Aleksey Zhuravlev

Founder & Chief Scientist

X · LinkedIn · ORCID · GitHub

What We’ve Built

One engine deployed in production on financial markets. Six application domains validated. All from first principles — zero training data.

The Story

No PhD. No lab. No funding. Just 30 years of an open question.

In my first year at university, a professor said: “How consciousness works – nobody has been able to explain it yet.” That sentence opened a gap. It didn’t close for 30 years.

I became a translator, worked in an entirely different field – but the itch remained. Over time, without any deliberate plan, I assembled my own curriculum from the best sources available: Andrew Huberman and Artem Kirsanov on neuroscience, 3Blue1Brown on mathematical intuition, ScienceClick and PBS on physics, Stanislav Drobyshevsky on anthropology – alongside a data science program specializing in NLP, and textbooks on machine learning, complexity theory, and network science.

Each of these was a standalone interest. All of them were quietly building context around that one unresolved question.

Once the theory reached its initial form, I subjected it to a systematic cross-analysis against eight formal academic courses and textbooks — each chosen for a specific lens on the mechanisms BMC describes:

• Robert Sapolsky — Human Behavioral Biology (Stanford): G-layer biology, stress, neuromodulation, hormonal thresholds
• Steven Strogatz — Nonlinear Dynamics and Chaos (Cornell): bistable ODEs, bifurcations, relaxation oscillations, Lyapunov exponents
• Jure Leskovec — CS224W: Machine Learning with Graphs (Stanford): graph neural networks, oversmoothing, community detection, adversarial robustness
• Wulfram Gerstner — Neuronal Dynamics (EPFL): STDP, Hopfield networks, winner-take-all circuits, AdEx adaptation
• Melanie Mitchell — Introduction to Complexity (Santa Fe Institute): edge-of-chaos criticality, Gödel’s incompleteness, effective complexity, scaling laws
• Michael Gazzaniga — The Cognitive Neurosciences: the Interpreter, split-brain, theory of mind, attention networks, ERP components
• David MacKay & James V. Stone — Information Theory, Inference, and Learning: channel capacity, mutual information, rate-distortion theory, belief propagation
• Richard Sutton & Andrew Barto — Reinforcement Learning: An Introduction: TD learning, actor-critic, general value functions, options framework

This cross-analysis produced 73 concrete updates to the theory, 37 new predictions, and 19 new formal terms. Every mechanism in BMC now has an explicit mapping to the relevant academic foundation.

At some point the accumulated density was enough: the gap closed. The result is the theory you’re reading about on this site. About 30 years of open Structural Information Tension – the very mechanism BMC describes as the engine of curiosity. The author is, in a sense, the theory’s first test case.

This is perhaps the first generation where such a path is even possible. A unifying theory of consciousness requires simultaneous knowledge of evolutionary biology, neuroscience, network science, complexity theory, AI, and philosophy of mind. No single academic department teaches all of this. But the combination of open-access lectures, modern textbooks, and years of accumulated context turned out to be enough.

The work is the credential

I have no institutional credentials to offer. What I have instead:

• 5 formal theory documents – Extended Meme Theory, Biomemetics, Network Memetics, AGI Foundations, and Swarm Memetics – totalling 20,000+ lines, fully available on this site
• 149 falsifiable predictions across 23 categories, 34+ testable with existing data
• 9/9 COGITATE retrodiction – the largest empirical test of consciousness theories (2025). BMC explains all results; IIT scored 4/9, GNW 1/9
• 5 competing theories related as limiting cases – IIT, GNW, HOT, AST, and Predictive Processing recovered as limiting cases via subsumption lemmas; each retains independent value
• Retrodiction score of 19.0 across 20 consciousness phenomena (next best: Predictive Processing at 11.0)
• Rust computational engine – 103 gate checks, 300+ unit tests, applied to emergent cognition, drug discovery, symbolic regression, and financial markets
• Emergent language from scratch – agents born with zero knowledge develop 85–97.5% Lewis signaling accuracy without gradient-based communication optimization, up to 533 concepts (DOI: 10.5281/zenodo.19181798)
• Language parasiticity confirmed – 10 survival experiments ($N$=8–150): signal memes optimized for M-fitness, not G-fitness. Separating test: BMC TopSim 0.72 vs REINFORCE 0.60 ($p = 0.011$)
• Cowan’s 4 ± 1 reproduced – working memory capacity $k \approx 4.4$ emerges as evolutionary equilibrium from metabolic cost vs planning benefit (DOI: 10.5281/zenodo.19310012)
• Communication need as replication pressure – reception without expression is indistinguishable from complete isolation ($p = 0.64$, 79 seeds). Expression even into void provides relief (DOI: 10.5281/zenodo.19309824)
• Persistent functional meta-cognition from zero knowledge – agents born with $|V_m|=0$ develop compositional language, cultural convergence, and SMC² in 9/9 seeds at $N=150$ ($p = 0.002$). Cultural Memory identified as the critical mechanism via ablation (DOI: 10.5281/zenodo.19559519)
• 5 published papers on Zenodo with DOIs – see Publications

If the theory holds up to scrutiny – and I invite that scrutiny – it doesn’t matter where it came from.

Get in Touch

• Email a.o.zhuravlev@gmail.com
• ORCID 0009-0008-4127-0740
• GitHub github.com/aozhuravlev