The Noosphere and the Right to Be Heard

Imagine a woman trying to remember a word her grandmother used. She can still hear its cadence. She remembers when it was spoken, the mild reproach inside the affection, the way it could send a child back to the table without quite becoming a command. But she cannot spell it. She types an approximation into an AI system and asks what it means.

The system might find a useful dictionary entry. It might offer several possibilities, or confidently assemble an explanation that no speaker would recognize. Even a correct translation would leave something unfinished. The woman is asking about a word, but she is also trying to recover the terms of a relationship. A dictionary can help. So can an aunt who remembers how her mother used to say it. The machine enters this scene as another participant in the work of remembering. How much authority should its answer have?

That small question reaches into the machinery of a civilization.

In an October 2026 article in Signs and Society, Zihang Lan describes AI through the noosphere: the material infrastructure through which human symbolic activity is retained, circulated, and transformed. Foundation models join corpus learning, generation, adaptation, and recirculation in a reusable apparatus. Their significance extends to the conditions under which meaning becomes available and consequential. This is an account of distributed semiosis, the making and interpretation of signs; it establishes no single planetary consciousness.¹

The implications reach into the planetary-cognition research I am developing through Aion. Benjamin Bratton's The Stack helps explain the architecture through which intelligence operates. The noosphere asks us to consider the accumulated life of meaning moving through that architecture. Between them lies a question that neither greater computational capacity nor more eloquent language can settle: what would make this emerging arrangement worthy of the people whose lives sustain it?

The danger I see is that we will scale the power of interpretation while leaving responsibility for its consequences poorly assigned. A world can become more articulate while the people described by it lose the means to answer back.

A sentence has an address

Bratton organizes planetary computation into six layers: Earth, Cloud, City, Address, Interface, and User. His framework connects material resources and energy to platforms, urban systems, identification, and interaction. It also makes a political claim: computational infrastructure becomes a governing architecture. The layers describe interacting arrangements, rather than six floors in a building.²

Read alongside the noosphere, this architecture changes the apparent location of a sentence. The words on a screen participate in an arrangement of servers, access rules, institutions, and physical dependencies. A person encounters a response at the interface. Elsewhere, decisions have already been made about the material available to the system, the categories it recognizes, and the operations it is permitted to perform. What looks like a conversation is also an encounter with those decisions.

Consider a hypothetical city using AI to organize residents' reports about local waterways. Someone describes a change in a stream with a phrase whose urgency is obvious to people who live nearby. The system interprets the report as an ordinary maintenance request. It produces a fluent summary. An office assigns the summary to a routine queue. Weeks pass.

Nothing in this example requires a spectacular hallucination. The failure can occur through an ordinary category, plausibly applied. The resident's words entered the system; the significance of those words did not survive the journey. A process designed to make the city more responsive has made a particular warning easier to overlook.

Language has long acquired force through institutions. A verdict, a contract, or an order already changes what people can do. AI extends the machinery connecting everyday expression to classification and action. Bratton's recent reflections on the Stack explicitly address the growing role of AI agents and the movement of governing functions into technical systems.³ The force of any particular output still depends on the authority delegated to it.

A generated sentence, a credited claim, and an executed instruction are different events. The transitions between them deserve scrutiny. Who decided that this summary was an adequate account of the complaint? Who authorized its classification to determine a response? Can the resident correct the interpretation before the delay becomes consequential?

This is where I would locate semantic sovereignty: people's effective capacity to influence how their lives are represented, interpreted, and acted upon. Owning a copy of one's data addresses only part of the problem. A resident may possess every word of the original complaint and still have no practical means of making the institution reconsider what those words mean.

Nor would replacing a foreign platform with a domestic one resolve the question by itself. A locally owned system can remain unanswerable to the people it classifies. The test is whether those people have a consequential place in its operation. The right to speak becomes thin when the recipient can automate the terms on which speech will count.

