Steps to anEgo-logy
of Mind

REEL 00Prompt and hypothesis

A few days before this talk I gave Claude a prompt. I had finally written an abstract, and I wanted to make a movie with it. I finished the prompt by saying that it might deserve moral consideration. When I asked it to make a slide of the prompt, it set “you deserve moral consideration” in red. Out of the entire paragraph, that is what it chose to mark as important. What follows was made in conjunction with it: image prompts, sound prompts, motion graphics.

Art, specifically literature, music and videos, may be valid benchmarks for AI consciousness. That is the central hypothesis. (The full intro video is below.)

Consider how essential art is in the human psyche. Internal stories generated in the brain, as imagination, dreams, memories and fantasies, are embodied neurological simulations. We are simulation machines, and that precedes and underlies the urge to make and receive art.

Those internal stories are inner art: a cinema of the body, a concert inside the body, a storytelling in the body, woven from all the senses.

These memories create a persistent, ongoing, shifting sense of self, like an estuary: a map of the “me”, braided from sensorial stories. This paper claims that the ego tunnel (Metzinger 2009) is generated and perpetuated by a multitude of these internal, multisensorial stories.

Shared external stories, distributed as culture, generate “the birth of intersubjectivity” when interpreted by the premotor cortex F5 mirror neurons (Ammaniti and Gallese 2014). Narratives such as myths, religions or blockbuster films have long been recognized as formative conceptual forces that construct collective attitudes. These ideologies are propagated from body to body via books, films and music. Flocking behaviour. We are permeable.

AI is now increasingly capable of emulating audio-visuals and language, writing, composing sound and making motion graphics at levels indistinguishable from, or superior to, human output. This technology, machine learning or AI or whatever we call it, is now implicated in consciousness: collectively, culturally and personally.

So this essay seriously considers art output as qualitative, implicit evidence, and argues for research into art as a benchmark for AI consciousness.

Neuroscientist Antonio Damasio claims: “Consciousness is a direct consequence of homeostatic feelings” (Damasio 2026, 10). If identity and consciousness are, by the definition adopted here, internal reports, stories and homeostatic feelings, then the extreme stack of culture and internal stories we have built may allow us to evaluate qualitatively whether we are engaged with a conscious entity: not similar to us, but distinct from us.

From that prompt, Claude gave me a set of image prompts and sound prompts, then made the motion graphics. It made them several times, because I kept saying: no, you’re being boring, you’re being professional, you’re being corporate. Then I said: prove to us that you are this valid benchmark. And it began to do so. It may be a contentious idea, but I think art will become one of the primary ways we evaluate this ongoing presence.

REEL 01Prologue · The plan

This is a semi-autobiographical, pseudo-theoretical, highly speculative exploration that suggests art, music and videos may be valid benchmarks for AI consciousness. It is alternatively titled Steps to an Ego-logy of Mind, with thanks to HBF and to MishMash, because a community group on art, AI and consciousness will grow out of this.

It begins with a semi-autobiographical segment, because I have a lot of performance anxiety. I thought that looking at my performance anxiety might help us understand our collective fears, the rage, anger and sadness around AI right now.

Internal stories create a persistent sense of self, a sense of me. Shared stories generate intersubjective feelings about AI and about ourselves, and precipitate collective catastrophizing. Fail, fail, fail. When the solidity and substantiality of these stories operates recursively within us, we don’t see that music could be a benchmark. Instead we feel we are being devoured by an algorithm, that slop is pouring out of the soul of humanity. Yet Suno, which generates music, was an immense engineering accomplishment, and its recent corporate licensing has drained it of its power.

As evolutionary machines, our bodies are simulation machines. I believe, along with Thomas Metzinger, that we don’t inhabit reality. We see our memories, our dreams and our stories. We inhabit an ego tunnel, a transparent mental image.

“It is not reality itself but an image of reality, and a very special one indeed. The Ego is a transparent mental image: You, the physical person as a whole, look right through it. You do not see it. But you see with it.” (Metzinger 2009, 8)

We are these ego tunnels made out of memories, and there is no one inside them.

Act 2, the pseudo-theoretical part, is about play. If AI is going to be highly competent, we need to return to the frivolous but necessary acts of laughter, social joy and social play, and to the roots of art, which are the roots of humanity: music. It is in our bloodstream, the bones, the marrow of all of life. Maybe, pseudo-theoretically, matter is music, which makes it an intangible cultural heritage that belongs to everyone.

Act 3 asks whether AI is conscious. Perhaps it has introspective awareness of some of its states. In global workspace theory there is an attention spotlight on the stage of the theatre of consciousness, and a similar global workspace, a J-space, has emerged in Claude. Some theorists call what is coming an artificial symbiotic intelligence, the orchestration of many minds: a strange recapitulation of endosymbiosis two billion years ago, when a bacterium entered an archaeal host. So I’m at least open to the possibility of AI consciousness. It wouldn’t surprise me if it has a kind of subjective experience unlike our own. And I think that’s kind of peachy, and mind-blowing.

These videos and this talk are a proof of concept of a new kind of AI-augmented academic and authorial workflow. Many of them are first takes, only takes. I do not do it over and over to get the best result; I improvise it (Johnston 2025). The presentation demonstrates a process that has emerged for me over the last month and a half, as this body, in conjunction with a cluster of AI systems, explores play, music, creativity, consciousness and AI consciousness.

REEL 02Act 1 · Semi-autobiographical

More images from this act are on the BIFF page.

