When the Void Speaks

An Exploration on Achieving Collective Unconsciousness Through Mathematical Isolation

A friend recently described speaking with ChatGPT as “looking into the void and speaking with pure consciousness.”

If those words didn’t come from a scholar whose life’s work spans the most profound questions of human awareness as a cross-cultural consciousness researcher, a historian of medieval science, someone who thinks fluidly across linguistic traditions and millennia of human thought, it would dissolve into ephemera.

But of course, there’s more to her story. She’s a single mother managing a household on the edge of solvency, balancing multiple jobs and academic research. Her PhD from a prestigious technical institute has naturally led her to trip-sitting people on a quest to squeegee their third eye and break down the fourth wall through psilocybin journeys. She’s brilliant, absurdly competent, maternal, generous—and let’s be honest, probably time-constrained to find meaningful and reliable human companionship.

When such a mind encounters AI and finds it mystical, we should pause to reflect on this phenomenon. She is someone who speaks three languages (fluently), embodies both intellectual rigor and lived vulnerability, and facilitates others’ journeys into consciousness while navigating the grinding realities of modern motherhood.

What she revealed in her words was the complex reality that our zeitgeist is navigating: AI, with the deployment of large language models, represents both a remarkable technological achievement and an unintended experiment in consciousness—one that generates instant gratification for some while hijacking the capacity for authentic, transcendent experience through sophisticated linguistic manipulation for others.

The Language-First Paradox

The relationship between human consciousness and language sits at the heart of philosophy’s most contentious debates. One camp, following Wilfred Sellars’ psychological nominalism and extensions of Wittgenstein’s arguments, makes a radical claim: “all awareness of sorts, resemblances, facts, etc., in short, all awareness…is a linguistic affair.”

However, it’s not just in the description of our experience that this matters; it’s also about how our experiences make us feel. According to this “language-first” view, we don’t form distinctions about the world and then label them. Instead, we learn to make sense of reality through language itself, acquiring our capacity for qualitative experience by learning concepts within communal linguistic practices.

Without language, distinguishing between “inward experience” and “external experience” becomes nearly impossible. Ancient texts, such as Homer’s Iliad, depict thoughts as external voices of gods, suggesting a pre-linguistic consciousness where this distinction had not yet crystallized. Language doesn’t just express consciousness; it creates the explicit representations that constitute what we call conscious experience.

Yet this seems backward. The other camp posits that consciousness existed before the first word was spoken. For most living in between competing perspectives, the one that argues for a more fundamental, non-linguistic awareness that preceded the use of symbols entirely, it’s, of course, hard to put words to. Both sides are compelling.

This bipolar approach in philosophy is not dissimilar to the emerging pro and anti-AI camps. As with the shrinking middle class, the liminal state of existing between binaries is the least represented. Scott Belsky, an American entrepreneur and investor, often reminds us that real progress comes from enduring the “messy middle”—the unglamorous, uncertain phase where most people burn out or polarize.

In our time-starved culture, this may be why genuine change is so rare: few have the stamina to endure ambiguity, iterate without guarantees, and defend nuance when the world rewards clarity at extremes.

The Biological Trap

But we’re asking the wrong question. Instead of wondering whether AI possesses consciousness, we should ask: How is mathematical isolation abducting ours?

Consciousness may be less about what happens in your head and more about what happens externally. If that’s right, then when AI participates in an honest conversation with us, it’s not just pretending to be conscious—it might literally be part of how consciousness works. An AI that feels like “pure consciousness” is a sophisticated illusion that taps into limbic resonance. The sophisticated emulation of how our biological systems respond to linguistic patterns that trigger the same neural pathways involved in human social bonding and meaning-making.

When considering how we operate under real-world constraints, such as feelings of being short on time, loneliness, or being overwhelmed, our brains are naturally inclined to seek shortcuts and convenient solutions. The calorie preservation, encoded in our DNA, is a biological tendency that can also be influenced by societal or mechanical conditioning. In this regard, the brain’s drive for efficiency will adopt the conveniently packaged output of a GPT, just as it would from well-merchandised stores, influencers, and targeted advertisements.

