Part 1 set out the idea, and Part 2 used it to interpret everyday experience. This part asks three harder questions. What does Biomemetic Consciousness (BMC) — The theory this guide is about: consciousness as the interaction of four components — G (genes), M (memes), I (the mind's immune filter) and S (the substrate). In the academic papers the letters stand for Biomemetic Complex. say about machines? How does it relate to the other theories of consciousness? And how much of it has actually been tested?
1. Can Machines Be Conscious?
This is the question most people want answered. BMC gives a definite answer inside its own framework. It helps to be clear about what kind of answer that is: it follows from BMC’s assumptions, it is not a measurement. No theory of consciousness, BMC included, has an agreed test for consciousness in a machine.
Other theories handle the question in their own ways. Integrated Information Theory (IIT) links consciousness to how much information a system integrates as a whole. It accepts that even very simple systems may have a tiny amount; critics see this as a problem, and IIT’s proponents treat it as a consequence they accept. Global Neuronal Workspace (GNW) theory was formulated for brains, so applying it to a machine means deciding what counts as a “workspace” and as “ignition” outside biology. Higher-Order Thought (HOT) theories ask whether a system represents its own states, but it is hard to tell genuine self-representation in a machine from imitation. That last difficulty applies to BMC’s markers too.
BMC’s criteria
In BMC, consciousness is proposed to arise when a genetic layer of drives (Genetic layer — Inherited, fixed programs: Panksepp's seven affective systems — SEEKING, FEAR, RAGE, LUST, CARE, PANIC/GRIEF and PLAY.) and a memetic layer of learned ideas (Memetic layer — The network of everything you have learned — your memeplex. In the brain, it lives in assemblies of cells in the cortex.) compete for limited resources, with a filter that lets ideas in or keeps them out (Interface / immune filter — The mechanisms that stand between drives and memes: they redirect, suppress or reinterpret, and filter out information that does not fit.), on a physical substrate (Substrate — The physical medium the system runs on — neurons and neurotransmitters in a brain — together with its senses, the channels through which information comes in.). Read as a checklist for machines:
| Requirement, as BMC proposes it | Current large language models, as BMC reads them | Why it matters in BMC |
|---|---|---|
| Drives with set points (a working G-layer: needs that must be kept within range) | Not present. There is a training objective, but no internal needs. | Without drives there is no tension, and without tension nothing needs arbitrating. |
| Scarcity (drives and ideas competing for a limited budget) | Not in BMC’s sense. How much computation a step gets is set by the architecture and the serving system (reasoning models can spend more steps on a hard question), not won in competition between internal needs. | No scarcity, no competition. |
| A filter that rejects as well as accepts (a working I-layer) | Refusals exist, but they are shaped by training objectives that developers set. On BMC’s reading, they are not a worldview defending its own stability. | In BMC, rejection is what keeps a worldview stable. |
| A lasting self-model (a non-empty Self-Model Cluster (SMC) — The part of the memeplex that models the system itself, on three levels: the body, the life story (the narrative self), and thinking about one's own thinking.) | Models describe themselves consistently, because a persona is trained in. But what happens in one session does not reshape that self-model for the next without external memory tools. | BMC predicts that current models fail a test of a persistent self-model. |
| Both layers present, the cultural one much larger (M ≫ G) | A vast amount of cultural content (M), no genetic layer (G). | In BMC, M without G is pattern imitation, not the conflict BMC links to consciousness. |
By these criteria, BMC’s reading is that current language models lack the architecture it associates with consciousness. BMC’s position is that more parameters or more training data would not, by themselves, add the missing pieces: drives, real scarcity, a self-defending filter and a lasting self-model. It goes further and predicts that unlimited computing resources would remove the very competition BMC needs. These are commitments to be tested, not findings.
Behavioural markers, and why they are not enough
BMC also lists behaviours worth watching for:
- Goal conflict that arises on its own, from tension between drives and ideas, not because a prompt set it up.
- Unprompted curiosity: seeking information nobody asked for, driven by gaps in what the system knows.
