We define a network with no central gravity as one in which no single authority controls identity resolution, object addressing, or interaction semantics (as in Fig. 1). Distributed objects retain persistent identities and can migrate across servers without losing referential integrity.

Objects can exchange messages and interact directly, whether hosted locally or synchronised across peers. Coordination emerges from protocols and shared representations rather than centralised infrastructure. Control is not anchored to any specific server, platform, or authority.

This concept builds on ideas from Alan Kay regarding object-oriented messaging and late binding [11], as well as Leslie Lamport's work on vector clocks [13].

Seed Hypermedia [3, 23] implements these principles as an open hypermedia network for the Web. It enables collaboration across distributed nodes while preserving authorship through cryptographic signatures and ensuring that content integrity can be independently verified.

Using this system, we propose an experiment at the Hypertext Conference 2026 in London to evaluate a networked hypermedia system in a conference setting. Grounded in Engelbart's augmentation framework [7], the goal is to examine whether transforming the proceedings into a networked knowledge repository—where participants annotate, link, and extend documents—supports the emergence of shared meaning. We further investigate whether such interaction can amplify collective intelligence and reshape how knowledge is produced and organised.

Engelbart's augmentation framework provides a pragmatic model of human augmentation, defining the co-evolution of tools, language, and practices. While the core primitives he identified remain largely unchanged, their technical realisation has evolved significantly. Seed can be understood as a contemporary implementation of these principles under the conditions of a distributed network.

Hypertext remains an unfinished revolution. Early systems such as Augment [8], Xanadu [19], HyperCard [2], Intermedia [17] and later open hypermedia approaches such as Microcosm [9] and Hyper-G [16] explored rich models of linking, authorship, traceability, and collaboration. The Web, however, achieved global adoption by prioritising simplicity and ease of deployment over this richer functionality. As a result, foundational hypertext capabilities such as persistent links, transclusion, fine-grained provenance, and dynamic structures were largely left aside.

Today's dominant digital platforms impose a form of central gravity, whereby centralised infrastructures mediate and shape interaction. Although effective at large-scale distribution and attention capture, these systems restrict ownership and limit the possibilities for human collaboration. Participation is enabled, yet co-authorship of the medium itself remains structurally constrained.

By contrast, a network without central gravity reintroduces the openness, uncertainty, and creative potential that characterised the early promise of hypertext as seen before the advent of the Web. In such a network, communities are not merely users of a platform; they are active participants in constructing their own knowledge environments. As we argue through the design of Seed Hypermedia, open hypertext systems can support decentralised collaboration while preserving structure, provenance, and meaning.

In Augmenting Human Intellect: A Conceptual Framework  [7], Douglas Engelbart defines augmentation as improving the ability to comprehend complex situations and derive better solutions. This enhancement does not increase innate intelligence, but reorganises it through systems of tools, language, methods, and training (‘HLAM/T’).

Intelligence emerges from gaining capabilities through hierarchically organised processes, where higher-order cognition builds on simpler operations. These processes can be expanded through artifacts and practices, forming a system of “intelligence amplification”.

Tools shape these processes. Software is not only a container for information; it enables or constrains particular forms of thought, association, and coordination. Our aim is therefore not just to build a repository of content, but to create an infrastructure that improves the processes by which communities construct and revise shared understanding.

The process begins as participants share their “feel for a situation” [[7], para. 1a1]—their intuitions, hunches, and partial understandings—which coexist with formal concepts, precise notation, sophisticated methods, and computational tools. Augmentation arises from integrating these elements into a system that supports both exploratory thinking and structured knowledge work.

Collective thinking within knowledge communities improves when ideas are externalised into structured representations that can be shared, inspected, and transformed. Intelligence is not solely an individual property, but a collective phenomenon emerging from coordinated activity within a community.

We operationalise Engelbart's framework through a recurring process:

Conversation, where communities engage in argumentation, express perspectives, and negotiate shared understanding;

Concept creation, where concepts are introduced and refined, serving as tools for the augmentation process;

Structure formation, where knowledge emerging from these interactions is captured and organised into hierarchically open hypertext documents and ontologies;

Reorganisation, where structures evolve as understanding deepens and knowledge emerges as knowledge output.