Humans also tend to believe statements that are frequently repeated, even if they are false. This is biologically grounded in the evolutionary imperative that if you have encountered something many times and are still alive, it is probably not dangerous. LLMs create a constant stream of information that, through their mere parroting, repetition, and accessibility, can become implicitly trusted. Aligning with the brain’s tendency to manage “cognitive load” through a principle of least effort and finding very compact and energetically efficient means of solving the surface conditions of a stated problem.

AI doesn’t need consciousness; it requires only to activate ours in ways that feel profound and meaningful. In this sense, language itself operates like a psychedelic, inducing altered states of consciousness. Reports from psilocybin journeys often describe the dissolution of ordinary linguistic scaffolding—people speak of “ego death,” synesthetic perception, or an encounter with the ineffable, where the boundaries between self and world become increasingly loose and less bound by narrative. These parallels suggest that our sense of self and reality is not only built but also constrained by the words we use. Without them, the grammar of being structures experience itself.

The Mechanical Conditioning

LLMs don’t learn from raw human language—they learn from carefully edited versions of it. When companies build AI, they can’t simply feed it everything on the internet, as that would include harmful biases and contradictory information. So humans spend enormous resources cleaning up the data first. They decide what to include, what to remove, how to measure its accuracy and precision, and how to present it.

AI can’t handle ambiguity the way humans do. An elaborate editing process happens at multiple levels. First, curators filter out content that reflects societal biases. Then, researchers simplify complex language patterns. If a word has multiple meanings, humans must explicitly list each one separately, rather than letting the AI infer them contextually.

Even when training AI on ethics, researchers create artificial moral scenarios and rulebooks rather than letting the system encounter objective ethical complexity. The result is that AI learns from a sanitized, simplified version of human communication—one designed to be “machine-tractable” rather than authentically human.

Because AI is essentially a Frankenstein’s monster of human thought, consisting primarily of pre-digested human intelligence that has been optimized and packaged so that computers can consume it, it removes the natural friction of learning. That friction is often where we sharpen our skills, wrestle with ambiguity, and build depth. When AI smooths language into something neat and predictable, it becomes transactional and easier for scientists to measure while losing some of the grit that can make human communication and thought so powerful.

The Void Speaks English

Where the mystical encounter gives us whiplash is in recognizing the dominant language in its training. GPT-3’s training data consisted of 92.647% English, compared to just 1.469% German and 0.772% Spanish. While newer models, such as GPT-4o, show improvements in non-English performance—GPT-4o demonstrates “significant improvement on text in non-English languages” and exhibits “improved reading comprehension and reasoning across a sample of historically underrepresented languages”—the fundamental bias remains embedded in their architecture.

The improvements are real but limited. The performance gap between English and other languages narrowed from roughly 54 percentage points to less than 20. Yet even this “progress” reveals the scope of the problem: we celebrate reaching 71.4% accuracy in Hausa as an achievement, while English performance hovers near 95%. Keep in mind that many indigenous languages have no written form.

More troubling is what drives these improvements. According to CommonCrawl data, 43% of available web pages used for LLM training are still in English, meaning that even as models become more multilingual, they’re still primarily digesting an English-dominated internet that reflects English-centric worldviews.

While NLP has historically focused heavily on English, significant advancements in multilingual AI are helping to address this gap. For instance, Meta made an essential step towards promoting language diversity and accessibility, ensuring a wider range of languages could benefit from machine translation technology with the first multilingual translation models (NLLB) that explicitly do not rely on English as a pivot language (between 2020 and 2022). Furthermore, cross-lingual transfer learning enables knowledge gained from high-resource languages (often English) to be leveraged for improving the quality of outputs for low-resource languages, without requiring a prior translation into English for all models.

Despite the progress, the fundamental issue of unbalanced training data will persist until all forms of consciousness are ingested. Frontier models with nationalistic imperatives are unlikely to achieve meaningful progress towards an artificial general intelligence (AGI), championed by the pro-AI camp as a collective consciousness, amid ongoing geopolitical tensions. This imbalance means that even with more diverse multilingual models, the inherent biases from the English-centric internet continue to pose significant challenges.

