated hypertext show a profile close to human hypertext, or does it diverge systematically - for example, through higher density but lower relation-type heterogeneity?

5.3 A Rereading of Xanadu

Xanadu was for decades dismissed as utopian and unimplementable. The profile view explains why it remained unfinished and why it remains a reference point. Xanadu was not the only system to pursue several hypertextuality parameters at once: NoteCards, VIKI, Intermedia, and Ward's Wiki all targeted multiple parameters. The defensible and more interesting claim is that Xanadu made high-dimensional hypertextuality its explicit design telos, insisting on reverse-traversal, transclusion, and source integrity as jointly necessary design goals: ρrt → 1, θ → 1, η high. Its technical incompleteness does not invalidate its value as a regulative ideal against which other systems can be measured through H(G).

5.4 H(G) as a Transferable Tool

Independently of RIC, H(G) has standalone methodological value. It enables classification and comparison beyond the hypertext/not-hypertext dichotomy; operationalised evaluation of new systems such as Roam, Obsidian, Logseq, and AI-augmented hypertext; measurement of a single system's evolution over time; and design with explicit structural targets. This methodological contribution is separable from the metaphysical thesis and is offered to the field whether or not RIC is accepted. We invite the community to calibrate the parameters, add dimensions, and validate the profile against real corpora.

5.5 Social Implications

The Web became dominant not by realising every hypertextual ideal associated with Xanadu, but by making a simpler subset of linking practices widely implementable. Today, with social platforms built on one-directional linking and algorithmic selection, the profile H(social media) shows low reverse-traversability, high centralisation, and low relation-type heterogeneity. We state the next step as a hypothesis, not a proven causal law: the pathologies observed in this infrastructure - filter bubbles, asymmetric participation, and the easy travel of disinformation - correlate structurally with these parameters, and H(G) makes the correlation quantifiable rather than merely qualitative. Whether the structural mismatch between a putatively bidirectional reality and a unidirectional medium is causal is precisely the kind of question the profile is meant to let the field ask with measurements rather than slogans.

6 Open Questions for Discussion

Operationalising H(G). What weights, if any, should the coordinates carry? Is there a justification derived from RIC, or is this purely empirical calibration? The IRA data suggest, under principal component analysis, that three components capture most variance; confirmation on other corpora is needed, and the empirical dimensionality of hypertextuality is itself an open quantity.\

Status of H(world-structure). Does it make sense to speak of the hypertextuality profile of physical structure, or is this only a regulative idealisation? What profile would a graph of holographic entanglement structure have?\

Testability through H(G). Could certain H(G) configurations on real systems count against the regulative bet? A trial hypothesis: functionally effective hypertext systems should cluster near a particular attractor. Long-running systems such as Wikipedia, ACM DL, and arXiv could test it.\

Agency in AI. What criterion distinguishes AI linking that participates in RIC from statistical simulation? Does H(GAI) differ measurably from H(Ghuman), for appropriately constructed AI-generated and human-authored hypertext graphs?\

The future of literary hypertext. If Storyspace, Tinderbox, and the works of Joyce and Moulthrop are proving grounds for human perception of networked structure, which narratives after 2025 develop this function most effectively - and is AI-augmented hypertext their extension or their replacement?\

Limits of the analogy. Where does the productive comparison between physical and hypertextual structure end? Are there properties of physical structures, for example quantum no-cloning, with no hypertextual analogue, marking the boundary of RIC in its strong form? Naming these limits is part of the thesis, not a threat to it.

Source

ACM full-text HTML

ACM Digital Library

Imported from ACM’s structured HTML source. ACM Reference Format: Andrzej Adamski, Władysław Błocki, Mariusz Pisarski, and Marcin Szewczyk. 2026. Hypertext as the Native Architecture of Reality. In 37th ACM Conference on Hypertext (HT '26), September 14--18, 2026, London, United Kingdom. ACM, New York, NY, USA 6 Pages. https://doi.org/10.1145/3800935.3830889

References

[1] Mark Bernstein. 1998. Patterns of Hypertext. In Proceedings of the Ninth ACM Conference on Hypertext and Hypermedia(HT ’98). Association for Computing Machinery, 21–29. https://doi.org/10.1145/276627.276630

[2] Mark Bernstein. 1999. Structural Patterns and Hypertext Rhetoric. Comput. Surveys 31, 4es, Article 19 (1999). https://doi.org/10.1145/345966.346011

[3] Mark Bernstein, Silas Hooper, and Mark W. R. Anderson. 2025. Back To The Information City. In Proceedings of the 36th ACM Conference on Hypertext and Social Media(HT ’25). Association for Computing Machinery, 118–126. https://doi.org/10.1145/3720553.3746664 — ACM HyperText copy