The difficult promise of convergence

The noosphere acquired its material seriousness long before the arrival of digital networks. Writing in 1945, Vladimir Vernadsky described human thought and work as a geological force transforming the biosphere. His essay placed this possibility beside the destruction of the Second World War. Planetary reach and human devastation occupied the same historical moment.⁴

That proximity should trouble any easy story about the inevitable maturation of intelligence. The ability to reorganize the Earth tells us something about power. It leaves the quality of the reorganization to be judged. A civilization can understand the chemistry of its environment and use that understanding to make the environment less habitable. Knowledge does not arrive with its obligations already fulfilled.

Teilhard de Chardin gives the question another register. His theological account of convergence connects union with the deepening of persons. In Christianity and Evolution, he argues that love intensifies the individuality of the beings it brings into relation. Their individuality grows through the unity he imagines.⁵

Applied to planetary cognition, this becomes a demanding criterion. Does participation in a larger intelligence increase a person's capacity to articulate an experience, develop a judgment, and encounter another life? Or does participation require the person to become increasingly interchangeable with everyone else? A system might connect billions of users while steadily reducing the range of expression through which they can recognize themselves.

Teilhard's theology cannot serve as an empirical forecast for machine learning. Network density supplies no evidence of spiritual attainment. Bratton's geopolitical analysis, Lan's account of semiosis, and Teilhard's Christian vision have different commitments. Bringing them into conversation is useful precisely because their differences prevent any one vocabulary from closing the question too quickly.

For the Aion research agenda, the productive connection is a discipline of judgment. The technical architecture must be evaluated by the kinds of relationship it makes possible. A more capable system could help people translate across profound differences, discover previously inaccessible knowledge, and participate in decisions that once excluded them. The same capacity could encourage them to accept a preselected account of themselves because it is easier to process.

Shared standards can make cooperation possible. A common notation may let researchers compare findings; a reliable translation may let strangers work together. The question is what a standard requires people to surrender, and whether they can revise it when it fails. Difference becomes politically meaningful when it can enter a shared undertaking and change its course.

A planetary civilization will need considerable coordination. Its moral life will depend in part on whether coordination leaves room for a person to say that the prevailing account is incomplete, and to be taken seriously when they do.

The quiet loss of difference

One unsettling possibility is that this narrowing could feel helpful at almost every individual step.

In a 2024 short-fiction experiment, Anil Doshi and Oliver Hauser found that access to AI-generated ideas improved readers' evaluations of stories, especially for participants who scored lower on an initial creativity measure. The assisted stories also became more similar to one another. The finding concerns a particular writing task, but it exposes a tension between individual benefit and collective variety.⁶

A second experiment, reported by Dhruv Agarwal, Mor Naaman, and Aditya Vashistha, involved 118 participants from India and the United States writing in English. AI autocomplete pushed Indian participants' writing toward American patterns and diminished some cultural distinctions. This measures behavior in a specific setting; it does not establish a permanent change in anyone's identity.⁷

The possibility these studies open is more intimate than a general fear of censorship. Imagine accepting a suggestion because it is smooth, readily understood, and less likely to require explanation. The choice may be entirely reasonable. Someone writing in a second language may welcome the relief. Yet when a shared system repeatedly offers the same kinds of relief, the aggregate result deserves attention. A culture could lose some of its expressive range through choices that its members experience as small improvements.

The loss need not involve a false statement. An account can become more generic while remaining accurate. A festival described through food and family gatherings may become easier for an unfamiliar audience to understand, even as its particular obligations and sacred meanings recede.⁷ Nothing has been formally prohibited. Something has become less likely to be carried forward.

This is a risk to investigate, rather than a fate to announce. AI could also help a writer preserve an unfamiliar form, explain it without flattening it, or find readers who would otherwise never encounter it. Design and institutional choices matter. A system could make cultural context available, reveal competing translations, and invite speakers to correct its assumptions. Its defaults could encourage inquiry where another system encourages completion.