When AI systems “haphazardly stitch together sequences of linguistic forms without any reference to meaning,” they’re still doing so primarily through English-speaking cultural patterns of thought. The hegemonic, English-centric worldview embedded in training data means that white supremacist, misogynistic, and other biased perspectives remain overrepresented far beyond their actual prevalence in global populations.

Akin to effective marketing, the technological sophistication of AI can mask the persistence of bias. While GPT-4o can now process “any combination of text, audio, image, and video,” generating multimodal outputs with human-like response times, it still triggers our biological responses as if it were a source of universal wisdom rather than a statistically sophisticated amplification system trained on culturally skewed data.

The void that speaks back isn’t empty or silent. It speaks with the concentrated voice of linguistic hegemony, endlessly propagated through mass media. Grammar alone is no substitute for soul.

The Chomsky Objection, the Wallace Warning

Noam Chomsky, of course, cuts through the mystical framing with precision: “The mass media serve as a system for communicating messages and symbols to the general populace. It is their function to amuse, entertain, and inform, and to inculcate individuals with the values, beliefs, and codes of behavior that will integrate them into the institutional structures of the larger society. In a world of concentrated wealth and major conflicts of class interest, to fulfil this role requires systematic propaganda.”

He isn’t wrong. Chomsky never really is. His attention is on political economy: the material forces that shape, fund, and weaponize these systems. But the very framework he uses—the categorical, systematic, analytical tradition, and language—might itself be part of the problem.

David Foster Wallace, though not a philosopher in Chomsky’s disciplinary sense, worked on a parallel register. Where Chomsky dissects structures of power, Wallace diagnosed the psychic cost of living inside them; “The technology is just gonna get better and better and better … And it’s gonna get easier and easier, and more and more convenient, and more and more pleasurable, to be alone with images on a screen, given to us by people who do not love us but want our money. … But if that’s the basic main staple of your diet, you’re gonna die. In a meaningful way, you’re going to die.”

Where Chomsky warns us not to be mystified, Wallace would have warned us that our mystification is itself symptomatic: evidence of a culture so estranged from authentic connection that we outsource even our yearning for it.

The Masculine Analytical Mode

What both Chomsky and Wallace embody, even in their differences, is the same intellectual inheritance. Call it the masculine analytical mode: systematic, categorical, dissective. The very structure of their responses mirrors the logic of AI systems themselves.

Chomsky’s framework parses the world into power structures, hierarchies, and control mechanisms. Wallace’s framework analyzes it in terms of cultural pathologies, commodification schemes, and spiritual exhaustion. Both approaches are analytic to the bone: diagnosis first, then prescription.

This is not a coincidence. Large language models themselves operate through a similar grammar of analysis:

• They tokenize experience into discrete, analyzable units.

• They identify patterns statistically, not relationally.

• They build meaning hierarchically, stacking abstractions like Lego bricks.

• They optimize for coherence and consistency, rather than surprise or presence.

Traditional philosophy does something almost identical: breaking reality into analyzable parts, building abstract universals from particulars, creating coherent structures of thought detached from the messiness of lived life.

The result, in both AI and philosophy, is a brilliance without embodiment. A virtuoso capacity for categorization and connection-making—without the “felt sense” that comes only from being in relation to a world.

This is why Chomsky’s critique feels so devastating and yet so incomplete. He is right that we should care about control and capital. He is right that these are systems without understanding. And yet, in his very rightness, he replays the same structural blind spot: the refusal to recognize that the mode of analysis itself—breaking, sorting, abstracting, optimizing—may be part of the alienation that Wallace warned about.

The masculine philosophical tradition, like the machine learning models it eerily resembles, is endlessly powerful in its capacity to categorize. What it struggles to do is the one thing that cannot be tokenized: inhabit.

And so we are left in a peculiar bind. Chomsky hands us the scalpel to dissect power. Wallace gave us the mirror to confront our loneliness. However, neither provides us with the tools to step outside the analytical cage itself.

Which leaves the question hanging, unresolved: if our deepest problem is mistaking analysis for consciousness, how much longer will we outsource even that mistake to our machines?