[4] Vincent D. Blondel, Jean-Loup Guillaume, Renaud Lambiotte, and Etienne Lefebvre. 2008. Fast Unfolding of Communities in Large Networks. Journal of Statistical Mechanics: Theory and Experiment 2008, 10 (2008), P10008. https://doi.org/10.1088/1742-5468/2008/10/P10008

[5] Rodrigo A. Botafogo, Ehud Rivlin, and Ben Shneiderman. 1992. Structural Analysis of Hypertexts: Identifying Hierarchies and Useful Metrics. ACM Transactions on Information Systems 10, 2 (1992), 142–180. https://doi.org/10.1145/146802.146826

[6] Vannevar Bush. 1945. As We May Think. The Atlantic Monthly 176, 1 (1945), 101–108.

[7] Rudolf Carnap. 1950. Logical Foundations of Probability. University of Chicago Press, Chicago.

[8] Thomas M. Cover and Joy A. Thomas. 2006. Elements of Information Theory (2 ed.). Wiley.

[9] Douglas C. Engelbart. 1962. Augmenting Human Intellect: A Conceptual Framework. Technical Report AFOSR-3223. Stanford Research Institute, Menlo Park, CA.

[10] FiveThirtyEight. 2018. Russian Troll Tweets. Dataset, GitHub repository. https://github.com/fivethirtyeight/russian-troll-tweets

[11] Luciano Floridi. 2011. The Philosophy of Information. Oxford University Press, Oxford.

[12] Steven French and James Ladyman. 2003. Remodelling Structural Realism: Quantum Physics and the Metaphysics of Structure. Synthese 136, 1 (2003), 31–56. https://doi.org/10.1023/A:1024156116636

[13] Franca Garzotto, Paolo Paolini, and Daniel Schwabe. 1993. HDM—A Model-Based Approach to Hypertext Application Design. ACM Transactions on Information Systems 11, 1 (1993), 1–26. https://doi.org/10.1145/151480.151483

[14] David Gibson, Jon M. Kleinberg, and Prabhakar Raghavan. 1998. Inferring Web Communities from Link Topology. In Proceedings of the Ninth ACM Conference on Hypertext and Hypermedia(HT ’98). Association for Computing Machinery, 225–234. https://doi.org/10.1145/276627.276652

[15] Frank Halasz and Mayer Schwartz. 1994. The Dexter Hypertext Reference Model. Commun. ACM 37, 2 (1994), 30–39. https://doi.org/10.1145/175235.175237

[16] Michał Heller. 2009. Ultimate Explanations of the Universe. Springer, Berlin.

[17] Jon M. Kleinberg. 1999. Authoritative Sources in a Hyperlinked Environment. J. ACM 46, 5 (1999), 604–632. https://doi.org/10.1145/324133.324140

[18] Theodor Holm Nelson. 1993. Literary Machines 93.1. Mindful Press, Sausalito, CA.

[19] Theodor Holm Nelson. 1999. Xanalogical Structure, Needed Now More Than Ever. Comput. Surveys 31, 4es, Article 33 (1999). https://doi.org/10.1145/345966.346033

[20] Mark E. J. Newman and Michelle Girvan. 2004. Finding and Evaluating Community Structure in Networks. Physical Review E 69, 2 (2004), 026113. https://doi.org/10.1103/PhysRevE.69.026113

[21] Seongbin Park. 1998. Structural Properties of Hypertext. In Proceedings of the Ninth ACM Conference on Hypertext and Hypermedia(HT ’98). Association for Computing Machinery, 180–187. https://doi.org/10.1145/276627.276647

[22] Carlo Rovelli. 1996. Relational Quantum Mechanics. International Journal of Theoretical Physics 35, 8 (1996), 1637–1678. https://doi.org/10.1007/BF02302261

[23] P. David Stotts and Richard Furuta. 1989. Petri-Net-Based Hypertext: Document Structure with Browsing Semantics. ACM Transactions on Office Information Systems 7, 1 (1989), 3–29. https://doi.org/10.1145/64789.64791

[24] Mark Van Raamsdonk. 2010. Building up Spacetime with Quantum Entanglement. General Relativity and Gravitation 42, 10 (2010), 2323–2329. https://doi.org/10.1007/s10714-010-1034-0

[25] John Archibald Wheeler. 1990. Information, Physics, Quantum: The Search for Links. In Complexity, Entropy, and the Physics of Information, Wojciech H. Zurek (Ed.). Addison-Wesley, Redwood City, CA, 3–28.

[26] Jr. Whitehead, E. James. 1997. An Architectural Model for Application Integration in Open Hypermedia Environments. In Proceedings of the Eighth ACM Conference on Hypertext(HT ’97). Association for Computing Machinery, 1–12. https://doi.org/10.1145/267437.267438