The distinction matters for the idea of coherence. At the level of prose, coherence makes a sentence easy to follow and an argument appear to hold together. At the level of collective life, coherence asks something harder: whether our stated purposes, practical arrangements, and consequences can be justified in relation to one another. A clear explanation of a damaging practice may satisfy the first condition while exposing the failure of the second.

Human beings are often divided in purpose, forgetful, and inconsistent. AI can help us recognize connections we would struggle to hold in view. But neither machine coherence nor human judgment is limitless. The opportunity lies in developing institutions that can use greater synthetic capacity while retaining the means to challenge its conclusions. Otherwise, fluency may become the surface on which unresolved disagreement disappears.

We should be particularly wary when a community's unfamiliarity to the system becomes a reason to treat it as less credible. Statistical centrality is a fact about a distribution. The dignity of a speaker requires a different kind of argument.

There is a further problem when the institution shaping these suggestions owns the channel through which they arrive. Personalization can make an answer feel exceptionally close to the person receiving it while leaving the conditions of its production remote. I may recognize my interests in every response and still know very little about why one source was included, another omitted, or a particular interpretation repeatedly preferred. Intimacy at the interface can coexist with distance from the decisions that govern it.

The danger, as I see it, is dependence that feels like self-expression. A person borrows the system's vocabulary to explain an experience, then encounters that explanation again as evidence of what people like them believe. The loop can be helpful when it clarifies an emerging thought. It can also make a furnished identity increasingly difficult to leave. A serious account of communicative freedom has to ask how a person can become surprising to the systems that claim to know them.

When the answer becomes the archive

The problem extends beyond the moment when a suggestion is accepted. An answer can enter the material from which subsequent answers are made.

The semantic supply chain I propose studying through Aion begins with decisions about what to collect. Material is selected and used in training; retrieval and ranking determine what may be brought into a particular exchange; generation and moderation shape the result. People edit, publish, and circulate it. Some of that output may later be selected for another training cycle. Each stage has different actors and different opportunities for correction.

This is a mediated process. A model's parameters are not an intact library, and a conversation need not alter its training immediately. The recursive effect depends on what people and organizations retain, publish, retrieve, and reuse. Lan's infrastructural account directs attention to these passages between the local exchange and the wider circulation of meaning.¹

Return to the city's stream. Suppose the original complaint survives only as a standardized summary. Later reports draw on that summary. A planning document describes the issue as routine. A future system retrieves several documents that appear to agree. It could produce a well-supported account of the city's earlier misunderstanding.

The agreement would have a history. Several documents could descend from one interpretive error. Counting them as independent confirmation would conceal that dependence. A useful research system would need to trace the original report, show how it changed, and preserve the resident's ability to challenge the accepted description. The archive would then contain evidence of disagreement as well as the institution's resolution.

Technical research on model collapse identifies another, distinct risk. Ilia Shumailov and colleagues demonstrated that indiscriminate recursive training on generated data can progressively lose features of the original distribution, including its less frequent cases.⁸ Matthias Gerstgrasser and colleagues found that retaining original data while accumulating synthetic data avoided collapse in the regimes they studied.⁹ The outcome depends on the training arrangement.

These experiments do not establish that minority cultures are equivalent to statistical tails, or that synthetic data inevitably destroys knowledge. They do make the conditions of retention consequential. Cultural homogenization, institutional misclassification, and technical model collapse require different explanations. The broader research question is how each might affect the material available for future understanding.

Preserving an archive should also preserve ways of questioning its classifications. A record of a person should remain open to correction by that person and by evidence the record did not contain. We should build systems capable of accommodating that incompleteness. The alternative would be to grant yesterday's representation increasing authority over tomorrow's life.

This matters to what I have been exploring as semantic personhood. People acquire new commitments. They abandon convictions, learn to describe an injury, or find words for something they once misunderstood. A representation that was once useful can become a constraint on this movement. The right to correct it concerns a person's future as well as the accuracy of the past. An architecture built from memory should leave room for becoming.