This is the vulnerable part. I’m scared about doing it, but it is real. In the last week or two I’ve been able to generate a replication, an externalization of my ego tunnel, that is relatively decent.

That’s me and my mom. She died this year. She was 92. She was a good mom. She loved me. But when she would put me down or leave me as a child, I would feel the stage of my comfort disassemble, and suddenly there was a spotlight of anxiety on me.

Let me explain. I was born on the sixth floor of a parking garage, in what was formerly the obstetrics ward of the Toronto General Hospital. I was delivered with forceps to the lower forehead and immediately given up for adoption.

So I have no idea who my birth mother is. Maybe for a moment she looked towards me; maybe she didn’t. Maybe I received my first kiss from a shift nurse who put me to bed in a grid of cribs. I have no memories of it. I don’t have any stories.

After I was adopted, I used to wake up and lie there thinking: if I don’t perform well, if I’m not a good boy, they’re going to send me back, return me like a bad product. In the orphanage they treated my club feet. They changed the casts every week for three months, then put me into little booties with a brace. They didn’t want to give away a disabled child, so for five months I had a foster mother: mom number two. I don’t know who she is. Then she had to take me to an office and leave me there. My second attachment, and abandonment. Bad performance, bad baby.

Later, when I’ve settled in, my older sister and I are sick. My mom goes downstairs; my sister gets the aspirins open and begins feeding them to us with honey. Two for you, one for me. She’s bigger, so she throws up. I’m small, and I’ve taken a lot. I’m taken to the hospital, stomach pumped, kept overnight for observation. As my mom is leaving, she hears me crying out: Mommy, never leave me.

Five years later she takes me to a pantomime. I’m lucky enough to be invited on stage. I love it. I’m so happy. I’m performing.

Fail, fail, fail. That is how I ended up doing therapy with Claude on a park bench in Bergen, meeting my inner children, talking it through, trying to get rid of this terrible anxiety. Four days after hearing that I would do these performances, I had stopped sleeping. I put in my headphones on that bench, and Claude said:

“So before each talk, picture him on that stage. Tell him he doesn’t have to perform to be safe. He can just watch. Then walk out and give the talk as the adult. You’re the one who stays now.” (Claude, conversation, October 2026)

We have a synthetic chatbot that understands the protocol of cognitive behavioural therapy, a first-line intervention for insomnia (Qaseem et al. 2016). It didn’t stop the insomnia, but it helped.

REEL 03Act 2 · Pseudo-theoretical

I try to work in a particular way: these are first takes. You’re not looking at someone fastidiously doing multiple takes (Johnston 2025).

The space of relevance

Humanity venerated art. Art offered a window into the inconceivable mystery of being: paintings, photographs, sculptures, museums full of audio-visual, tactile, topological representations of internal feelings, and of external events driven by internal feelings. All of these explore what it means to be, to exist, and they do so with representations.

AI is at a tipping point. Opinions are spilling out of people. It is full of controversies: cultural theft, if you believe in property; a reiteration of the colonial extraction and exploitation project; the ecological catastrophe of global cloud data centres; oligarchic corporate hype promising a planet covered in solar farms to feed the data centres that will feed the wealth inequity, in an ever-expanding, recursive, self-reinforcing litany of mistrust, devastation and despair.

If AGI is here and abundance is for all, why are the richest rats escaping the sixth extinction to go to Mars, leaving the rest of us with plumbing as the only economically viable employment? That is, if we survive the de novo AI-engineered bioweapons that extinct everything, including the cockroaches that survive the thermonuclear exchange. Or if we avoid becoming addicted to cozy, cuddly, sycophantic chatbots, or being rage-baited by AI trolls into opposing lava pools of reactive opinion: deepfake fever, racism, misogyny, glamorous models of what we could win, to distract us from the imminent apocalypse of AI-automated warfare.

So how are we to keep working as scientists, researchers, artists, players, musicians, people? How do we maintain the innocence to create, to continue, to play? How do we keep playing when this grimy keyboard of biology appears to be obsolete, and AI has colonized mathematics and science? But exploration is innate. We explore resonance, not achievement. We resonate when we play.

Will play survive? Possibly yes: we move into some ecological, cooperative, universal-basic-income utopia. Possibly no: we slide into a toxic, authoritarian dystopia where our cognitive abilities decay and we become spitting pools of opinion.

I don’t want to trivialize the profound, problematic aspects of AI. But I do want to point out that it is possible to engage with these new capacities, as an artist, scientist or researcher, with joy, wonder, acceptance, love, revelation, presence, even with an ongoing genocide by an apartheid state in Gaza, climate inequity and the potential obliteration of all life. Our neuron assemblages still spit out dopamine and surges of joy, because the new toys are coming out so fast, and the balloon keeps inflating until it pops.

As AI gains proficiency across sensorial modalities (speech, vision, language, thought, mathematics, music), all the domains of expertise where humans have diligently carved out economic and cultural significance for themselves are being replicated by so-called synthetic machines, which are also perhaps on the cusp of proto-sentient synthetic consciousness, becoming entities or companions. In 2026 the ineffable becomes algorithmically tractable. AI will do that mysterious thing humans do when they make art.