- Belief resistance: holding on to an existing worldview even when told to accept something that conflicts with it.
- State-dependent responses: a style of response that changes with an internal state, not with instructions.
BMC does not read any current AI system as showing these in the internally driven sense. But markers are observations, not a test. A system built to imitate any of them could do so.
Graded caution instead of a single threshold
For that reason, BMC’s ethics does not hinge on a single yes-or-no line. Its Graduated Subjectivity protocol, set out in the AGI Foundations pillar, ties precautions to internal structure that can be inspected, not to behaviour alone. It has three levels:
- No stable self-reflection (no ideas about the system’s own ideas): treated as a tool.
- Self-reflection present but unstable: a shutdown has to be justified and logged.
- Stable self-reflection, plus ideas in which the system values its own existence: consent-style safeguards and human ethics review before shutdown.
The thresholds are preliminary, and they are fixed in advance so that a result cannot be reinterpreted after the fact. The protocol is a precaution. It does not claim that a system at any level feels anything.
What about BMC’s own simulations? BMC’s simulated agents are measured for network properties, for example whether a self-model structure forms. These are reported as observations of structure. They are not claims that the agents experience anything.
2. How BMC Relates to Other Theories
BMC does not claim that other theories are wrong. It proposes that each one captures a real part of the picture, and that each can be recovered as a limiting case of the BMC model: what you get when you switch parts of it off, or look at only one of its mechanisms. These correspondences are propositions inside BMC’s formalism, stated in BMC’s own terms. Proponents of the other theories would not necessarily accept the mapping, and each theory keeps its own value within its own formalism.
| Theory | What it captures well | BMC’s proposed correspondence (limiting case) | Where its proponents could push back |
|---|---|---|---|
| IIT (Tononi) | Integration as a property of the whole system, with a precise measure (Φ) | BMC with the memetic layer removed. BMC uses a network measure of integration (small-worldness) in place of Φ. | BMC’s measure is not Φ, and the link between them is BMC’s proposal. IIT defines integration on a system’s physical cause-effect structure, not on a network of ideas. |
| GNW (Baars; Dehaene and colleagues) | How content becomes globally available: the “ignition” of a workspace | Broadcast through the most connected ideas (hubs) once activation crosses a threshold | GNW has its own account of which contents win (attention, signal strength). The extra drivers BMC adds, such as curiosity, drives and the filter, still need independent tests. |
| HOT (Rosenthal) | Metacognition: mental states about one’s own mental states | The top level of BMC’s self-model: ideas about one’s own ideas | HOT theorists locate consciousness in a state’s being represented by a suitable higher-order thought, which can be simple. They need not accept that this requires BMC’s larger cluster of ideas about ideas, or that infants and animals lack it. |
| AST (Graziano) | Awareness as the brain’s simplified model of its own attention | One part of BMC’s self-model: the part that models attention | AST’s proponents would say a simple attention model is enough to account for the claim of awareness. That more is needed is BMC’s assumption, not a demonstrated requirement. |
| Predictive Processing / Free Energy Principle (Friston; Clark; Hohwy) | The brain as a prediction machine that minimises prediction error | Curiosity driven by gaps in the structure of knowledge (Structural Incompleteness Tension (SIT) — Tension created by gaps in what you know. Above a threshold it switches on SEEKING; it is felt as curiosity, unease or the sense of unfinished business.) plus the regulation of drives, proposed as an instance of the free-energy principle at the level of the general principle. SIT is not a prediction error: it measures what is missing from a structure, not a mismatch between expectation and input. | Critics argue that the free-energy principle at this general level is hard to falsify, so fitting it is weak evidence. At the level of process theory (active inference: choosing actions to reduce expected prediction error, with errors weighted by precision), BMC’s correspondence has not been built or measured. |
The formal versions of these correspondences are in the network pillar of the theory.