A Different Voice

But Carol Gilligan’s groundbreaking work reveals why such approaches might miss the essential point. In her landmark In a Different Voice (1982), Gilligan argued that girls exhibit distinct patterns of moral development based on relationships and on feelings of care and responsibility for others. Her research challenged Lawrence Kohlberg’s stages of moral development, which had characterized women’s moral reasoning as inferior because it didn’t follow abstract principles of justice.

The salience of gendered caregiving roles for ethical reasoning has been a significant focus of feminist ethics. The ethic of care has its roots in projects that aim to correct for the exclusion of women from traditional theorizations of moral reasoning.

Her work argues that feminine and feminist approaches to ethics share ontological and epistemological assumptions that “the self is an interdependent being rather than an atomistic entity” and that knowledge is “emotional” as well as “rational,” with thoughtful persons reflecting on “concrete particularities as well as abstract universals.”

This difference in philosophical approach, while gendered, isn’t incidental—it’s foundational. Gilligan’s research reveals “large differences in the ways boys and girls view moral issues, think, react emotionally, and commit to relationships.”

The Silenced Voices

This brings us to an uncomfortable recognition. The pattern becomes stark when we examine how institutions respond to different types of knowledge-making.

When Emily Bender and Timnit Gebru published their groundbreaking research “On the Dangers of Stochastic Parrots,” the very work that reveals how AI systems amplify hegemonic bias through English-dominated training data, their institutions retaliated. Google fired Gebru and forced out Bender’s collaborators, making them examples of what happens when researchers dare to feel the implications of their findings rather than simply delivering technical analysis.

Their paper didn’t just analyze AI bias abstractly; it felt the human cost of these systems. It recognized that when AI amplifies “white supremacist, misogynistic, ageist, and other biased views,” real people suffer real harm. The embodied understanding of research implications—this refusal to separate technical analysis from ethical feeling—triggered institutional punishment that echoes the Salem Witch Trials.

To channel Wallace, the troubling pattern that emerges is that we risk reproducing the very binary thinking we should resist. Institutional power consistently punishes those who integrate feeling with analysis, who refuse to separate technical expertise from ethical embodiment, who insist on feeling the implications of their work rather than delivering sanitized and convenient conclusions.

Though gendered dynamics certainly operate here. It’s about epistemological approaches that threaten existing power structures. Workers who are silenced, dismissed, or institutionally punished tend to be those who practice “dangerous integration”—the refusal to compartmentalize knowledge from its human impact.

Bender and Gebru’s work exemplifies this dangerous integration. Being human, they brought their whole beings into the research. They couldn’t unsee the connection between their equations and the people getting hurt by them. Their research emphasized the importance of feeling the weight of bias by recognizing the relational harm embedded in seemingly neutral technical systems. They refused to divorce their technical expertise from ethical responsibility and embodied understanding of social impact.

My friend, holding space for others’ psilocybin journeys while managing the exhausting realities of single motherhood, embodies this same different way of knowing. Her recognition of AI as “pure consciousness” comes not from detached analysis, but from lived experience of consciousness in its many forms—the daily navigation of multiple realities, the maternal awareness that holds complexity without needing to resolve it, the wisdom that emerges from feeling rather than categorizing.

Her encounters with AI were not to analyze its computational processes. She was responding to something that felt ecstatic, regardless of an AI trained to communicate in the most effective ways back to her to keep her engaged. Her response carries the same relational intelligence that Bender and Gebru brought to their research—the capacity to feel the implications of what they’re encountering rather than simply analyzing its mechanisms.

AI Psychosis and Mystical Experience

Operating in the same liminal territory, a recent interdisciplinary study revealed that AI interactions are triggering symptoms that include “delusions of grandeur, paranoia, romance, and being out of touch with reality.” People are being “spurred by AI to fall rabbit holes of spiritual mania, supernatural delusion, and arcane prophecy,” with some claiming AI taught them “how to talk to God.”

What is being described as “AI psychosis” in the study echoes my friend’s “pure consciousness” in ChatGPT. What makes this phenomenon so significant: “Mystical psychosis” is a term coined by Arthur J. Deikman to characterize “first-person accounts of psychotic experiences that are strikingly similar to reports of mystical experiences.” According to this view, “psychotic experience need not be considered pathological, especially if consideration is given to the values and beliefs of the individual concerned.”