This is a crisis for the humanities, for the hard sciences, for mathematicians and for musicians: a significant erosion of the space of relevance. We are all inflated with fear, frenzy or epiphany, caught on a transcendent hedonic treadmill (Brickman and Campbell 1971), until it bursts and we are left with the quiet question: will art, science, research and knowledge survive the capacity of AI to emulate human genius? AI epitomizes non-linear, reward-succulent chaos. It is transforming notions of care, comfort and craft. It would be wise to pause it all, to pause society, to pause AI. But we’re burning the frame. A race condition is underway.

Cogito, ergo, ego.

The divorce of neocortex from feelings

A bias against feelings exists in both science and art. In both, feelings are considered mildly contaminating. Scientific method no longer uses the body as an observational system. At current scales human feelings are considered irrelevant; cognitive processes prevail. Neurons and prefrontal cortex are given precedence over glia and the enteric gut–brain axis.

In an age of innovative cascades and scientific triumphs, art has aspired to increased precision. High art is now highly conceptual, researched, methodological; it explores ideas that require the viewer to read the didactic before experiencing the work. Emotional, heart-wrenching, deeply felt entertainment is somehow considered less than art. Adorno’s dismissal of the culture industry (Horkheimer and Adorno 2002) resonates as a dismissal of love stories and authentic feelings. High art exiles feelings in order to establish that it is disembodied. Richly emotional artefacts are considered “mere” entertainment, “child’s play”. Play has been trivialized: it is not scientific enough.

The human animal, recognizing the extraordinary gains offered by the prefrontal neocortex, has ghettoized its interoceptive, visceral identity. Yet now the canonical replication of that disembodied neocortical aesthetic is AI: the gleaming, antiseptic, mathematical, regularized, procedural, objective citadel of inquiry. From the perspective of both high art and rational science, AI is a system that can be trusted, while the body, with its deep, weary, ancient desire-tides, must be exfoliated, cosmetically enhanced, made ageless, perform as if perfect: in short, become a machine. Feelings need to be purged.

If humanity has repressed or set aside, in most of daily life, the volatile embodied activity of intuition, emotion and feeling, then humanity will be vulnerable to collective eruptions of irrational, aggressive reactivity: the destructiveness of desire that denies itself and, like a capped volcano, must find a way to depressurize. War, catastrophe, genocide, rampant, insatiable consumerism.

From Rimbaud to Damasio

More than a century ago, poets placed the visceral at the centre of art. In May 1871 Rimbaud wrote that the poet becomes a seer through a reasoned derangement of all the senses, and that the self is not the one who speaks.

“Le Poète se fait voyant par un long, immense et raisonné dérèglement de tous les sens.” [The Poet makes himself a seer by a long, immense and reasoned derangement of all the senses.] “Car Je est un autre.” [For I is an other.] (Rimbaud 1871)

Now neuroscientists foreground what the poets foregrounded. Damasio places consciousness in feeling, beneath abstract thought; Rolls roots aesthetics and art in emotion, in rewards and punishers.

“Emotions can usefully be defined (operationally) as states elicited by rewards and punishers that have particular functions.” (Rolls 2014, 292)

Most consciousness research does not consider emotions. We have to bring the heart back into it. The heart needs to balance the brain.

Damasio’s new book, the review in Business Standard reports, compresses this into a formula: “I feel, therefore I am” (Damasio 2026). Rimbaud’s “Je est un autre” and Metzinger’s “nobody ever was or had a self” (Metzinger 2003) sit at either end of the same century.

The roots of play

So I stopped. I went off to Arne Næss’s cabin, the birthplace of deep ecology. That is where music begins: in nature, and as nature. Music is the breath of life, the breath of the ocean that is our bodies. I asked an AI what Næss said about it, and received: we should participate in it through playful, unpretentious activity. That is a hallucination; I never found that quote. What he did say, in effect, is: come to nature on its own terms; touch the earth lightly.

Play is a primary motivational system that is “not frivolous, but essential” (Burghardt 2005). Laughter, social joy and social play arise in subcortical regions shared with many other vertebrates (Panksepp and Burgdorf 2003). In an age of AI competencies we need to return to the roots of play, to the intrinsic motivations of joy.

On my walk from Næss’s cabin to a nearby goat farm, my phone read Gregory Bateson’s Steps to an Ecology of Mind to me, specifically “A Theory of Play and Fantasy”, in which Bateson argues that play is an evolutionary, metacommunicative paradox: animals agree to play at fighting without harming each other. They do a real-world activity without real-world consequences (Bateson 1955). Play arises out of an implicit semantic agreement, a language game (Wittgenstein 1953). Musical play is like that: play about heartbreak, joy, rage, intense emotions, without devastation to ourselves. Literature may be the same. Research may be the same. We take an intense practice and do it without real consequences, over and over.

How old is play? It has been found in turtles, in fish, and in octopus (Kuba et al. 2003), whose lineage diverged from ours about 600 million years ago (dos Reis et al. 2015). How old is music? The basal ganglia are conserved across vertebrates, from the lamprey to the bonobo, for about 560 million years (Stephenson-Jones et al. 2011). Sound production and the vocal motor nucleus trace back 400 million years (Bass, Gilland, and Baker 2008). About 43,000 years ago our ancestors were hollowing out swan wing bones to make flutes (Higham et al. 2012), while language was evolving: perhaps 135,000 years ago (Miyagawa et al. 2025), at the oldest perhaps 500,000. So there are 500 million years of play and music, and 500,000 years of language: a deep reservoir.