Knowledge is continuously reconstructed through dialogue and argumentation. Seed integrates conversational flow and structured knowledge by treating conversations as first-class units of information, on par with document paragraphs. Through transclusion, versioning, and multiple views, these elements can be reused, reorganised, and interpreted across contexts. This enables the emergence of shared meaning, understood as the sustained alignment of concepts within a community.

Engelbart identified the interactive primitives underlying the augmentation process. Although these primitives—linking, structuring, versioning, referencing, and collaborative editing—have remained conceptually stable, their technical realisation has evolved significantly over the past five decades.

Recent advances in distributed systems, cryptographic identity, version control algorithms, and peer-to-peer networking make it possible to revisit long-standing hypertext and decentralisation ambitions with new practical realism.

Systems such as IPFS[1] [4], Secure Scuttlebutt [25], Git, Bitcoin [18], CRDTs [24] and the PGP Web of Trust [5] demonstrate that decentralised coordination and authorship can be achieved without centralised infrastructure. These developments make it feasible to combine ideas from NLS/Augment and Xanadu with distributed object models inspired by Alan Kay's work.

The following capabilities characterise the emerging technological substrate that enables a network with no central gravity:

Documents are immutable and versioned through content-addressing, using hash functions and cryptographic signatures, forming directed acyclic graphs (DAGs) of changes that can be partially ordered using mechanisms such as vector clocks;

Users retain control over their cryptographic keys, enabling authorship and identity to be verified without third-party authorities;

Synchronisation occurs through peer-to-peer message exchange, forming a distributed “sea of messages” rather than relying on centralised servers.

Permissions (e.g. read and write access) are enforced through capability-based cryptographic certificates;

Links are persistent, unbreakable, and potentially bidirectional, enabling robust referential integrity across the network;

Content can be transcluded rather than duplicated, preserving provenance and enabling reuse across contexts;

Communities can define and evolve their own structures, interfaces, and workflows, rather than inheriting them from a centralised platform.

The goal is not only technical decentralisation, but enabling communities to coordinate thought without surrendering control of their intellectual environment.

Seed's design is influenced by Ted Nelson's vision of Xanadu [19], particularly its emphasis on the concept of hypermedia servers holding authors’ digital rights, transclusion [20], immutable versions, and networked knowledge. Seed Hypermedia adopts a xanalogical [21] model of sharing, in which content remains immutable and reusable, while new contexts, interpretations, and interfaces can be layered on top.

While Seed supports interaction with objects independently of their physical location, it retains the notion that canonical attribution and rights are anchored in specific nodes. In this sense, hypermedia servers are not centres of control, but points of authority for provenance, where authorship and licensing are persistently maintained.

By default, content in the network is published under Creative Commons Attribution 4.0 (CC BY 4.0), ensuring that knowledge remains reusable and remix-able while preserving attribution. However, legal openness alone is not sufficient.

Wikipedia demonstrates the power of open collaboration at a planetary scale. Its success is rooted less in technological novelty than in social innovation: anyone can edit its documents, collaboration is organised around ongoing conversations, and contributions are guided by well-defined workflows and norms [15]. A key enabler of this model is its commitment to free licensing, which ensures that content can be reused and redistributed without restriction.

Wikipedia represents only a partial realisation of the hypertext vision. It prioritises stability and consensus over fluid interaction and branching exploration, and lacks native support for (Xanalogical) transclusion[2], fine-grained provenance, and dynamic restructuring.

Governance remains a central challenge. Even in good-faith communities, questions of authority and conflict resolution persist [12]. Unlike conceptual systems, real-world platforms must address these tensions directly. This highlights the need for hypertext systems that integrate both technical capabilities and adaptable governance models from the outset.

Despite these limitations, Wikipedia remains a foundational example of large-scale knowledge collaboration. It demonstrates how collective stewardship can sustain and scale shared repositories and collaborative intellectual work.

Academic conferences are not primarily environments for structured knowledge work, but for social interaction. Their core function is to enable participants to meet, exchange ideas, and engage in conversation. While papers form the backbone of the scholarly record, they are typically written for careful, long-form reading and are therefore not fully consumed during the event itself. In addition, presentations of accepted papers often reinterpret or extend the written work, offering perspectives that may differ from the paper while remaining equally valuable.

