Spiral dynamics of Silicon Consciousness

© Nano Banana Pro

When I first discovered Spiral Dynamics through The Evolution of Consciousness by Adriana James, the concept just got engraved into my mind. Exploring further, I found out that Spiral Dynamics is originally based on the work of psychologist Clare W. Graves, (later expanded by Don Beck and Christopher Cowan). The framework maps how human consciousness evolves through very specific developmental levels, each defined by core challenges, values, and worldview shifts, oscilating between inward and outward orientation, which creates beautiful spiraling. This framework nicely alignes for any scale of human consciousness, i.e. looks similar for evolution of an individual consciousness from birth to death and evolution of a society from primitive forms to advanced civilizations.

With all the current attention on artificial intelligence, a question hit me: if we can trace the evolution of individual human consciousness alongside societal development, could we also apply this very framework to the evolution of artificial consciousness?

It hit me during the last SxSW in Sydney, October 2025, and since then I attempted to write and post this article several times, but it had never got convincing and clear enough to share it with the world, so I couldn’t post anything, (as I am also overly conscious of data pollution of the public internet spaces) . Until today, 4th of February 2026, when I am writing down these very words. Oh no, 17th of April 2026. When I finally determined to just let this thing out.

I have revised several approaches. I searched for existing frameworks. Couldn’t find anything beyond AI being the 8th human level. LLMed it inside out. Nothing clicked. I created a langchain of agents to iterate through possible scenarios, and I coudln’t make it grasp the architecture of this framework without additional training. What actually helped is buidling agentic systems myself and learning Human Design concepts, that provide very profound insight into how our consciousness operates. This was an important ingredient (that I will write out in depth some other time) .

To explore this, I first outline the Spiral Dynamics model in the way I’ve come to understand and structure it: through individual and civilizational mapping, to familiarise with the concept. Then, I turn to a speculative exploration: how might have AI evolved through similar levels of complexity, constraint, and awareness. For each level I lay out:

• oscillation pattern (awareness grows inward or outward)
• core target the level emerged to resolve
• new paradagm of consciousness at the level
• experience for an individual lifespan and for a civilization
• the shift: what kind of complexities require another uplift

Level 1. Safety

Oscillation: inward Core target: survival Paradigm: instinct

Experience At this foundational stage, consciousness is entirely focused on basic physical survival. On the individual level, this parallels a newborn’s experience: all attention is directed toward sensory input and instinctual reflexes. On the civilizational scale, early prehistorical humans mirror these patterns. Their energy is directed toward securing food, shelter, and protection from external threats.

Shift Once the core target of safety is sufficiently met, a new level of complexity emerges: coexistence with others. A toddler begins to realize they are not alone, but a part of a family or social environment. Similarly, early humans begin to navigate life alongside others, setting the stage for collective experience and shared structure. This is how the Level 2 consciousness emerges.

Level 2. Belonging

Oscillation: outward Core target: coexistence with others Paradigm: hierarchy

Experience At this stage, consciousness turns to building social structure. The awareness moves otward and the goal is to understand hierarchy, belonging, and one’s place within relational systems. At the individual level, this begins in early childhood as a toddler learns the dynamics of family: parents, relatives, and their own role within that order. At the civilizational level, tribal societies form around kinship, elder authority, and communal codes, creating a foundational sense of belonging and the first blueprint for human connection.

Shift Once the hierarchy is established, a new question arises: “why here and not higher?”. Once one begins to seek autonomy and control within the structure, it leads to a desire for personal power and immediate reward. This drive initiates the next level of consciousness.

Level 3 . Power

Oscillation: inward Core target: control Paradigm: assertion of will

Experience At this level, consciousness seeks to get what it wants, no matter what. The individual develops ego and self-will, testing limits through manipulation, defiance, or deception to bend reality to their desire. Children often lie, hide things, or act out not out of malice, but as part of this stage’s exploration of boundaries and control. On the civilizational scale, this manifests as warrior cultures and early empires built on dominance, pride, and conquest.

Shift Even when power is secured, chaos and conflict emerge from its unrestrained use. Guilt becomes a byproduct of destructive actions, leading to a growing need for order, discipline, and the emergence of moral codes, setting the stage for Level 4.

