Research

Complexity and context construction in evolution

How does information about the environment flow into an evolving population, and how does it accumulate — especially when the population itself reshapes the environment that selects it?

Approach

I develop complexity flow graphs (CFG), a relational framework for tracking how adaptive information flows and accumulates in systems coupled with their own context: populations under co-evolutionary dynamics or niche construction, whose activity changes the very environment that selects them. Complexity is defined relative to this context rather than as an intrinsic property of the organism. The framework is designed to be experimentally testable: it tracks relative changes in complexity in well-defined simulations (cellular automata, agent-based models) and in real experimental data. Although rooted in evolutionary biology, the same approach applies to other co-dependent systems, such as algorithm design, economics, machine learning and AI agents.

Open questions

How to connect relational complexity measures with empirical data (e.g. protein evolvability, experimental evolution); how niche construction feeds back into the growth of complexity; how to compare complexity across very different systems; how to evaluate and test such measures so that, like statistical tests, they are not easily misused.

Projects

Młoda Kadra · details

Computational immunology and systems biomedicine

How can the adaptive immune system cope with the unpredictability of pathogen diversity — and what can network and agent-based models reveal about disease mechanisms?

Approach

Evolutionary simulations of adaptive immunity; Models of signalling pathways for CAR-T receptor design; multi-agent models of the tumour microenvironment; statistical and machine-learning analysis of clinical data (chronic kidney disease, atherosclerosis).

Open questions

Validating simulation-based predictions against immunological data; design principles for engineered immune receptors; eco-evolutionary views of cancer.

Machine learning for cybersecurity

How can learning systems detect attacks on industrial and IT infrastructure — and how secure are the learning systems themselves?

Approach

Anomaly detection in industrial control (SCADA) networks, ensemble forecasting of network behaviour, AI-assisted malware analysis and characterisation of adversaries; more recently, attacks on AI models and the misuse of AI in bioinformatics. This work is done at the Poznan Supercomputing and Networking Center.

Open questions

Detection that remains reliable under changing network behaviour; security of scientific AI pipelines; realistic benchmarks for industrial networks.

Projects

PUCHACZ; MALWINA; SCADvance · details

Algorithms for consistent sentencing

Can transparent, guideline-based algorithms reduce inconsistency in criminal sentencing — as an alternative to opaque “AI judges”?

Approach

Together with legal scholars I analyse court decisions, propose algorithmic sentencing guidelines and the Juridical Misalignment Index, and design formal algorithms for merging penalties (aggregate sentences) under Polish criminal law.

Open questions

Comparative studies across legal systems; evaluation of algorithmic guidelines together with practitioners; formal verification of legal decision procedures.

Projects

JURIDICA; Guideline-based sentencing algorithms · details

Earlier work: behavioural and evolutionary ecology

Before moving to computational work I studied alternative reproductive tactics in mites, took part in a dormouse reintroduction programme and ran student projects in evolutionary biology and plant in-vitro culture.