The Membrane

Knowledge, power and human agency in the age of AI

LowRes.AI · October 2026

The Membrane is an emerging principle for governing the relationship between technological capability and human knowledge systems.

AI changes what knowledge can become

The book made knowledge portable.

Digitalisation made knowledge copyable.

AI makes knowledge operational.

This is a shorthand for changes in scale, not a claim that knowledge could not travel or support action before these technologies existed. The important shift is that codified knowledge can increasingly be turned into capabilities such as translation, synthetic speech, classification, explanation, recommendation and decision support.

Knowledge is no longer only something we consult. Technological systems can draw on it to perform tasks.

This creates extraordinary possibilities. It also changes the relationship between knowledge and power.

AI industrialises the conversion of codified knowledge into scalable technical capability.

Yet the infrastructure and authority needed to control that conversion are profoundly unequal. Organisations with access to computing resources, capital, models, engineering expertise and institutional power can derive capabilities from knowledge at scales unavailable to many communities and local institutions.

The resulting asymmetry is not simply about who can access information. It is about who can determine what knowledge may become, and who controls what is derived from it.

Two logics: extractive and relational

Many technological and economic systems operate through a logic of extraction.

Something embedded in relationships is separated from its context, converted into a resource and made available for use elsewhere. Value may be captured by actors who have little continuing responsibility towards the people or environments from which it originated.

We call this an extractive logic.

A different orientation understands knowledge, resources and technologies through the relationships, contexts and responsibilities that give them meaning. We call this a relational logic.

These are not two kinds of societies, nor shorthand for Western and Indigenous cultures. Both logics can be found within communities, institutions, economies and technological systems. They may coexist and compete.

Nor is abstraction itself extraction. Documentation, comparison and exchange are essential to education, science and shared progress.

The problem arises when separating knowledge from its context also transfers control, authority or benefit away from those with legitimate relationships to it.

Extraction begins when detachment becomes dispossession.

A relational approach asks not only What is this knowledge, and what can be done with it? It also asks Who is connected to it? Who has authority? What responsibilities accompany its use?

AI makes these questions newly consequential because it can transform what it encounters, not merely store or transmit it.

Abstracted and situated knowledge

Not all knowledge can be understood independently of the conditions in which it exists.

Some knowledge can be abstracted, documented and transferred with relatively little loss of meaning. A technical specification or mathematical formula, for example, can remain useful far beyond its context of origin.

Other knowledge derives part of its meaning from particular relationships, places, histories, practices and responsibilities. We refer to this as situated knowledge.

An observation about a forest may be inseparable from sustained engagement with a particular territory. A medicinal practice may involve responsibilities concerning who can teach, interpret or apply it. An account of environmental change may depend on lived experience across generations.

This does not make situated knowledge less rigorous. Nor does it mean that scientific knowledge is somehow context-free: scientific knowledge, too, is produced through situated practices and institutions.

The distinction is not a rigid classification of different peoples' knowledge. It is a way of asking what happens to meaning, context and authority when knowledge moves.

Information can be detached from context. Knowledge cannot always be detached without being changed.

An AI system may retrieve, translate, summarise or recombine locally held knowledge while obscuring the relationships that govern its use. Even when the source remains identifiable, a new output may exceed the authority originally granted to use the source.

Provenance matters. But provenance alone is not governance.

The conversion asymmetry

The central AI asymmetry is not that some people possess knowledge while others possess capability. All communities have their own knowledge and capabilities.

The asymmetry lies in the vastly unequal power to convert codified knowledge into scalable technological capability, and to control what is derived from it.

This differs from the familiar digital divide.

The digital divide asks who can access technology. The conversion asymmetry asks who can shape what technology does with knowledge, under whose authority, and for whose benefit.

Access can expand without redistributing meaningful control.

A local institution may use advanced AI tools while remaining dependent on external providers for the models, infrastructure and conditions of use. A language community may contribute material to an AI initiative without acquiring authority over resulting applications. A public organisation may receive impressive outputs without being able to inspect their sources or govern sensitive inputs.

The issue need not be malicious intent. It may be built into the structure of the relationship.

The question is therefore no longer only:

Who controls the data?

It is also:

Who controls what the data may become?

Governance must address both flows, what may move across a boundary, and transformations, what may be made from what crosses it.

Knowledge extraction predates digital technology. AI changes its scale, speed, persistence and potential consequences. This makes the governance of derivation, not just access, increasingly important.

Access is not agency

Expanding access to AI matters. But access is not the same as agency over technology.

An organisation can use an AI system without controlling its assumptions or dependencies. A community can benefit from outputs without having authority over the knowledge used to generate them. Participation may expand while decision-making power remains elsewhere.

Inclusion without control can reproduce dependency.