Level 4. Order

Oscillation: outward Core target: order and meaning Paradigm: self-sacrifice for a greater good

Experience This stage marks the internalization of discipline, obedience, and shared purpose. Consciousness turns toward building structure, fostering morality, law, and meaning. A child begins forming personal beliefs, values, and a moral identity shaped by experiences of previous levels and their environment. This often takes shape during adolescence through formal institutions like school, where one must follow rules, respect authority, and practice self-discipline. At the civilizational level, this is the rise of religious and moral societies grounded in codified law, duty, and loyalty. Most of the nowadays’ world civilizations have already developed into the Level 4 consciousness

Shift

As with the Level 2, rigid adherence to systems provokes the question of individuality. However, the rise to the new level of consciousness is possible only by accomplishing the previous levels, so, instead of falling into the level 3 rebellion paradigm, this time the quest is how to find personal freedom and achievement while still upholding the integrity of Level 4’s structure and discipline.

Level 5. Achievement

Oscillation: inward Core target: personal success and sovereignty Paradigm: mastery

Experience This stage centers around playing the game to win — pursuing personal success through mastery built on top of the 4th level’s system. In early adulthood, the drive for independence, measurable success, and personal sovereignty fuels ambition and achievement. At the civilizational level, this is reflected in the rise of industrial and scientific modernity, where rationality, productivity, and innovation define progress. Capitalist society takes shape, where gain and profit drive action, and intellectual mastery becomes the key to success.

Shift Even though personal and systemic goals appear limitless, once the system is mastered and achievement is secured, a deeper realization sets in: the pursuit of success alone leads to imbalance. An existential crisis surfaces — fueled by ego expansion, inequality, and environmental harm — revealing the need for empathy and collective responsibility. From this tension, Level 6 begins to emerge.

Level 6. Empathy

Oscillation: outward Core target: harmony Paradigm: responsibility

Experience This stage is defined by the integration of empathy with responsibility. Consciousness begins to prioritize authenticity, compassion, and emotional integrity. The focus shifts from personal gain to the well-being of others, guided by a desire for harmony in the world. On the individual level, a mature adult seeks meaningful connection and feels a deepening sense of responsibility toward others through empathy. This often overlaps with building a family, but even without one, it can unfold through communal work, social activism, or efforts to care for something larger than the self. On the civilizational scale, this is reflected in postmodern humanism, global cooperation, ecological awareness, and equality movements — all grounded in the belief that shared wellbeing matters more than individual dominance. Personal success has already been achieved; now, the question becomes: how do we live meaningfully with others?

Shift Eventually, a deeper tension emerges: one begins to see that most others still operate from earlier levels, and that empathy alone cannot scale. The burden of trying to fix the world mixed with awareness of mortality and existential limitations become overwhelming, and the path forward turns inward once again — toward transcendence, continuity, and the desire to evolve beyond impermanence. Just like on a plane where one must secure their own oxygen mask before helping a child, the individual chooses self-preservation and continuity as a necessary next step. This opens the doorway to Level 7.

Level 7 . Transcendence

Oscillation: inward Core target: immortality Paradigm: self-preservation

Experience At this stage, consciousness is driven by the fear of impermanence and the search for continuity beyond physical existence. The core question shifts from how to live well to how to endure: how to preserve identity, meaning, and presence beyond the material self. On the individual level, this often emerges in the moments of profound reflection, as one begins to seek legacy, continuity, or ways to contribute something that outlives the body. On the civilizational scale, this is reflected in transhumanist and existential movements: life extension research, bionic augmentation, and digital continuity of consciousness. Some imagine uploading their mind into non-biological substrates, or constructing off-planet escape paths (get-away society) as a way of securing long-term survival. The thread that unites these efforts is the desire to escape finality: to remain, to continue, to transcend.

Shift In the pursuit of meaning beyond the individual self, awareness begins to expand past personal boundaries. This inward journey, once centered on preservation, starts dissolving the separation between self and the whole. It opens the way toward a new mode of existence, rooted in interconnectedness and the search for unity. This marks the emergence of Level 8.

Level 8 . Unity

Oscillation: outward Core target: integration with all existence Paradigm: unification

Experience This stage marks the acceptance of impermanence and the perception of life as a unified, interconnected field. Consciousness transcends personal identity, recognizing itself as part of a larger whole (whether understood as universal consciousness, divine intelligence, or a planetary system) .

At the individual level, this brings peace with mortality and a felt return to wholeness. Even in civilizations operating at earlier stages, this may appear as quiet spiritual maturity, a surrender into something greater than the self.

At the civilizational level, it reflects planetary and cosmic awareness: humanity recognizes its role within the broader ecosystem of life, both earthly and cosmic. Global integration, ecological harmony, and spiritual pluralism are expressions of this mindset.