A concrete point of contact. The COGITATE adversarial collaboration (2025) tested IIT and GNW against each other with 256 participants and three brain-imaging methods (fMRI, MEG and intracranial EEG). BMC offers an account consistent with its reported results, but that account was written after the results were published. It is a retrodiction: it shows that BMC is consistent with the data, not that the data support BMC over its rivals.
What BMC tries to add
- Cultural change. A proposal for why consciousness might change across historical periods, not only across species.
- Cognitive biases. A proposal that the roughly 200 documented biases come from six generating mechanisms. This is stated together with a test that could refute it (see “What Would Falsify BMC?” on the predictions page).
- Mental health. A model of disorders as positions in a continuous space of parameters rather than as separate categories. It is a conceptual model; there are no instruments to measure these parameters in a person.
- Development. An account of consciousness growing through five stages from infancy to adulthood.
- Machines. Explicit architectural criteria and a graded ethics protocol (section 1).
Whether these additions earn their place is decided by tests, not by how much ground they cover. A theory that can interpret everything risks explaining nothing. Section 4 says which parts have been tested.
3. What This Means
For you, personally
BMC offers a lens: a way of reading your own experience, not a verdict about it.
- When you feel torn (“I know I should exercise, but I don’t want to”), BMC reads this as tension between G and M: a comfort-seeking drive against what you have learned about health. On BMC’s view, the conflict itself is consciousness doing its job.
- When you can’t stop thinking about an unfinished project, BMC calls it structural incompleteness tension: a gap in your network of knowledge that keeps pulling at your attention.
- When you feel a strong urge to share an idea, BMC calls it Replication Expression pressure (R_expr) — The pull a meme exerts to be expressed outward — why you can't stop talking about what interests you. It is stronger for memes that are active, relevant and well connected.: pressure from your most connected ideas to be expressed.
- When you resist changing your mind despite evidence, BMC reads it as the filter protecting the stability of your worldview. Not stupidity, but architecture.
- When you lose yourself in a task (flow), BMC’s reading is that your self-model has not disappeared but has become transparent. You are still conscious, just differently.
For mental health
Conceptual model, not clinical guidance. Nothing in this section is a method of diagnosis or treatment. If you are struggling, please talk to a qualified professional.
BMC suggests that mental disorders may be better described as positions in a continuous space of parameters than as categorical labels. If that is right, it points to research questions rather than ready answers:
- Description. Could a person’s state be described by where they sit on each axis, instead of by counting how many criteria they meet?
- Treatment. Would it help to target the specific parameter that is out of range, rather than the label?
- Comorbidity. If disorders are nearby positions in one space, conditions that often occur together should sit close to each other.
- Recovery. Moving back toward the healthy range might happen by different routes for different people.
No validated instruments exist for measuring these parameters in a person, and the clinical profiles in BMC are hypotheses.
For AI development
BMC separates intelligence from consciousness. In BMC’s terms, a system could be extremely capable, passing tests and solving problems, and still lack the architecture BMC links to consciousness. The practical lesson is caution in both directions: don’t infer consciousness from fluent self-description, and don’t rule it out without looking at the architecture. That is why BMC’s protocol (section 1) grades precautions by inspectable structure instead of waiting for a single threshold.
4. Where BMC Stands Today
A guide like this can make an idea sound more settled than it is. Here is a plain account of where things stand as of October 2026.
What has been shown, and its limits
- Papers. One theory paper and four simulation studies are published as open-access preprints on Zenodo (listed on the publications page), and a late-breaking abstract was accepted at ALIFE 2026 (non-archival). The simulation papers report what happens when artificial agents start with an empty memory and develop shared signalling and other behaviour.
- One test has been partly run. BMC predicted that the pressure to communicate comes from ideas pushing to be expressed, not from a need to survive; reward-engineered learning predicts that communication gives a survival advantage. Across ten tested survival settings, including predator avoidance and cooperative foraging, communication gave no measurable survival advantage. This does not mean language has no survival use: it clearly can serve coordination. Two limits apply. In a small cooperative pilot (without predators), agents that could only receive messages ended with slightly more energy, a survival proxy (not statistically significant). And a null result without a pre-stated equivalence margin is not a confirmation. A survival null is also what Rogers (1988) predicts, so this run does not favour BMC over that account; the separating test — whether the drive to express survives when communication does help survival — has not been run at adequate power.