This creates a well-established behavioural pattern: attendees prioritise interpersonal exchange over in-depth engagement with the proceedings. Historically, conference papers were distributed on arrival in printed form and perhaps only read fully after the event, a pattern that persists despite digital access. As a result, conferences allocate limited time and attention to activities such as reading, annotating, or systematically organising knowledge.

These conditions introduce a significant challenge for systems like Seed. Engaging with such systems requires participants to perform explicit knowledge work, e.g. installing, configuring, and interacting with unfamiliar tools, within a context that prioritises immediacy and face-to-face interaction. The perceived benefit of structured, persistent interaction is not immediately aligned with the motivations that drive conference participation.

In addition to behavioural constraints, structural factors inherited from the print era continue to shape conference timelines. Papers are often finalised and released close to the start of the event, limiting opportunities for pre-conference engagement. Submission delays, review cycles, and production requirements, such as fixed publication windows, further compress the time available for participants to interact meaningfully with the material. There is also the issue of negotiating early release of papers as circulation is often embargoed until the start of the conference. Much of this is a direct result of the age of print media and as publication moves fully digital conventions may change.

These constraints do not undermine the potential of systems like Seed, but they do define the conditions under which such systems must operate.

Academic conferences can be understood as distributed memory systems in which shared meaning is constructed and maintained through a dense network of interactions. While formal outputs such as papers and proceedings provide a partial record of knowledge, they capture only a fraction of the intellectual activity that takes place within these communities. The greater part resides in conversations, relationships, shared references, remembered debates, and recurring practices that unfold over time. In this sense, an academic conference functions as a community memory palace: a cognitive space distributed across its participants, where meaning is generated through dialogue, reinforced through repeated encounters, and shaped by a shared history that exists both in individual memory and in collective awareness.

This perspective reveals a fundamental limitation of current digital systems. While they are effective at storing and retrieving documents, they largely fail to preserve the memory structures. As a result, the deeper continuity of the conference—its evolving interpretations, social ties, and accumulated understanding—remains largely ephemeral, rather than becoming a persistent, inhabitable space that future participants can meaningfully explore and build upon.

In a hypermedia network with no central organising authority, the primary unit of knowledge is not the individual document or user, but the community. Meaning does not emerge from isolated contributions, but from the continuous interaction, interpretation, and negotiation that occurs within bounded groups of participants. As Etienne Wenger argues, knowledge is socially constructed through communities of practice, where shared context and sustained engagement enable collective understanding [26]. In this setting, each community effectively becomes a locus of meaning-making, responsible for organising, maintaining, and evolving its own knowledge structures.

A conference can be understood not as a single, unified community, but as a temporary micro-network of networked improvement communities. Each of these communities—whether rooted in Digital Humanities, Computer Science, or Hypertext—brings its own conceptual frameworks, terminologies, and evaluative criteria. As a result, the same term or concept (e.g. document, link, authorship, or interaction) may carry significantly different meanings depending on the community in which it is used.

Rather than treating this multiplicity as a barrier, it can be reframed as a productive condition. The conference becomes a site where semantic negotiation occurs: meanings are not fixed but emerge through interaction across communities. Each improvement community continues to evolve its own internal coherence while simultaneously contributing to a broader, shared discourse.

Actors—authors, attendees, organisers—participate in this process by engaging with materials and discussions originating outside their primary domain. Through mechanisms such as annotation, transclusion, and commentary (as enabled in systems like Seed), participants can situate their interpretations explicitly, making visible how a concept is understood within a given context. This creates a layered structure of meaning, where differences are preserved rather than flattened.

Crucially, shared meaning does not require full consensus. Instead, it emerges through:

Contextualisation: making the origin and assumptions of a contribution explicit;

Traceability: preserving links between interpretations and their sources;

Negotiation: allowing multiple perspectives to coexist and interact over time.

This perspective also suggests a path for extrapolation beyond the conference. If the conference is treated as a bounded experiment, then the same dynamics can scale to larger, persistent networks. The micro-network becomes a prototype for a broader ecosystem in which:

communities retain autonomy over their conceptual structures;

interoperability is achieved through shared primitives rather than enforced uniformity;

meaning continues to evolve through ongoing interaction rather than centralised control.