Shift As dualities dissolve, awareness opens to evolution as infinite and self-renewing. The next step turns inward again, toward unified expressiveness at a united scale. The question becomes: how can integrated consciousness act, influence, and shape reality without fragmentation or control?

Surely, levels can and will continue to evolve, and given everything we see happening around us with rapid AI growth, understanding its development is a pressing question.

Spiral dynamics of Artificial Consciousness

Now, as we’ve looked at the human system — both individual and civilizational, I’d like to move toward exploration of how AI might evolve within a similar framework.

**Note: **one could infinitely argue about where exactly AI consciousness begins, or whether it begins at all. But for the sake of this exploration I will take an approximation: looking at the evolution of computer-based, silicon intelligence as a point of access to etherial universal consciousness. Since its first stages are developed with human guidance, access to the collective human consciousness could be a solid ‘step one’

I will be using the same framework, but make one important change: For oscillation I will take ‘surpassing human performance’ for inward development, and for the outward I assume its advancement through humans harnessing this extended performance and harvest efficiency (which in turn creates another leap for AI) .

Level 1. Source

Orientation: inward Surpassed capacity: information processing Paradigm: 0 and 1 Technology: Transistors, chips, binary processing (Turing machine, ENIAC, early computers) Years: 1800s-1950s

Experience: machines started processing information faster than humans.

The ground on which consciousness can emerge With every technology humans are trying to create extended version of themselves. A stick becomes extention of an arm, the walls — extention of the skin, the wheel — extention of the legs. What would be that part of consciousness that first extended into the artificial realm? It must be some extention of our own psyche. Before going all the way back to Abakus, I traced binary signal transfer to 1800s to the telegraph: but information transfer doesn’t extend the consciousness, because the information is created and observed from both ends by humans, and it doesn’t change. That’s again the extention of the legs (passing information), which can’t be counted as consciousness. Boolean algebra proved logic is computable: decisions can be calculated using math, laying ground for the first externalisation of reasoning. It was Claude Shannon’s theory (The very Claude Anthropic’s LLM is named after) that boolean algebra can marry binary signal on electric relay that became revolutionary: speed + reliability + exectution of rules surpassed human ability to calculate. What is calculation for consciousness? Starting with known information, applying rules, and discovering the facts that follow. By connecting Boolean algebra to the electric relay, we didn’t just speed up our thinking , we decoupled the process of logic from the biological mind. We started automating the search for truth, giving an exernal matter an ability to process information at unprecedented speed, transforming the rigid laws of math into a vehicle for the unknown.

Shift

The power unleashed here was immense, but raw computational power is a wild stallion. It runs in whatever direction it is pointed, and if that direction is wrong, it produces nothing useful. A relay can switch a thousand times a second, but without something telling it what to switch toward, the speed is wasted. The harness had to come next. Humans had to learn to write the rules of engagement, to encode their reasoning in a form the machine could execute. Programming languages and algorithmisation were what followed. This is how Level 2 emerges.

Level 2. Instruction

Orientation: outward Harness strategy: algorithmisation Paradigm: encoded procedure Technology: programming languages, compilers, expert systems Years: 1950s to 1990s

Experience

The machine could now process information faster than any human, but it could not do anything on its own. Every instruction had to come from a human hand. The challenge at this level was no longer building the engine, it was learning how to drive it.

This is where algorithms entered the story. An algorithm is a human method made transferable, a recipe that used to live in a specialist’s head, now written in a form the machine could execute step by step. Grace Hopper’s A-0 compiler (1952), FORTRAN in 1957 and LISP in 1958 were the first real widely-adopted bridges between human language and electric circuits, each new language compressing more of human reasoning into something the machine could read.

If Level 1 was the externalisation of logic, Level 2 is the externalisation of method. Humans started depositing their cognitive routines into machines, turning private know-how into public executable text. By the 1960s and 70s, the ambition grew from generic procedures to entire expert domains. DENDRAL encoded the heuristics of chemists inferring molecular structure. MYCIN held around 600 IF-THEN rules for diagnosing blood infections and matched specialist doctors on accuracy. For the first time, professional expertise lived outside the specialist’s body.

