Artificial Life · 1997 – Present Evolutionary Computation & Communicative Agency

Artificial Life, Open-Ended Evolution & Communicative Minds

A 30-year thread investigating how digital organisms, virtual morphologies, self-referential prompts, and communicative agents evolve—stretching from 1990s cybernetic biochemistries (Creatures) and mixed-reality TV physics (BAMZOOKi) to modern evolutionary AI at Google DeepMind.

01 // Google DeepMind Research & Publications

Morphology Evolution, Cognitive Scaffolding & Self-Referential AI

Papers and experimental systems exploring how evolution, external representations, and intrinsic communicative drives shape intelligence.

Cognitive Science, Agency & Consciousness 2024 – 2025

External Representations & "Wanting to Be Understood"

Why do biological and artificial agents externalize internal states into persistent environmental artifacts—drawings, symbols, and artworks—and re-perceive them as cognitive scaffolding? With Chrisantha Fernando et al., this research stream investigates both the origin of external representations and intrinsic communicative motivation, where agents actively model whether their partner has understood them and interactively repair misunderstandings.

Adaptive Behavior (2024): External Representations ↗ Wanting to be Understood (arXiv:2504.06611) ↗ Meta-Problem of Consciousness (arXiv:2506.12086) ↗

3D Morphology & Brain Co-Evolution DeepMind · MuJoCo

Evolving 3D Virtual Creatures & Sumo Wrestlers (EvoBorg)

Inspired by Karl Sims' seminal 1994 evolved virtual creatures and Dylan's own Creatures and BAMZOOKi heritage: co-evolving both the physical 3D bodies (directed acyclic graph genotypes of limbs, joints, and sensors in MuJoCo physics simulations) and neural network controllers using asynchronous distributed genetic algorithms as virtual creatures learn to wrestle each other in competitive sumo arenas.

Co-Evolution of Body & Brain MuJoCo Competitive Physics

ICML 2024 & Gemini 1.0 Tech Report 2023 – 2024

Promptbreeder: Self-Referential Self-Improvement via Prompt Evolution

A general-purpose self-referential evolutionary mechanism where a population of LLMs evolves both task-prompts and the mutation-prompts that govern how prompts mutate—outperforming hand-engineered chain-of-thought strategies on reasoning benchmarks and contributing to mathematical reasoning in the Gemini 1.0 Technical Report.

Promptbreeder (arXiv:2309.16797) ↗ Gemini 1.0 Report (arXiv:2312.11805) ↗

Continual & Transfer Learning 2017

PathNet: Evolution Channels Gradient Descent in Super Neural Networks

A foundational DeepMind study demonstrating how microbial genetic algorithms can evolve active modular pathways through a giant neural network while gradient descent trains weights along those pathways—enabling continual transfer learning across Atari and Labyrinth games by freezing winning pathways and evolving new ones.

PathNet Paper (arXiv:1701.08734) ↗

NeurIPS 2023 · Multimodal Benchmark 2023

Perception Test: Diagnostic Benchmark for Multimodal Video Models

An open-source diagnostic video benchmark designed to probe real-world perception, memory, physics intuition, counterfactual reasoning, and semantic abstraction across multimodal foundation models using real-world filmed scenarios inspired by developmental psychology.

Paper (arXiv:2305.13786) ↗ GitHub Repo ↗

02 // Pioneering Digital Life, Mixed-Reality TV & Embedded Sims

Creatures (1997), BAMZOOKi (2003) & A-Life on Pi

Before foundation models, Dylan spent over a decade building two iconic British cult classics of artificial life and physics-based virtual creatures.

CyberLife Technology / Creature Labs 1997 – 2003

The Creatures Series (Creatures 2, 3 & Docking Station)

Served as Senior Researcher & Head of Artificial Life on the BAFTA-nominated Creatures artificial-life series. Engineered heterogeneous neural network brains, artificial biochemistries (hormones, drives, chemo-receptors, and emitters), and mutable digital chromosomes so "Norns" could learn language, forage, and evolve across generations on home PCs.

Neural Network Brains Artificial Biochemistry Digital Genetics

BBC / Gameware Development 2003 – 2009

BAMZOOKi (BBC Mixed-Reality TV & Zook Kit)

Technical & Creative Product Lead across four series of the BBC's BAFTA-nominated mixed-reality children's television show. Children downloaded the free 3D physics and Inverse Kinematics "Zook Kit" to engineer their own virtual creatures at home, then brought them into the BBC studio to race, hurdle, and sumo-wrestle on a physical table via real-time mixed-reality broadcast rendering.

4 BBC TV Series Real-Time Mixed Reality Physics & IK Creature Kit

Raspberry Pi & SenseHAT 2016 – Present

A-Life on Pi

Lo-fi, open-ended artificial-life simulations, continuous cellular automata, and emergent pixel ecosystems running autonomously on tiny Raspberry Pi boards and 8×8 LED matrices.

Visit alifeonpi.com ↗