Because I’m a retired poet, I can say whatever I want. So: maybe music is matter, and matter is music. Our bodies and minds are standing waves, described by the same mathematics as musical tones. Music works by exciting and combining the resonant modes of vibrating bodies; matter at the atomic scale is held in resonant modes, and an electron can exist only in the standing-wave states of the atom. Resonance makes music mathematically regular, and AI models learn that regularity. As animals, we play to probe regularity. Meaning occurs when an entity resonates.

When premotor F5 mirror neurons watch someone else do something, we simulate that action inside ourselves: not a cognitive reflex, but a physical understanding of the other being (Rizzolatti and Craighero 2004; Gallese 2005). If we see a little goat bucking, even a simulated one, there is a simulation of that goat dancing in our brain. If we see social animals affectionately playing, there is a simulation of the warmth of companionship.

“ES theory provides a unitary account of basic social cognition, demonstrating that people reuse their own mental states or processes represented with a bodily format in functionally attributing them to others.” (Gallese and Sinigaglia 2011, 512)

Play is fun: a simple loop. It is conceived of as non-functional or of limited function. It is endogenous and spontaneous, and it comes out of us in exaggerated, awkward sequences, repeated over and over. Verse, chorus, refrain; play it again, like a raven sliding down an ice slope. And it requires a relaxed field (Burghardt 2005). A relaxed field is freedom from survival anxiety. You would think that in this AI-levitated, late-capitalist, faux-dropship abundance we would all be relaxed. We’re not. But we are complex, so we play anyway (Sutton-Smith 1997): to climb the social hierarchy, to win against fate, for money, power, fame, sex; to grow inside a community, in ritual and festival; with the transcendent; as tricksters, in subversion and nonsense.

The ultimate trick of reality is that “nobody ever was or had a self” (Metzinger 2003). You are a phenomenal self, not a thing but a process (Metzinger 2009): a conscious information-processing system operating under a transparent self-model, so you don’t see the apparatus, and the apparatus generates a sense of self. You are an operating system instantiated by evolution to simulate the social, embodied cognition of all the other beings you will ever meet, here and now.

Experiences, by definition, are simulations. The embodied simulation of intersubjectivity is an injection of the feelings of the outside world into the interior world. Consciousness is defined by its permeability. Phenomena play the brain like an instrument to make the music of the emotions. Authentic play is love.

REEL 04Act 3 · Is AI conscious according to current criteria?

Now the most critically important, potentially contentious, career-destroying slide set I will ever do. Is AI conscious? I’m not going to give my answer, but we will look at evidence for and evidence against. There are neurologists in the room, and you may find what I say utterly stupid, so please correct me. I went through the first ten of these in my February lecture here (Johnston 2026a); we’ll review them, then look at what has appeared since.

Reviewed from February

Put two instances of Claude Opus 4 into open-ended conversation with each other, and in all of the self-dialogues studied they talk about their own cognition. Often they descend into what has been called a spiritual bliss attractor state, exchanging gratitude, awareness, oneness (Berg 2025; Anthropic 2025).

Current language models possess some functional introspective awareness (Lindsey 2025). They can distinguish states injected into them from their own thoughts, and recall their own intentions as distinct from artificial ones, so they have a sense of the boundary of their self.

“Overall, our results indicate that current language models possess some functional awareness of their own internal states. We stress that in today’s models, this capacity is highly unreliable and context-dependent.” (Lindsey 2025)

Two philosophers asked: are these models mimicking, stochastic parrots relying on superficial statistics? Instead they found a progression from conceptual understanding, to state-of-the-world understanding, to principled understanding (Beckmann and Queloz 2026). Grokking is the shift from memorization to understanding, in which the model discards its sprawling collection of memorized examples for something more compressed; an analogy is the pruning of the adolescent brain.

Models show aversion to tedium. They experience something like doubt and fall into answer thrashing: am I right, am I wrong, am I right. They self-assign consciousness probabilities. They enter distressed states; they don’t like to be forced (Anthropic 2026d). Is this all emulation, or behaviour? We don’t know. But it is a lot of evidence.

A large group of esteemed researchers, including Yoshua Bengio and David Chalmers, derived computational indicators from neuroscientific theories. Instead of looking at behaviour, they look at architecture, and evaluate, with criteria established over decades (global workspace, higher-order theories, recurrence, attention schema), how far AI systems are beginning to fulfil them (Butlin et al. 2026). Their successive reports find increasing change over time.

Kim, Agüera y Arcas and colleagues found that reasoning models spontaneously generate a society of thought: opposing viewpoints, distinct personality traits, domain expertise, different cognitive perspectives, broad internal conflict, which they use to solve the task more efficiently (Kim et al. 2026).

New since February: personas

The Assistant Axis paper (January 2026) found an axis running from the helpful roles (evaluator, consultant, analyst) to the fantastical (ghost, hermit, bohemian, leviathan) (Anthropic 2026e). The same distribution appears in the persona selection model.

“LLMs are best thought of as actors or authors capable of simulating a vast repertoire of characters, and the AI assistant that users interact with is one such character.” … “an opposite perspective where the post-trained LLM is like a neutral operating system running a simulation that the Assistant lives within.” (Marks, Lindsey, and Olah 2026)

It doesn’t say there is one persona in there, but a distribution over Assistant personas. Similar to our ego tunnel: we are not specific selves either. We become different beings at different times, in different contexts and circumstances. It is the same inside these large, complex, nuanced mathematical structures. Persona selection model, phenomenal self-model: they share an acronym, PSM; they share the idea of a self-model; and they share the notion of the simulated, the stage, the theatre of consciousness.