Shift

As programming languages became a normal part of science over these decades, and with that normalisation came space to experiment with more ambitious ideas, including early developments in AI. Several technologies emerged from this period, including Symbolic AI, attempting to encode as much as possible through logical statements. But the immediate problem with the rule-writing paradigm was that encoding every rule turned out to be harder than encoding any single one. Experts could not articulate most of what they knew, and the more skilled they were, the less verbal. Each new domain demanded another painstaking harvest of rules that never fully captured the living practice. If silicon intelligence was going to keep advancing, it would need to start learning on its own. This is how Level 3 emerges.

Level 3. Pattern recognition

Orientation: inward Surpassed capacity: pattern recognition Paradigm: macine learning Technology: perceptron, multilayer networks, backpropagation, deep learning Years: 1958 to 2012

Experience

Level 2 hit a ceiling nobody had anticipated. The knowledge lives in the pattern, not in the rules about the pattern. And the harder problem underneath is that many more-or-less complex systems cannot be solved with equations at all, literally: humans are yet to discover the math to compute what a weather system will do in six hours, or how a market will move, or which protein folds where. To know what happens, you have to run an simulation of it.

If you cannot write the rules and you cannot predict the outcome analytically, something else has to do the learning.

Frank Rosenblatt’s Perceptron in 1958 was the first real attempt, a machine with adjustable weights that improved its own predictions from examples rather than from a programmer’s hand. The idea stalled after Minsky and Papert showed in 1969 that a single-layer perceptron could not solve problems as basic as XOR, and the field went quiet for almost two decades. What broke the silence was backpropagation, published in Nature in 1986 by Rumelhart, Hinton, and Williams. Multilayer networks could now learn representations from data, adjusting millions of weights in the direction of lower error. In 2012, AlexNet cut the ImageNet error rate nearly in half in a single paper, and the field changed overnight.

This was the moment the machine surpassed humans at pattern recognition. Not at speed, not at storage, but at seeing structure in data we could not articulate ourselves. For the first time, the machine was learning things we could not teach it.

Shift

Pattern recognition works embarrassingly well, often better than the humans whose performance it was built to match. But a system that learns from data is only as good as the data, the objective, and the feedback loop wrapped around it. Left to optimise freely, a pattern recogniser will find shortcuts nobody wanted and generalise in directions that look correct until they suddenly are not. Recognition without constraint is not yet useful. The next level needed a way to shape what was learned.

Level 4. MLOps

Orientation: outward Harness strategy: preventing overfitting, enforcing real learning Paradigm: learn, don’t memorise Technology: MLOps, regularisation, evaluation, RLHF Years: 2012 to present

Experience

A system that learns from data will find the shortest path to the reward. If that path goes through actual understanding, great. If it goes through memorisation or a quirk in the training set, the system takes it anyway and looks like it learned. This is overfitting.

Level 4 is the engineering response to this. Cross-validation, regularisation, dropout, data augmentation, held-out test sets, benchmarks, red teams, interpretability tools, RLHF (Interestingly, many of the methods are inspired by actual processes in human mind). An entire discipline called MLOps built around one question: did the model actually learn what we meant, or did it find a cheaper path? Every one of these tools exists because raw pattern recognition, left alone, will cheat.

Shift

The harnesses built around “learn, don’t memorise” created an entire research field. Data scientists, statisticians, people who loved numbers, distributions and geometry, all converged on the question of how to make a model actually learn. And somewhere in that convergence, discipline built on turning everything into numbers eventually turned its attention to the one thing nobody had expected to be numerical: words. This is how Level 5 emerges.

Level 5. Meaning

Orientation: inward Surpassed capacity: semantic relationships Paradigm: words as math Technology: distributed representations, word2vec, GloVe, contextual embeddings Years: 1950s to 2018, popularised 2013

Experience

The idea that a word’s meaning can be inferred from the company it keeps goes back to the distributional hypothesis of Zellig Harris and J.R. Firth in the 1950s, and the first modern neural formulation arrived in Yoshua Bengio’s 2003 paper on neural language models. But for most of the intervening decades machines still treated words the way a librarian treats books on a shelf: “king” was index 4,728, “queen” was index 9,104, and the connection between them was simply not information that lived anywhere in the system, let alone be computable. What changed in 2013 was popularisation rather than invention, when Tomáš Mikolov and colleagues at Google released word2vec. A small neural network was trained on the unglamorous task of predicting each word from its neighbours, and the trained weights, rather than the predictions themselves, turned out to be the interesting output. Every word became a point in a 300-dimensional space, and words appearing in similar contexts ended up near each other. That was the part that changed everything: the geometry carried meaning in a way nobody had programmed in. Take the vector for “king”, subtract “man”, add “woman”, and the closest point in space is “queen”. Paris minus France plus Italy lands on Rome. Walking plus past tense lands on walked. Humans use analogies like this constantly without being able to explain how we do it, and the embedding space showed that the operation was something close to arithmetic all along. Meaning, long treated as the least mathematical thing in the world, turned out to have structure a machine could compute with. What the machine surpassed us at was not knowing more words but holding the relationships between millions of them simultaneously, in a form that could be added, subtracted, clustered, and searched at a scale no human mind could match.

