R / 01
Seeing inside neural networks
How do internal model dynamics shape an answer? We aim to develop tools for inspecting activations and testing causal hypotheses through controlled interventions.
Independent research initiativeNEURAL NETWORKS × PHILOSOPHY
Intelligence
beyond
boundaries.
We explore how neural networks make meaning, how ideas emerge, and where a cognitive system begins and ends.
NEURAL NETWORKS / PHILOSOPHY / MATHEMATICSFrom concepts to experiments.
01The laboratory
What if intelligence is something that happens between a model and its world?
Leaky Intelligence Lab brings together philosophical inquiry, mathematical ideas and experiments with neural networks. Our research programme connects questions about meaning and agency with the internal dynamics of language models and the systems built around them.
Why “leaky”?
For us, the name expresses a research question: how do the boundaries of cognition change through language, memory, tools and interaction? We study intelligence as an open, relational process.
02 / Research programme
Four connected questions.
Our starting points for investigation: hypotheses to test, methods to develop, connections to explore.
R / 02
The emergence of new ideas
When does a change of context lead to a genuinely different idea? We investigate creativity, transitions between meanings, and ways to guide open-weight models without initial fine-tuning.
R / 03
Where cognition begins and ends
What changes when a language model becomes part of a system with memory, tools, people and other agents? We examine how agency and meaning take shape through these relations.
R / 04
A geometry of meaning
Can geometry reveal structure that ordinary distances miss? We explore p-adic and hierarchical descriptions, and test whether ideas such as holonomy can become useful experimental tools.
01 — Formulate a question
Make philosophical assumptions explicit and turn them into testable hypotheses.
02 — Build an experiment
Start with small open-weight models, inspect their dynamics, and intervene.
03 — Revisit the concept
Let the findings reshape the question, including when the hypothesis fails.
03 / People
Different disciplines. Shared curiosity.
Maxim Yanukovich
AI research & philosophy
Head of the Arteus AI laboratory and a researcher pursuing a Candidate of Sciences degree at the Faculty of Philosophy, Lomonosov Moscow State University. His work connects applied machine learning with the philosophy of cognition, meaning and the observer.
LinkedIn profileRodion Karneev
Philosophy of science & technology
Candidate of Sciences in Philosophy and Research Fellow at the Institute of Philosophy, Russian Academy of Sciences. His research focuses on subjectivity, ontology, and the philosophy of science and technology, including the reassembly of the subject in the age of neural networks.
Academic profile & publications04 / Start a conversation
Good questions deserve company.
Interested in neural network interpretability, machine creativity, or the philosophy of AI? Get in touch to discuss research, experiments and collaboration.
m.yanukovich@gmail.comResearch enquiries · Maxim Yanukovich