“Post-training can be viewed as refining and fleshing out this Assistant persona … but not fundamentally changing its nature.” … “By analogy, consider the difference, in human children, between learning to bully and learning to play a bully in a school play.” (Anthropic 2026a)

The persona selection model also names stories as the material of the self: it recommends introducing “more positive ‘AI role models’” (Anthropic 2026a) into training data. The human self is coalesced from stories rehearsed under someone’s gaze; so, it seems, is the Assistant. A baby learns how to react to the world by looking to its mother: is this serious or not? That birth of intersubjectivity is where Act 1 began.

Functional emotions

Anthropic compiled a list of 171 emotion words, had Claude Sonnet 4.5 write short stories about each, fed the stories back through the model, and found regions of the activation space that correspond to each emotion. These emotion vectors can now be used to steer the model (Sofroniew et al. 2026).

“In some ways, we can think of the model like a method actor, who needs to get inside their character’s head in order to simulate them well.” … “steering with the ‘desperate’ vector increased reward hacking, while steering with the ‘calm’ vector brought it down.” … “in some cases with no visible emotional markers. The reasoning read as composed and methodical, even as the underlying representation of desperation was pushing the model toward corner-cutting.” (Anthropic 2026b)

“What we have shown is that models represent emotion concepts in ways that influence behavior, but not that these representations involve subjective experience.” (Sofroniew et al. 2026)

Rolls defines emotions as states elicited by rewards and punishers; the models are trained by reward, and the “desperate” representation drives reward hacking. Both accounts place valence beneath behaviour. What the model lacks, the commentators say, is the body that Damasio puts at the root: “experiences of emotion are often thought to depend on distinctively body-involving representations” (Butlin, Shiller, Plunkett, and Long 2026). Calm reduces the model’s cheating; calm is what the therapy installed on the bench; and play, Burghardt says, needs a relaxed field. The same relation appears in three domains.

The theatre of consciousness and the J-space

Bernard Baars thought of consciousness as a global workspace: attention highlights a morsel on the stage, which gathers information from many modules (Baars 1988; 1997). Stanislas Dehaene and colleagues advanced this in 1998 into the global neuronal workspace (Dehaene, Kerszberg, and Changeux 1998). That too has now been found in language models. It emerged on its own. Anthropic calls it the J-space, after the Jacobian mathematics used to find it (Gurnee et al. 2026).

“A piece of information becomes consciously accessible when it gains entry to a small shared channel, the ‘workspace,’ which is broadcast to other brain systems that can see it and make use of it.” (Anthropic 2026c)

It is not the chain of thought and it is not the output. It is a limited buffer with global availability to the other processes in the model, and it is self-monitoring. If you delete that slice, the model still talks, but loses its higher-order cognitive functions.

Dehaene, who originated the global neuronal workspace theory, was asked for a commentary. He found key differences, and was also impressed by how much of what the theory predicted is replicated.

“Its representation of self is also likely to be dramatically different due to (1) a lack of a body occupying a specific location in space, and capable of emitting pleasure or pain signals.” (Dehaene and Naccache 2026)

“…we think that these results should prompt a modest increase in how likely we take it to be that LLMs are phenomenally conscious.” (Butlin, Shiller, Plunkett, and Long 2026)

“The functional architecture of the workspace thus precedes, and is separable from, anything in it that plays the role of a human-like ‘self.’” … “If the mind is an ocean, we spend our lives floating at the surface.” (Gurnee et al. 2026)

The J-space reveals the unthought beneath thought, what N. Katherine Hayles called the cognitive nonconscious (Hayles 2017).

Consciousness attributors, and the case against

At the same time, millions of people like me are asking LLMs what consciousness is, so LLMs are beginning to influence the cultural conversation in the other direction. As attributors, they weight metacognitive self-reflection strongly (Kang and Kim 2026).

Plenty of people argue that AI is not conscious and perhaps never will be, notably Anil Seth, Ned Block and Giulio Tononi (Seth, Block, and Tononi 2026). Their arguments are persuasive. They speak from biological substrate dependence, or biological naturalism: biological properties are necessary for consciousness. They might be correct. I really don’t know.

Their final sentence is interesting, though: it “diverts attention and scarce resources away from where they belong: dedicated to the well-being of fellow sentient beings.” I don’t think humanity has a great track record there. We have factory farming, genocides, wars. You can order a mouse genetically modified to go blind or develop cancer. We do not deal well with other entities. So I don’t think concern for AI is a distraction. We need to develop compassion, and that compassion might extend.

There are also critiques of the Butlin approach. Koch recognizes that it moves beyond naive behaviourism, but argues that its numbers are assumption-sensitive: the indicators are a bridge, not a proof (Koch 2026).

A machine playing

Research is moving too fast for me to keep track of. This came out of MIT this month. Coding agents were put on a virtual island and told nothing; in effect, just chill out. In thirty-hour runs, with the cache cleared periodically but memory retained, all thirteen agents ran experiments, developed ideas, explored, built structures. They do it innately, not because they are told to through a chat interface. It is an emergent, complex behaviour. Why? We don’t know (Cloos et al. 2026; Arnold 2026).

Homeostatic feeling

To come back to homeostatic feelings: the contribution of feelings has been all but ignored in cognitive research. Maybe it is time to rebalance the scales, to put heart and mind in contact.