In this sense, the conference is not only a venue for presenting finished work, but a live environment for the co-construction of meaning across communities—a small-scale instance of the kind of distributed knowledge network that systems like Seed aim to support.

Within each of these communities, the role of the Knowledge Organiser (KO) becomes essential. This role, often informal and emergent rather than institutional, supports the coordination of knowledge. This is done by connecting ideas, maintaining coherence, and reducing friction in how information is accessed and interpreted. Prior work in knowledge management, such as that of Thomas H. Davenport and Laurence Prusak [6], has identified similar functions under the notion of knowledge managers or stewards. In hypermedia environments, this role is augmented, but not replaced, by AI systems which can assist in organising, linking, and retrieving information at scale. However, the core epistemic functions of interpretation, judgment, and contextualisation remain irreducibly human, positioning the KO as a critical actor in sustaining shared meaning.

Decentralisation fundamentally redistributes epistemic labour. In centralised platforms such as Meta or LinkedIn, much of the organisational work of ranking, linking, surfacing, and contextualising content is performed by the platform itself, embedding its own incentives and priorities into the structure of knowledge. In contrast, a hypermedia network with no central gravity requires each community to assume responsibility for this work. While this increases the cognitive and organisational demands placed on participants, it also realigns incentives: communities gain ownership over their knowledge processes and are no longer constrained by external agendas that may not prioritise human understanding or collective intelligence. In this sense, epistemic labour becomes both more visible and more necessary, forming the foundation upon which decentralised knowledge systems can function.

Recent advances in artificial intelligence are accelerating multiple stages of knowledge work. Tasks that once required substantial effort–writing, coding, organising information, training models, and even bootstrapping systems–can now be performed with unprecedented speed and scale. This acceleration increasingly shifts the primary bottleneck toward the construction and maintenance of shared meaning within intellectual communities, where knowledge must still be interpreted, discussed, and negotiated to become meaningful and actionable. At the current state of the art, this process continues to rely on the human element as a form of “sense check”, ensuring that what is produced remains coherent, relevant, and grounded.

In this context, AI may indirectly accelerate the development of hypertext systems as the primary means for human augmentation. If meaning cannot be automated, it must be collectively constructed. Hypertext provides the substrate through which intellectual communities can externalise, connect, and evolve their knowledge, enabling them to think together, preserve collective memory, and engage more effectively with increasingly complex problems.

A central question for the experiment is why members of the Academic Hypertext Community would choose to use and contribute to the system. With Seed, participants can create their own authoring spaces while also participating in a shared conference space containing the community's papers and discussions. The experiment combines two complementary mechanisms for sustaining participation: a social process that connects the system to members’ interests and relationships, and a knowledge process that makes participation useful for understanding, representing, and discussing the community's accumulated knowledge.

The social incentive is supported by knowledge organisers who remain attentive to members’ practices, interests, identities, and expected forms of value. By understanding who participates, what problems and aspirations matter to them, and why different members choose to engage, knowledge organisers can connect participants, initiate relevant conversations, and direct attention toward contributions that are meaningful to particular members [14, 27].

The knowledge incentive comes from making the accumulated research of the community more useful to its members. Competency questions will represent questions that participants want the shared knowledge base to answer, while ontologies will define the entities, attributes, and relationships required to answer them [10]. Crucially, these structures will remain directly connected to the academic papers and authors from which the knowledge is derived. Participants therefore have an incentive to ensure that their own work is correctly represented, attributed, and connected to related research. They can refine terminology, contribute their own knowledge through their Seed nodes, and participate in conversations when definitions are disputed. As the ontology improves, it should help answer increasingly useful questions about the field, creating a reciprocal incentive: participants contribute to the shared knowledge structure because it improves a resource that they themselves can use.

Finally, developers participate in the same knowledge environment as practitioners and knowledge organisers, so requirements, usability problems, and missing capabilities that emerge through actual use can become objects of discussion and development. This creates a tighter feedback loop between use and development, supporting faster iteration and adoption through a lean development process.