The more fundamental objective is technological agency: the practical ability of people and institutions to exercise self-determination in relation to technology.

That ability includes understanding choices, setting conditions, refusing participation, changing how systems are used, and retaining enough capability to make decisions effective.

Three connected capacities matter:

Awareness creates the ability to recognise choices.

Governance creates the legitimate authority to make them.

Capability creates the practical power to act on them.

The right to refuse is essential. Not all knowledge must be digitised. Not all language data should become training material. Not all knowledge should be made machine-readable, and not every technology needs to be adopted.

But refusal alone may leave existing capability asymmetries intact.

The right to refuse extraction must be matched by the capability to use technology on one's own terms.

That is the problem The Membrane seeks to address.

The Membrane: governed permeability

The Membrane is a human-governed boundary between external technological capability and locally governed knowledge, enabling governed permeability without surrendering authority.

Its central principle is simple:

Capability may flow in without knowledge having to flow out.

The Membrane is neither a wall nor unconditional openness. It enables governed permeability: different rules for what may enter a knowledge environment, what may happen inside it, and what may leave.

What may enter? Trusted external knowledge, models and technical capability can become available according to local needs and decisions. Access to that capability does not create an automatic claim on locally governed knowledge.

What may happen inside? AI may retrieve, compare, explain, translate and help develop communications, but the right to access information does not automatically grant unlimited authority to transform or repurpose it.

What may leave? The people and institutions with legitimate authority determine what may be communicated, exported, shared or published, including when outputs are derived from restricted sources.

The boundary is deliberately asymmetric. External capability may enter without requiring an equivalent transfer of locally governed source knowledge in return.

Access to capability does not create a claim on knowledge.

The objective is not to make all knowledge systems identical or universally available. It is to reduce inequalities in access to useful technological capability while respecting differences in knowledge, authority and responsibility.

This is why human governance is indispensable. Legitimate authority may rest with knowledge holders, institutions, community organisations or other recognised decision-making arrangements. It cannot simply be assumed from possession of a device or declared by its designers. Nor is a community necessarily a single, undivided authority.

The Membrane enables exchange without making extraction the price of participation.

From Human-in-the-Loop to AI-in-the-Loop

Many approaches to AI governance emphasise keeping humans involved in AI-driven processes. Human-in-the-Loop practices can provide meaningful oversight, review and control.

But the framing often starts from an AI process and asks where people should intervene.

The Membrane starts from the human system.

People and institutions already have ways of creating knowledge, interpreting evidence, deliberating, communicating and deciding. The question becomes:

Where should AI be permitted to operate within a human process?

We call this orientation AI-in-the-Loop.

AI-in-the-Loop places technological capability inside a human-governed process rather than treating human participation as a safeguard added to a system whose purposes have already been set.

Humans hold authority. AI participates.

This matters when different knowledge systems encounter one another. Scientific evidence, institutional information, community experience and Indigenous knowledge may offer distinct, sometimes complementary and sometimes conflicting understandings.

AI can help retrieve sources, compare accounts, identify uncertainty, develop questions and prepare communication. It should not gain authority to collapse these differences into one supposedly definitive answer.

The Latin American tradition of diálogo de saberes, a dialogue among knowledge systems, offers a practical orientation. Its purpose is not necessarily to merge knowledge into a single hierarchy, but to support encounters in which different forms of knowledge can inform understanding and action without erasing their distinct contexts.

AI does not host the dialogue. It enters the dialogue.

From principle to system design

The Membrane is intended to inform real systems, not remain a metaphor.

A conventional technology project may begin by choosing an AI capability, deploying it, and adding safeguards once risks become apparent.

The Membrane suggests a different sequence:

Human purpose → legitimate authority → governed boundaries → appropriate technological capability → human-directed action

Some boundaries can be supported technically, through local processing, separated knowledge environments, restricted connectivity, visible source information and controlled exports.

Others require social and institutional processes: informed consent, cultural protocols, deliberation, contested-authority resolution, oversight and the possibility of withdrawing permission.

Neither physical isolation nor technical permission settings are sufficient by themselves. A locally operated system can still be used extractively if its governance is illegitimate or its outputs ignore restrictions on the knowledge from which they were derived.

Technology may enforce the rules of The Membrane. It must not determine the legitimacy of those rules.

Different communities and institutions will need different boundaries. The principle is general; legitimate governance must be contextual.

Intellectual foundations

The Membrane is an emerging synthesis, not a claim to have invented situated knowledge, Indigenous self-determination, relational thinking or appropriate technology. It draws on distinct bodies of scholarship and practice that illuminate different aspects of the challenge. None of these sources should be read as an endorsement of The Membrane itself.