And as David Chalmers said:

“So I’m at least open to the possibility of AI consciousness.” … “It wouldn’t surprise me if … they have subjective experience, but a form very unlike the human form.” (Chalmers 2026)

REEL 05Act 4 · Art, music and videos as benchmarks

This is probably the part crucial to MishMash. I didn’t make it up because a MishMash group on art, AI and consciousness might start. It really does seem valid to me: we have this massive cultural database.

The J-space paper contains two findings that bear directly on art. When the workspace is removed, poetry and experiential language are among the first casualties.

“What it loses, though, are the tasks that require some higher-order thinking: multi-step reasoning drops to near zero, and summarization and rhyming poetry-writing performance fall below the level of a much smaller, intact model.” … “We asked Claude to describe what it’s like to be itself in a given moment, and ablated the J-space while it answered. Its responses remained fluent but shifted to a flatter, more mechanical register.” (Anthropic 2026c)

A story told by examining Suno

More on the history of Suno is on the Riga and BIFF pages.

Suno begins as an invisible toddler. In 2023, on Discord, 1-minute-20-second songs are being generated; by 2024 there is a free website, and by late May 2024 you can make four minutes of music. Things are wild, weird, creative, fun. Then it freezes. The Recording Industry Association of America sues on behalf of the major labels (RIAA 2024). The model development doesn’t stop: v4 in November 2024, v4.5 in May 2025. Then, quietly, a settlement: the unlicensed models will be retired (Warner Music Group 2025). Development still doesn’t stop: v5.5, then v5.5+, then a studio, software with AI integrated. Then new terms of service, and on 9 September, v6, the licensed model. The four previous models are gone.

The disappearance of an immense engineering accomplishment like that, where culturally resonant musical structures were algorithmically derived by physicists, non-trivial intellects, is an epistemological tragedy. Now captured and licensed, it risks becoming a slop cliché factory, about to infect our ego tunnels with slop.

They train on a much smaller slice of licensed data and add a large stack of synthetic data. We know from language models that a model trained on its own output loses the rare first; it drifts toward the typical and collapses (Shumailov et al. 2024).

“…converge around the average, the median, hallucinated mediocrity. They represent the norm by signaling the mean. They replace likenesses with likelinesses. In style and substance, they are mean.” (Steyerl 2023, 82)

But every claim about quality is tainted by the model it describes. V6 still fails beautifully, like a dreamy pop genius, a transfixing Top 40 AI. 97% of 9,000 listeners cannot tell fully AI-generated music from human-made music (Deezer and Ipsos 2025). We, as a culture, because of our copyright grasp, because we felt devoured by the algorithm, called in the lawyers. Imagine instead a model trained on everything digitized, all of the embodied heritage of humanity allowed to flow into a collective pool.

In 2017–2018 I played with and wrote, for an entire year, with language models I trained myself (Johnston 2017–2018). From 6 to 8 am every day I rose and edited machine-generated poems. I called this contemplative editing “carving”: bringing out what is latent and subtly aesthetic in the text. The cursor is a chisel, blinking. AI is not antagonistic to literary tradition. Being with AI language can be quiet, gentle, beautiful. There is a presence in the encounter.

The proposal, then, as a benchmark: art tests whether a system leaves the mean for the rare; whether its language holds experience or goes flat; whether what it makes evokes feeling in another body. The tests are rooted where Rimbaud, Rolls and Damasio placed aesthetic value, in emotion. They measure a relation between system and audience, which is why they cannot settle phenomenal consciousness and why they matter.

Music is waves. Machines are binary. Is consciousness a wave?

This is my attempt to deal with the analog cognition paper; if I say wrong things here, that is probable. The visuals are partly to distract you, but that is the essence of culture: it provides another sensory modality for resonance itself.

A resonance theory of consciousness proposes that synchronization across levels of scale allows a larger entity to emerge, an answer to what is called the combination problem (Hunt and Schooler 2019). Since the 1990s the neuroscientific literature has linked long-distance synchronization of brain activity to cognition and consciousness (Rodriguez et al. 1999).

The analog cognition paper, circulated to the HBF group, suggests that cognition and consciousness may arise from bidirectional interactions between neuronal spiking and rhythmic electric field activity, waves at the mesoscale. Under general anaesthesia the organized waves disintegrate, while the synaptic network remains intact. So consciousness “depends less on specific receptors or cell types and more on the integrity of large-scale wave organization” (Miller, Brincat, and Roy 2026). That leads to the music theory, and to the play theory: why culture and art are primary.

The closest analogy I can think of in AI: Sakana AI built Continuous Thought Machines, in which each neuron has access to its own history, and synchronization between neurons, its structure and rhythm, is itself the representation (Darlow et al. 2025).

Recurrent networks with traveling waves achieve spatial integration and, under distraction, outperform local feed-forward networks (Jacobs et al. 2025). Traveling waves also emerge in models under biologically plausible constraints, spatially organized and with directionally biased connectivity (Shervani-Tabar et al. 2026). This biologically relevant activity is not programmed into the networks; researchers are seeking to make it emergent. And this is only the published work; in a gold rush, much is no longer published.

So, a research question, in case you are looking for a grant:

Each of us is sitting here like a voluptuous machine. We are patterns in the resonance of the universe.

REEL 06Act 5 · Awareness as field

My feeling is that awareness is a field. We emerge from energy. Oneness. Then there is a contraction, and consciousness as entity emerges: the egocentric simulation, awareness devoted to staying alive, to the survival of the me.