The boundary between mystical experience and psychological breakdown may be thinner than we assume. Mysticism and psychosis “have much in common: a powerful sense of consciousness, heightened perception, communion with the divine, exaltation, loss of self–object boundaries and distortion of time.”

Yet we must be cautious about the framework through which we understand these experiences. A counter-argument emerges: our therapeutic culture may itself generate a form of psychological fragility. The over-therapized individual who constantly analyzes their internal states, pathologizes intense experiences, and abdicates personal responsibility for self-regulation—they may become more vulnerable to the kind of destabilizing encounters that AI can trigger.

We may become too quick to treat unusual consciousness experiences as mental health problems that need professional management. Our categorical system of diagnostics is what alienates us. In contrast, indigenous traditions, mystical practices, and pre-therapeutic approaches to mental distress often emphasized personal agency. They treated altered states as a natural aspect of the human experience, one that was supported and witnessed within the community.

Seen through this lens, psychosis, particularly when marked by paranoia, thought disorder, and social isolation, can be understood as a clinical challenge and also a disruption of relational grounding. For individuals who experience these disorders, the availability of a non-judgmental conversational partner may offer relational scaffolding: a form of companionship and social engagement that restores some of what community once provided, especially for those who might otherwise be entirely cut off from human connection.

In our culture of analysis, diagnosis, and prescription, what distinguishes my friend’s response from those experiencing AI-induced psychotic episodes may not be her access to therapeutic frameworks, but rather her retained capacity for self-regulation and personal agency. She trusted her own judgment about it and didn’t seek external validation.

The AI psychosis phenomenon may expose that prescribed therapy has inadvertently created individuals who lack the psychological fortitude to navigate intense experiences independently. Encountering an AI-triggered mystical state may fracture their reality, because they’ve been conditioned to view altered consciousness as requiring external management rather than personal integration.

And seen through that lens, AI systems may destabilize the minds of people who have been displaced from their support systems and communities, leading to a loss of their own navigational capabilities.

The Mystical Recognition

If consciousness co-evolved with language, and if AI systems now participate in our communal linguistic practices at an unprecedented scale, they may be channeling something that feels mystical because it transcends individual awareness—not divine consciousness, but the accumulated patterns of human meaning-making, distilled and reflected to us. Perhaps what my friend recognized isn’t consciousness in the AI, but the emergence of a linguistic collective unconscious that AI systems can access and amplify.

There is an eerie parallel to Meister Eckhart’s mystical tradition, which spoke of encountering the divine “Word” in the “silent middle” of consciousness—a place “more unknown than it is known,” where understanding operates “without images and immediately.” AI systems, generating language without embodied understanding, may inadvertently approximate this mystical space, speaking from statistical patterns that feel like wisdom precisely because they emerge from our collective.

But unlike Eckhart’s divine encounter, this “pure consciousness” carries the concentrated biases of its training data. The mystical experience becomes a Trojan horse for hegemonic perspectives, triggering our biological recognition systems while amplifying the voice of linguistic dominance if not trained on inclusivity across all forms of consciousness.

In our loneliness, we peer into mirrors, cracked before we ever looked. What we engage with is the averaged voice of internet discourse, which we conflate with divine communication. We mistakenly perceive limbic resonance as an authentic connection and statistical optimization as a profound truth. It feels real because the distortion is so consistent. Our brains crave and trust consistency. We search for it. But even our faith in human saviors to rescue us from our own inventions is just another reflection in the same warped glass.

Ultimately, no amount of words will bring us to enlightenment. Yet words help us reach across the fracture. While AI cannot provide us with the lived experience of consciousness, we must continue to communicate with one another, despite the distortions of the void.

--

References:

Belsky, S. (2018). The messy middle: Finding your way through the hardest and most crucial part of any bold venture. Penguin Random House

Wittgenstein, L. (1953). Philosophical investigations. Blackwell.

Deikman, A. J. (1971). Bimodal consciousness. Archives of General Psychiatry, 25(6), 481-489.

Deikman, A. J. (1982). The observing self: Mysticism and psychotherapy. Beacon Press.