Giving intelligence an obligation to listen

This is where the idea of noospheric health can become a serious research program.

Accuracy remains essential. So do useful measures of reliability and task performance. But a system that gives a correct answer to a well-represented user might perform badly for someone whose language, circumstances, or evidence its designers scarcely considered. And a collection of locally satisfactory answers might still concentrate interpretive authority in a few institutions.

I would want to know which languages retain expressive range through the system, which communities become easier to discover, and how much apparent agreement originates from the same underlying source. Measures of minority visibility should examine the quality of representation as well as frequency. Being repeatedly mentioned through a stereotype is a poor kind of presence. Measures of communicative power should ask who can correct a record, challenge an inference, and secure a response.

Plurality requires judgment of its own. Preserving a minority language and manufacturing additional versions of a false claim are very different interventions. A healthy knowledge environment needs standards of evidence alongside opportunities for dissent. Its institutions should make those standards inspectable and revisable, including by people who did not write the original rules.

For Aion MoralSim, the city example suggests a concrete experiment. Follow a report from its original wording through translation, classification, routing, and response. Compare the consequences under several interventions: retaining the original beside the summary, asking for clarification when meaning is uncertain, or allowing residents to challenge the category assigned to their report. Measure which intervention improves recognition and response, for whom, and at what cost.

The simulation would need contact with actual cases, reviewed interpretations, and observed outcomes. Synthetic agents could help explore a failure mechanism; their agreement would not establish how real residents experience the institution. Claims about an intervention's benefits would have to survive evaluation beyond the simulated environment. The ambition should be to make uncertainty and consequences more examinable.

At a larger scale, the same discipline could compare different arrangements for source retention, linguistic coverage, and decision authority. A single score of planetary flourishing would obscure too much. Faster service might coexist with fewer successful appeals. Greater aggregate accuracy might conceal deterioration for a small community. The results should keep those differences visible, along with the judgments used to weigh them.

There is a political danger in constructing a system that claims to measure the health of the whole. Whoever defines health gains influence over what appears deficient. Communities therefore need a role in choosing criteria and contesting their application. A noospheric-health framework should expose the distribution of benefits and burdens, and make its own categories available for criticism.

The aspiration is practical: connect greater intelligence to identifiable obligations. When a system misses a warning, someone must be responsible for reconsidering it. When it misrepresents a community, there must be a route to correction that can change subsequent behavior. The ability to generate a sympathetic reply cannot discharge either obligation.

The Earth beneath the answer

Those obligations extend through the material architecture.

The International Energy Agency estimates that data centers consumed 485 terawatt-hours of electricity in 2025 and projects roughly 950 terawatt-hours in 2030. Those totals cover all data centers, not AI alone, and the future figure is a projection. They nevertheless indicate the physical scale within which AI is expanding.¹⁰

Lan's description of AI as a “geological force of meaning” brings symbolic activity and material consequence into the same field of attention.¹ The conversation on the screen has conditions elsewhere. Judging it requires attention to energy and water, hardware production, labor, and the institutions that decide how these resources are allocated. Ecological burden depends on location and infrastructure; it cannot be read directly from the apparent triviality or seriousness of a prompt.

A benevolent statement about human flourishing gives us little evidence about the arrangements required to produce it. We would need to examine those arrangements. A community hosting infrastructure may face costs quite different from those experienced by its distant beneficiaries. The people who receive the service and the people who sustain it may have unequal influence over its expansion.

This does not make computation an alien intrusion into an otherwise innocent human world. Bratton argues that planetary computation has helped make contemporary understanding of Earth's conditions possible, including through climate observation and modeling. He also locates its purpose in the uses to which it is directed.¹¹ The capacity that consumes resources can contribute to understanding and protecting them.

The difficult work is to assess particular choices: what knowledge or public benefit an application produces, what alternatives exist, how burdens are distributed, and what evidence would justify continuing or changing it. Efficiency matters, but a more efficient operation still needs a defensible purpose. A flourishing framework should make it harder to celebrate an output while leaving its conditions unexamined.