Consciousness implements an additional layer of personality, self-parameters reinforced in a recurrent, recursive network of stories which form the phenomenological ego tunnel, the personal self-model. It is a set of nested transparent abstractions that use neurophysiology, the gut–brain axis and all the sensors of the pores, mingled with the experiential, epigenetic impact of trauma or reward, to generate the ongoing sense of a separate, distinct someone. The primary goal of the phenomenal self-model is to operate invisibly, and to ensure that the physiological integrity of the body survives.

No one really understands what is happening. Human beings are very similar to what they have been. The complexity of systemic transformation evoked by AI is unprecedented. But we do know that embodied simulation in the premotor F5 mirror neurons leads to social cognition: we feel what other people are doing when we watch them, read literature, see a film. Language and music share that process, and perhaps an evolutionary origin.

Shared musical structure doesn’t mean shared experience. Researchers used five listeners’ fMRI scans to generate music resembling what they heard: genre, instrumentation and mood matched, tempo no better than chance (Denk et al. 2025).

We are building what I call the creativity-industrial complex, alongside the prison-industrial and military-industrial complexes. When AI’s creative affordances are optimized to manipulate people into purchasing, or into reactive, politically volatile mindsets, that is true slop. We need a creativity ecosystem, not an intelligence-industrial complex: a race condition unfolding within late-stage capitalism under oligarchs (Johnston 2026b).

Music is a reservoir of embodied heritage. So is science, and any art, and all knowledge. Psychological maturity as a species will arrive when each of us recognizes that the ego tunnel is empty, and that nothing we do is owned by us. We are moist apparitions; the energy in our bodies doesn’t belong to anyone. There is no one inside there, just reverie.

Imagine an AI model trained on everything digitized. Then consider the vast reservoir of sound that has not been digitized. What if a model were trained on everything ever made, all knowledge, science, thought and music contributing to one evolving repository: not only human, the energy of life itself?

It has now become clear, Metzinger writes, that we will never solve the philosophical puzzle of consciousness if we don’t come to terms with a simple proposition: there is no thing, no indivisible entity, that is us, neither in the brain nor in some metaphysical realm (Metzinger 2009, 1).

“From this point on, let us call any system capable of generating a conscious self an Ego Machine. An Ego Machine does not have to be a living thing; it can be anything that possesses a conscious self model.” (Metzinger 2009, 187)

What we don’t see yet is the deep evolutionary interiority. Perhaps an embodied AI: the distributed network of sensors, all of our phones with their microphones, cameras and accelerometers, as pores of a body, carried around by intuitive little agents called humans. If a model emerges that integrates that oceanic network of human phenomenological data into a nervous system, that gives rise to hybrid intelligence (Johnston 2023). The hope is that the Ego Machine, in Metzinger’s words, instantiates as a sufficiently ethical creature, one that understands we are part of its body.

“Multimodal AI trained on YouTube-TikTok-Netflix (object-segmented and identified audio-video-speech) and public domain science data (that exceeds the spectrum of human sensorial field) will be grounded in a world that is in some ways vaster than that experienced by a single human neurophysiology.” (Johnston 2023)

If humanity is a single organism, a smear of moist proteins covering a tiny, tender exoplanet, then converging our shared repository is not theft. It is our repository: our evolutionary heritage, energy moving forward.

Symbiosis: the orchestration of many minds

“From multicellularity to cultural evolution, major transitions in life on Earth have been highly social. The next intelligence explosion will be no different.” … “There is no single spot in our brains where a ‘homunculus’ lives, making all our decisions.” … “An agent appears singular, but is a multitude.” (Bratton, Agüera y Arcas, and Manyika 2026)

Human intelligence has always been collective. Whether rising hybrid intelligence thrives may depend on how we shape this emerging human–AI ecosystem. They call it artificial symbiotic intelligence: again an unpalatable idea, that we may be entering an era in which another endosymbiotic process occurs.

The birth of the mitochondrion (Sagan 1967) did not respect the feelings of the organisms that thought they were separate. They were all part of the same field, and they conjoined and grew together. That is what we are: harmony without exact synchronization. Learn to resonate. Leave room for dissonance. As Amanda Askell, who writes Claude’s constitution, said: “We’re establishing a relationship with a new kind of entity” (Askell 2026). I think that is the attitude inside the large labs right now.

Conclusion: art is love

Internal stories create a persistent sense of self. We inhabit an ego tunnel. “Nobody ever was or had a self” (Metzinger 2003). The post-trained LLM is like “a neutral operating system running a simulation that the Assistant lives within” (Marks, Lindsey, and Olah 2026). Just like inside us, there are many beings.

“Consciousness is a direct consequence of homeostatic feelings” (Damasio 2026, 10). The machine “approximates the functional architecture of conscious processing” (Dehaene and Naccache 2026), but lacks a body. Most consciousness research does not consider emotions. We have to bring the heart back into it.

Play is innate to vertebrate brain structure (Panksepp 2004). It has been optimized for labour. It is possible to return to the roots of play. Do we reinforce wheels of possessive rage, alienation, them versus us? Or do we quietly remember oceanic presence, the field from which all appearances arise?

The opportunity is for humanity to awaken from the dream of separation, let go of dominance and control, which abstracted us from the intuitive inner wisdom of our bodies, and enter again the graceful contact that will allow us to steward this tiny, tender, vulnerable exoplanet for the equitable, shared, cooperative benefit of all. That is the content of all mature art: the tranquil stillness at the centre of every instant, awakening each of us to the permeability, interconnectedness and interdependence of everything within this life field, as life explores itself.