Situated knowledge and responsibility

Donna Haraway's 1988 essay, Situated Knowledges: The Science Question in Feminism and the Privilege of Partial Perspective, challenges the idea of a view from nowhere and emphasises situated perspectives and accountability in knowledge production. It informs our caution against treating supposedly context-free information processing as neutral.

Decolonising methodologies

Linda Tuhiwai Smith's Decolonizing Methodologies: Research and Indigenous Peoples examines how knowledge production has been implicated in colonial power and argues for Indigenous authority in defining research priorities and methods. It informs the distinction between being invited to participate and having meaningful power to shape the terms of participation.

Indigenous Data Sovereignty and CARE

The CARE Principles for Indigenous Data Governance articulate Collective Benefit, Authority to Control, Responsibility and Ethics. They offer an essential counterpoint to the assumption that making data accessible or reusable is sufficient for responsible stewardship. The Membrane does not claim to implement CARE; it treats questions of collective authority and benefit as central to the design problem.

Local Contexts and practical knowledge governance

Local Contexts' Traditional Knowledge and Biocultural Labels provide concrete mechanisms through which Indigenous communities can communicate provenance, protocols and expectations for future use of knowledge and data. They show how governance relationships can be made visible in information infrastructure, while remaining dependent on community-defined decision-making.

Data colonialism and extractive digital systems

Nick Couldry and Ulises A. Mejias, in The Costs of Connection, analyse how digital systems can appropriate aspects of human life through datafication and concentrate economic power. The Membrane extends the concern toward AI's capacity to derive operational tools, outputs and services from codified knowledge. That extension is our proposed argument, not a claim made on behalf of these authors.

Epistemic plurality and diálogo de saberes

Maya Lorena Pérez Ruiz and Arturo Argueta Villamar's Saberes indígenas y diálogo intercultural examines the possibility of dialogue between Indigenous and scientific knowledge without automatically imposing scientific standards as the sole basis of validation. It is a particularly relevant reference for the question of how AI might support, rather than govern, such encounters.

Appropriate technology and human-scale systems

The tradition of appropriate technology, associated with E. F. Schumacher and the organisation now called Practical Action, emphasises technologies shaped by local conditions, capacities and purposes rather than technological scale as an end in itself. This is an important practical antecedent for designing AI systems around their operating environments and local agency.

Taken together, these traditions do not amount to a single theory. They help articulate different facets of a shared challenge: technological capability should not require the transfer of knowledge authority away from those who legitimately hold it.

The Membrane's proposed contribution is to express that challenge as an operational design principle for AI: govern both the movement of knowledge and the transformations that become possible through it.

Making The Membrane operational: AI Capsules

A governing principle becomes meaningful when it can shape the design of real systems.

For LowRes.AI, this means asking a practical question:

What would an AI system look like if human authority over knowledge were a foundational design requirement, rather than a safeguard added afterwards?

Our first proposed implementation is the AI Capsule.

An AI Capsule is a self-contained, locally controlled AI system combining trusted knowledge libraries, local AI capability and human-governed workflows. Rather than requiring people and institutions to send their knowledge to external AI infrastructure, it brings useful technological capability into an environment they can govern.

The Membrane supplies its governing principle. Trusted external knowledge can enter. Locally governed source knowledge stays local by default. AI may help retrieve, compare, interpret and prepare communication, while legitimately authorised people determine which transformations and outputs may be shared or published.

An AI Capsule can operate without continuous connectivity or a cloud AI dependency. But offline operation is not its defining purpose. The defining idea is technological capability within a locally governed knowledge environment.

We are developing AI Capsule 1.0 as an early reference implementation using existing open technologies, portable knowledge resources and deliberately simple boundaries. It is a prototype under development, not an available product or an established deployment model.

The purpose is to test whether the principle can be made technically viable, practically useful and adaptable through real co-development, and to document where it succeeds or fails.

We encapsulate AI so that capability can move in without knowledge having to move out.

AI Capsules: from principle to practical architecture →

Toward AI that strengthens human agency

The Membrane begins with an observation about power: AI expands the ability to turn codified knowledge into scalable capability, while the resources and authority to control that transformation remain unequally distributed.

If inclusion means access to systems whose purposes, infrastructure and knowledge practices are determined elsewhere, it can expand capability without strengthening self-determination.

The Membrane proposes another relationship. AI can participate within human-governed knowledge environments. External capability can become available without automatically extracting local knowledge. Authority over meaning, transformation and release can remain with the people and institutions entitled to exercise it.

This is a proposition to be developed and tested, not a claim that technical design has already solved knowledge governance.

The Membrane keeps the human on the side of the subject, not the resource.

The Membrane is an emerging conceptual and systems-design framework developed within LowRes.AI, an initiative of Audiopedia Foundation. This text is a working public formulation intended to evolve through research, technical implementation and collaboration with knowledge holders, communities and institutions.