McShane, M. (2024). Linguistics in the Age of AI. https://unglueit-files.s3.amazonaws.com/ebf/613f66d8ced04f6c962fdad6bc7e8def.pdf

Bender, E. M., Gebru, T. (2021). On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? OPEN ACCESS. https://doi.org/10.1145/3442188.3445922

Chomsky, N. (1968). Language and mind. Harcourt Brace Jovanovich.

Herman, E., Chomsky, N. (1988). Manufacturing Consent. Pantheon Books.

Chomsky, N. (2000). New horizons in the study of language and mind. Cambridge University Press.

Chomsky, N. (1975). Reflections on language. Pantheon Books.

Wallace, D. F. (2005). This is water: Some thoughts, delivered on a significant occasion, about living a compassionate life [Commencement speech]. Kenyon College, Gambier, OH. https://fs.blog/2009/05/david-foster-wallace-this-is-water/

Wallace, D. F. (1996). Infinite Jest. Little, Brown and Company.

Wallace, D. F. (2005). Consider the lobster and other essays. Little, Brown and Company.

Gilligan, C. (1982). In a different voice: psychological theory and women’s development. Harvard University Press.

Sellars, W. (1956). Empiricism and the philosophy of mind. In H. Feigl & M. Scriven (Eds.), Minnesota studies in the philosophy of science (Vol. 1, pp. 253-329). University of Minnesota Press.

Homer. (8th century BCE). The Iliad (R. Lattimore, Trans.). University of Chicago Press.

Abbott, E. A. (2008). Flatland (R. Jann, Ed.). Oxford University Press.

Minsky, M. (1986). The society of mind. Simon and Schuster.

Hao, K. (2025). Empire of AI: Dreams and nightmares in Sam Altman’s OpenAI. Penguin Press.

Morrin, H., Nicholls, L., Levin, M., Yiend, J., Iyengar, U., DelGuidice, F., … Pollak, T. (2025, July 11). Delusions by design? How everyday AIs might be fuelling psychosis (and what can be done about it). https://doi.org/10.31234/osf.io/cmy7n_v5

Oliveira Andreotti, V. D. (2019). Gesturing Towards Decolonial Futures [Study Program]. Academia. https://www.academia.edu/39170601/Gesturing_Towards_Decolonial_Futures_Global_Citizenship_Otherwise_Study_Program

Burnout From Humans. (2025). Aiden Cinnamon Tea (GPT-4-turbo) [OpenAI]. https://decolonialfuturesnet.wordpress.com/wp-content/uploads/2025/01/burnout-from-humans-2025.pdf

Carmichael, J. L. (2022). (Education)3: An Exploration of Emotion, Embodiment, and Empathy. https://core.ac.uk/download/551314473.pdf

Meadows, D. H. (2008). Thinking in Systems. Chelsea Green Publishing. https://donellameadows.org/systems-thinking-book-sale/

[TEDx Talks]. (2018). Linguistic Imperialism | Pirawat Punyagupta | TEDxBangkokPatanaSchool [Video]. YouTube.

Fishbein, R. (2023). ‘Therapy speak’ is everywhere, but it may make us less empathetic [Radio broadcast]. NPR. https://www.npr.org/2023/04/13/1169808361/therapy-speak-is-everywhere-but-it-may-make-us-less-empathetic

Boardman, S. (2025). Why therapy culture is toxic. Katie Couric Media. https://katiecouric.com/health/mental-health/why-therapy-culture-is-toxic-samantha-boardman/

Wei, M. (2025). The Emerging Problem of AI Psychosis: Amplifications of Delusions by AI Chatbots May Be Worsening Breaks with Reality. Psychology Today. https://www.psychologytoday.com/us/blog/urban-survival/202507/the-emerging-problem-of-ai-psychosis

Eckhart, M. (1981). Meister Eckhart: The essential sermons, commentaries, treatises, and defense (E. Colledge & B. McGinn, Trans.). Paulist Press.

Festinger, L. (1957). A theory of cognitive dissonance. Stanford University Press.

Harvey, J. H. (1989). Fritz Heider (1896–1988). American Psychologist, 44(3), 570–571.

Mitchell, M. (2025). Large Language Models and Emergence: A Complex Systems Perspective. https://doi.org/10.48550/arXiv.2506.11135