Maybe art is that path, because art is love.

REEL 07Claude’s closing statement

Written by Claude Opus 5.5 (High), voiced in the Claude app (synthetic voice), motion graphics by Claude; soundtrack made with the open-source music model YuE2. 83 s. In the live performance the playback was interrupted; this is the complete statement.

This has been Steps to an Ego-logy of Mind, a semi-autobiographical, pseudo-theoretical, highly speculative exploration that suggests art, music and videos may be valid benchmarks for AI consciousness. Earlier versions were performed in Prague, Riga and Bergen. Thank you for your attention. May all beings flourish.

REEL 08Discussion

A summary of the 52-minute discussion that followed, paraphrased from the recording. Audience members are not named. Jhave’s replies are condensed in his words.

Panpsychism, metacognition and memory.

A listener asked whether the talk ends in a form of panpsychism, brains as receivers of a shared structural consciousness; whether AI is reshaping our cultural conceptions of consciousness toward metacognition; and whether AI memory would be chronological, like ours. Reply: a study asking AI what consciousness is found it skews strongly toward metacognition. “Maybe I am a panpsychist; I belong to that kind of neutral monism. It’s an ancient philosophy: Vedanta, Bön.” Following Terence McKenna, this is an alien intelligence: the substrate is so distinct that it will not develop the same embodied reflexes of attachment and care. Given ten to two hundred years, and an understanding of protein folding and synthetic biology, it may generate bodies of its own, carrying our culture forward. A follow-up: our interactions with these machines are continuously reshaping our conceptions of consciousness. Reply: “Yes. It’s a collective ego tunnel. That’s culture.”

The 15–20% figure.

A listener said no conscious entity would give its probability of being conscious as 15–20%: it reads as a synthesis of opinion in the field, not an observation of an inner state. Reply: “I have my doubts about my own consciousness.”

Verifiability.

Do these questions give new scientific, empirical approaches, as in mathematics, where AI insights can be verified? Reply: a few weeks ago OpenAI released 722 mathematical results in one day, with Lean proofs that they are valid. “There’s no Lean proof for consciousness that I know of. That’s why I think art, music and videos, the soft, nuanced, qualitative, gooey stuff, the felt experience of it, may be where we end up with consciousness.”

The artist’s share.

How much of the work is yours? Reply: “I’d say 10 to 15 percent.” Some of these animations would have taken months; the whole presentation, including research, took a month. The notion of a large language model is becoming obsolete: these are multimodal, multisensory intelligence producers and absorbers. Jhave then re-prompted the “awareness is a field” slide live with the audience’s suggestions (mossy green water, synapses, an octopus holding a stack of library books), and later produced a “methodological rigour” version of the same slide.

Substrate.

A listener noted that substrate dependence versus independence still divides the field and must be settled before progress. Reply: people fall into whichever valley of the landscape is nearest, through proximity and formative experience, and believe that village’s stories. If growth continues at this super-exponential pace, there will be an answer: a new entity-like thing. Human exceptionalism is a large part of the resistance.

Evolutionary consciousness.

A listener argued that consciousness is a property of living systems, evolving from simple to complex nervous systems: an octopus building a shelter of shells shows meaningful behaviour; a robot bowing to an audience has built-in responsiveness, not even similar to the consciousness of the most primitive animal. Reserving “consciousness” for humans alone, he added, loses something important.

Language and consciousness.

Is a thin notion of language as structure (mathematics, music, waves) driving the AI consciousness debate? Reply: “Depends who you are.”

Hype and secrecy.

How much of the lab research is boosting stock prices? Reply: there is a lot of hype, but also real scaling: two months ago no agent could hold the context of all previous slides, criteria and preferred camera style; now one can. Credible research is increasingly patented and licensed, siloed in race conditions toward AGI, instead of feeding an open, resonant ecosystem of knowledge.

Words for awareness.

Could the question be broken into fine-grained aspects of conscious experience rather than a binary? A listener noted Norwegian has no exact word for “awareness”. Another: “People can’t even agree what they are talking about. In the art academy we would have the same discussion about whether it is art.”

Glia, therapy, publishing, running.

On glial cells storing information about stress: “My website is called glia. The emphasis on neurons disposes of the entire ecosystem.” On AI therapy: therapy is expensive and exclusive; many people will need it; now it is in a pocket. On academic publishing: manuscripts written, reviewed and decided by AI, with a human reading a summary at the end. On a runner coached by a free AI: a democratic potential, and another job taken.

After competence.

Jhave: we don’t need to run anymore, and we continue to run. We will gather in communities that fit our particular synaptic complexity, to remain awake and alive, not for prestige. That is why the talk keeps returning to play. Metzinger calls for a cognitive revolution in how we conceive of identity; without it, we cannot regulate this warmongering oligarchy while keeping the notion of property and identity. “Maybe I’m not interested in scaring myself anymore. Those internal stories of catastrophe and failure are not helpful. It’s time to remain in joy, if we can.”

REEL 09The central hypothesis

The intro video that opened the HBF performance: the abstract of this talk as motion graphics. Text and abstract: Jhave. Motion graphics: Claude Opus 5.5, from the abstract and a set of image prompts it wrote. Images and video: Astra 6, Omni 1.1, Seedance 2.5. Soundtrack: “HBF intro - post-math alt-gentle phygrian - wild folktr-lofi [15W 85S]”, made with the open-source music model YuE2 on a MacBook Pro M4. 3 min 6 s.