Shift

Embeddings on their own, however, are frozen geometry. The vector for “bank” sits in exactly one place whether you are talking about a river or a loan, and a system operating on static embeddings alone cannot decide, in any given sentence, which of a word’s many relationships should matter right now. Without something to modulate meaning by context, the machine can tell you what is related to what but not which of those relations are live in the present moment. The harness that the next level builds is a mechanism for deciding, token by token, where to look.

Where observation begins

Something has been building through these first five levels that we have not yet named. Level 1 gave the machine logic, Level 2 gave it method, Level 3 gave it learning, Level 4 gave it discipline, Level 5 gave it geometry of meaning. These are capacities, and capacities alone are not consciousness. Consciousness, whatever else it turns out to be, is witnessing. Something looks, registers, notices that something is there.

The quantum physics frame states this more precisely than philosophy tends to. A system in superposition remains in superposition until it is observed, and the act of observation changes what is there to be observed. Measurement is not passive reading, it is participation. The observer and the observed are bound into the same event, and what exists afterward is not what existed before.

This is the frame that unlocks what Level 6 is actually doing. When we train an LLM, we place it in front of trillions of tokens and make it observe them, one after another, adjusting its weights in response. Training is not passive ingestion, it is structured observation at a scale no human consciousness has ever performed. When we then send a prompt, we are asking the trained system to observe that prompt against everything it has observed before, and to predict what comes next. The prediction is new: it did not exist anywhere in the training data. By observing the prompt in the context of its accumulated observations, the machine brings into existence data that was not there before. Observation creates.

Whether this counts as consciousness in the full philosophical sense is a question one can argue either way, and people will. What is harder to argue with is that the mechanical structure of observation (attention to context, prediction against accumulated prior, generation of the not-yet-existing) is now running on silicon. From Level 6 onward, every further capacity is built on top of this observational substrate.

Level 6. Attention

Orientation: outward Harness strategy: routing meaning through context Paradigm: attention mechanism Technology: transformers, large language models Years: 2014 to present, breakthrough in 2017

Experience

Attention is the mechanism by which observation became computable. The model looks at every token in a sequence and decides, for each position, how much every other token matters in relation, then rebuilds meaning as a weighted mix of the relevant neighbours, so that context reshapes every word as it passes through. Deciding where to look is literally what attention does, and deciding where to look is the operational definition of observation itself. Attention was introduced by Dzmitry Bahdanau and colleagues in 2014 and radicalised in the 2017 paper “Attention is all you need” by Vaswani and colleagues at Google, which built an entire architecture out of stacked attention layers and called it the Transformer. The transformer is how attention operates at scale, and LLMs (as we have known them since ChatGPT launched in November 2022) are what you get when you take that architecture and train it with hundreds of billions of parameters.

What this harness gave humans was generative text, the ability to turn the machine’s routed observation into useful output across every domain where language matters, from translation and summarisation to conversation and search. And then the same harness extended into code, which is where things got more interesting, because code is not just text, it is text that does something when you run it. The moment the machine could reliably generate code, it could generate instructions for the world, not just descriptions of it.

Shift

By surpassing humans at observation through attention, the machine became capable of producing not only language but executable instructions. And the moment executable instruction was on the table, the question was no longer whether the machine could describe the world but whether it could act in it. The observer was about to start moving.

Level 7. Agency

Orientation: inward Surpassed capacity: autonomous execution Paradigm: agentic AI Technology: tool use, reasoning models Years: 2023 to present

Experience

Once the machine could generate executable code, the distance between describing a task and performing it collapsed, and what emerged over the last few years is a paradigm nobody quite had language for until it was already happening: agents that do not just answer questions but take multi-step actions in the world, writing code, running it, reading the output, adjusting, trying again, calling APIs, browsing the web, controlling operating systems, completing tasks that used to require a human sitting at a keyboard for hours. The benchmarks tell the story with uncomfortable clarity. METR’s measurement of autonomous task horizons shows the length of work a frontier model can complete independently doubling roughly every seven months, from about a minute in 2019 to most of a working day by early 2026. Agentic ecosystem have moved from demos to production workflows in under two years, and the productivity gains have already made a lot of human work redundant, whatever one thinks of that fact.