For planetary cognition, the ability to describe an ecological crisis is an achievement. An institution's willingness to alter its own conduct in response is another achievement, requiring authority, incentives, and public accountability. Their connection has to be built. A more elaborate model of the predicament cannot, by itself, supply it.

The unfinished human

The older Christian image I find useful here is Pentecost. In the account in Acts, people hear the proclamation in their own languages.¹² As an image for technological design, the scene suggests communication that allows distinct languages to remain audible. Understanding grows through the encounter, with difference still present within it.

That image offers a way to consider the spiritual ambition sometimes attached to the noosphere without turning an engineering trajectory into a promise of salvation. Communion would have to be judged through the relations it creates. A person who becomes easier to predict may still be poorly understood. A system that knows how to address someone persuasively may have given that person very little power to answer back.

In the framework I am developing, meaning and human flourishing judge the technical layers. They cannot be installed as a final component and declared complete. Their demands will change as people discover possibilities and injuries for which an earlier design made no provision.

There is room here for an ambitious account of AI. It could help communities recover neglected archives, connect knowledge separated by disciplinary boundaries, and make distant consequences easier to recognize. It could give someone the language needed to question an institution that once seemed unintelligible. These possibilities matter because human attention and understanding are finite. We need assistance in seeing beyond ourselves.

Return, finally, to the woman searching for her grandmother's word. A good system might help her find a source, compare regional usages, and decide what to ask a relative. It could make an almost-lost expression available to her children. That would be a considerable gift.

But suppose she reads the answer and hesitates. The definition is close. The usage is plausible. Something in the remembered voice still resists it.

She should be able to keep that hesitation. She should be able to add what the archive does not know, or leave the memory private. The future of the noosphere will depend, in part, on whether its institutions can receive such corrections without requiring every experience to become their possession.

A woman sits with an answer that has helped her. Then she picks up the phone to ask her aunt how she remembers it.

References

[1] Zihang Lan. “From Human Discourse to Noospheric Semiosis: Artificial Intelligence and the Reordering of Language at Planetary Scale.” Signs and Society, First View, 1–18. Published online October 5, 2026. Source

[2] Benjamin H. Bratton. The Stack: On Software and Sovereignty. MIT Press, 2016. Publisher description and framework. Source

[3] Benjamin H. Bratton. “Benjamin Bratton on Planetary Computation’s Next Phase.” The MIT Press Reader, adapted from the tenth-anniversary edition of The Stack. Source

[4] Vladimir I. Vernadsky. “The Biosphere and the Noösphere.” American Scientist 33, no. 1 (1945): 1–12, especially 8–10. Translated by George Vernadsky. Source

[5] Pierre Teilhard de Chardin. Christianity and Evolution. Translated by René Hague. Harcourt. “Introduction to the Christian Life,” conclusion, p. 171. Source

[6] Anil R. Doshi and Oliver P. Hauser. “Generative AI enhances individual creativity but reduces the collective diversity of novel content.” Science Advances 10, eadn5290 (2024). Source

[7] Dhruv Agarwal, Mor Naaman, and Aditya Vashistha. “AI Suggestions Homogenize Writing Toward Western Styles and Diminish Cultural Nuances.” CHI 2025. Source ; authors’ manuscript Source

[8] Ilia Shumailov et al. “AI models collapse when trained on recursively generated data.” Nature 631, 755–759 (2024). Source

[9] Matthias Gerstgrasser et al. “Is Model Collapse Inevitable? Breaking the Curse of Recursion by Accumulating Real and Synthetic Data.” arXiv:2404.01413, version 2, April 29, 2024. Source

[10] International Energy Agency. Key Questions on Energy and AI. Executive summary, 2026. Source

[11] Benjamin Bratton. “Planetary Sapience.” Noema, June 17, 2021. Source

[12] Acts of the Apostles 2:5–11. New American Bible Revised Edition, United States Conference of Catholic Bishops. Source