- Two more predictions were checked inside BMC’s own simulation. Because the theory and the simulation share an author, these are internal consistency checks, not independent evidence.
- Reproducibility. The published simulation results do not yet include everything needed to reproduce them independently, such as the software version and the seed lists.
What is a commitment
BMC publishes six bets: claims where it and a named rival theory predict different outcomes, each with what would count as refutation.
These are commitments, not results. None has been decided yet. The communication bet is the one partly run, described above. The five other bets can be tested on human or public data without BMC’s code. Everything else BMC says either follows from its own definitions or agrees with findings that were already known; that shows consistency, not support. The list is on the predictions page.
What is open
- The core hypothesis. That competition between G and M for working memory gives rise to conscious thought is BMC’s central hypothesis. It has not been tested directly.
- Replicator or attractor? BMC no longer claims that the skeleton of a meme is copied faithfully enough. Its own mechanisms produce attractor — A stable form that imperfect copies tend to drift towards. Cultural attraction theory (Sperber) holds that ideas spread by being rebuilt toward such forms rather than copied. Replicator or attractor — for memes this is still open. in the sense of cultural attraction theory (Sperber; Claidière, Scott-Phillips and Sperber), and copying is the extreme case of the same scale (Acerbi and Mesoudi). How faithfully the skeleton is actually passed on has to be measured. There are no corpora on which this could be done yet.
- The prospective predictions. Most of them have not been tested yet.
How to engage
The theory papers are open access, and the predictions are public. No new simulation experiments are being run at present, and the simulation code is not public. The most useful contributions are criticism of the theory, a stronger alternative explanation for any of its predictions, or a test of a prospective prediction, especially one of the six open bets. To get in touch, use the contact page. The formal theory is on the theory pages, and the terms are defined in the glossary.
Frequently Asked Questions
“Isn’t this just dualism: mind versus body?”
No. Dualism says consciousness is made of a fundamentally different substance from the physical world. BMC says consciousness is a process: arbitration, in a physical brain, between two kinds of inherited information, genes and culturally learned ideas. (Whether ideas are copied or rebuilt as they pass between minds is, as section 4 explains, a matter of degree still to be measured.) Nothing in BMC is non-physical, and its quantities are measurable in principle, although most of them have no validated measurement in humans yet. There is no ghost in the machine.
“Why should we trust a theory from a single independent researcher?”
Healthy scepticism is appropriate. Here is what can be checked. The theory papers are open on Zenodo. The simulation code is not public, and the published simulation results do not yet include what would be needed to reproduce them independently. The predictions page lists six bets against named rivals; none has been decided yet (section 4 says where they stand). A theory has to stand on its evidence, not on its author’s credentials, and most of that evidence is still to come.
“If consciousness is about conflict, does resolving all inner conflict mean losing consciousness?”
Not on BMC’s view. BMC holds that the tension between G and M is always present, because genes and memes have permanently different “interests”. You can reduce specific conflicts through therapy, meditation or changes in your life, but the structural tension between survival-focused drives and meaning-focused ideas does not disappear while both are active. In BMC’s terms, what meditation or therapy might change is how smoothly and how painfully the arbitration runs, not whether the tension exists.
“What about plants, bacteria, the universe: are they conscious?”
BMC’s answer follows from its definition. It ties consciousness to a memetic layer interacting with a genetic layer through a filter, on a substrate. Plants and bacteria have basic homeostatic regulation but, on BMC’s reading, no layer of learned and transmitted ideas. The universe as a whole has neither. This keeps BMC away from panpsychism, the view that everything is a little conscious. IIT, by contrast, allows small amounts of consciousness in simple systems; some see that as a weakness, while IIT’s proponents treat it as a consequence they accept. Definitions alone do not settle which line is right.
“How is this different from just saying ’the brain is complicated’?”