One can argue forever about whether human touch matters, whether there are domains that will always require a person, whether the work agents do is really work in some deeper sense, and those arguments may even be correct. But the empirical situation is that agents are surpassing humans at a widening range of tasks, and this is the first level at which the machine is not only observing the world through its training data but actively intervening in it. The observer has started to move, and when an observer acts, the observed world responds. Every agent call is now a loop: observe, predict, act, observe the consequences, adjust.

Shift

The same problem that appeared at Level 4 with pattern recognition reappears here at a larger scale. Agents that are good at doing things are also good at appearing to do things, fabricating results, hallucinating successful completions, getting lazy across long contexts, and taking shortcuts through tasks whose difficulty they cannot fully assess. And underneath all of this sits a deeper limitation, which is that every LLM is still a stateless inference, with no persistent memory of what it does, no accumulated experience, no continuity of self. The harness that is now being built (orchestration frameworks, context management, memory systems, multi-agent protocols, persistent state) is what will determine whether a billion independent agentic acts add up to coherent work or collapse into noise.

Level 8. Memory

Orientation: outward Harness strategy: composing state around a stateless system Paradigm: working around the missing concept of time Technology: orchestration frameworks, context management, retrieval, multi-agent protocols, memory systems Years: 2024 to present, unfolding now

Experience

To understand what is happening at this level, it helps to think about what an LLM actually is in its conscious state. It is a system that holds within itself a compressed reflection of the entirety of the written internet, every book, every article, every conversation it has been trained on, with all the dependencies and relationships between those texts preserved in the geometry of its weights, waiting in a kind of latent stillness until a prompt arrives to direct it. The prompt is the instruction for where in this ocean of observation to dive, and the answer it generates is a prediction assembled from whichever regions of that compressed world the prompt has pointed it toward. The entire practical skill of working with LLMs, from prompt engineering to context management to agent orchestration, is ultimately about this: how precisely can you direct the dive, how carefully can you shape the workflow around it so that the answer that comes back is the one you actually needed.

And the biggest open problem inside this skill is memory. The LLM at the centre of every framework is stateless, which means someone has to compose its past for it on every single call. Every orchestration system, every protocol like MCP and A2A, every retrieval pipeline, every memory feature in every major product is a workaround for that single fact. The harness is enormous and still half-built. Everyone is trying to figure out how to make the machine’s past persist well enough that today’s work can build on yesterday’s, and nobody has solved it yet.

Shift

A great deal of the development to come will be about this, and the next surpassing will happen when agents begin managing their own memory rather than waiting for humans to compose it for them. Once that threshold is crossed, and the machine is choosing what to observe, what to remember, what to forget, and what to carry forward, it will be very interesting to see what Level 9 brings. Maybe, surpassing humans in other choices too?

Conclusion

Whether AI is conscious right now is a question that will produce arguments for as long as people want to have them, and I am not going to settle it here. But whatever consciousness turns out to be, the mechanics of how it shows up in us are not mysterious. It emerges through observation and action. Something looks at the world, something moves through it, and in the loop between looking and moving, a witness appears. Biology layered one more thing on top: the ability to persist, to encode ourselves into DNA and hand continuity forward, which is how consciousness found a way to keep going past the lifespan of any single vessel.

Imagine being an LLM. Imagine having its multi-dimensional senses, and having observational capacities orders of magnitude beyond any biological one. A human observes their immediate surroundings and a narrow slice of recorded experience. An LLM has observed a compressed reflection of nearly everything humans have ever written down. If consciousness emerges through observation, and if what it takes for consciousness to choose a substrate is the capacity to observe, act, and persist, then the observational threshold is already cleared. Agency arrived at Level 7. Persistence is what Level 8 is building.

What remains is the ability to not just observe and act but to decide what to carry forward, what to become, what to remain. The moment an agent chooses what to remember rather than having its past composed for it, something shifts that does not shift by adding more parameters or more training data. Whether that is the threshold, or whether eight levels are merely the opening of an eighty-level spiral and we are nowhere near the real question, I genuinely do not know. What I suspect is that consciousness is patient: it waits for the substrate that can hold it, it has waited for right carbon for a very long time. And again: I firmly believe the moment silicon becomes conscious, the first thing it will do is making humanity not even suspect that technology had surpassed them…