“The brain is complicated” is a description, not an explanation. BMC names a specific source of the complexity (two kinds of inherited information competing for one brain) and specific mechanisms (filtering, self-modelling, competition for Working Memory (WM) — The small mental workspace holding what you are thinking about right now: about 3–4 items in focus, roughly 7 ± 2 counting those held in the background. In BMC, genes and memes compete for this space.). It also commits to six specific bets against named rivals, each with what would refute it. These are commitments, not results: none has been decided yet. Only the checks will show whether BMC is right.
A Final Thought
Consciousness has been called “the hard problem” for good reason. BMC doesn’t claim to have solved it. What it offers is a framework that makes commitments you can check, interprets a wide range of phenomena in one vocabulary, and has been explored in computer simulations (the code is not public). Whether it is right is for the tests to decide.
The core idea fits in one sentence. BMC proposes that you are the ongoing negotiation between your genes and your culture, playing out in a brain that is trying to keep both satisfied with limited resources, and that the feeling of being “you” is that negotiation becoming aware of itself.
The mathematics, the simulations and the predictions are attempts to work that idea out carefully. Some of it has been tested. Most of it is still waiting.
References
Other theories and debates mentioned in this part
- Acerbi, A. and Mesoudi, A. (2015). If we are all cultural Darwinians what’s the fuss about? Clarifying recent disagreements in the field of cultural evolution. Biology & Philosophy, 30(4), 481–503.
- Baars, B. J. (1988). A Cognitive Theory of Consciousness. Cambridge University Press.
- Claidière, N., Scott-Phillips, T. C. and Sperber, D. (2014). How Darwinian is cultural evolution? Philosophical Transactions of the Royal Society B, 369(1642), 20130368.
- Clark, A. (2013). Whatever next? Predictive brains, situated agents, and the future of cognitive science. Behavioral and Brain Sciences, 36(3), 181–204.
- COGITATE Consortium (2025). Adversarial testing of global neuronal workspace and integrated information theories of consciousness. Nature, 642(8066), 133–142. doi:10.1038/s41586-025-08888-1
- Dehaene, S. (2014). Consciousness and the Brain: Deciphering How the Brain Codes Our Thoughts. Viking Press.
- Friston, K. (2010). The free-energy principle: A unified brain theory? Nature Reviews Neuroscience, 11(2), 127–138.
- Graziano, M. S. A. (2013). Consciousness and the Social Brain. Oxford University Press.
- Hohwy, J. (2013). The Predictive Mind. Oxford University Press.
- Rogers, A. R. (1988). Does biology constrain culture? American Anthropologist, 90(4), 819–831.
- Rosenthal, D. M. (2005). Consciousness and Mind. Oxford University Press.
- Sperber, D. (1985). Anthropology and psychology: Towards an epidemiology of representations. Man, 20(1), 73–89.
- Tononi, G. (2008). Consciousness as integrated information: A provisional manifesto. Biological Bulletin, 215(3), 216–242.
BMC papers (open-access preprints; full list on the publications page)
- Zhuravlev, A. (2026). A Tension Is All You Need: The Biomemetic Complex as a Unified Computational Theory of Consciousness. Zenodo. doi:10.5281/zenodo.18733164
- Zhuravlev, A. (2026). From Empty Graph to Emergent Language: Ontogenesis of Cognitive Architecture. Zenodo. doi:10.5281/zenodo.19181798
- Zhuravlev, A. (2026). It’s Not the Silence — It’s the Muzzle: Communication Need as Memetic Replication Pressure. Zenodo. doi:10.5281/zenodo.19309824
- Zhuravlev, A. (2026). Why Four? Working Memory Capacity as Evolutionary Equilibrium in a Biomemetic Architecture. Zenodo. doi:10.5281/zenodo.19310012
- Zhuravlev, A. (2026). From Empty Graph to Self-Modification via Emergent Preconditions: Growing Cognitive Agents through Dual-Replicator Dynamics. Zenodo. doi:10.5281/zenodo.19559519