Neurocrypt Genesis: A Platform for Artificial Abiogenesis and Emergent Computation

Neurocrypt Genesis is a Rust-based artificial life simulation environment and genetic programming framework designed to investigate emergent computation through evolutionary dynamics. By evolving self-replicating, ATP-regulated digital agents within a controlled virtual ecosystem, Neurocrypt serves as an experimental testbed for digital abiogenesis, open-ended evolution, and stochastic-to-structured intelligence transitions. Its conceptual underpinnings draw from synthetic biology, origin-of-life research, and computational neuroscience.

Theoretical Foundation and Research Aims

Can adaptive computation emerge from minimal initial structure under selection pressure alone?

This inquiry mirrors the theoretical premises of abiogenesis, positing that self-organizing systems capable of memory, decision-making, and information processing can arise from a stochastic genotype space through mutation, reproduction, and environmental fitness gradients.

The system is architected to eliminate algorithmic scaffolding or hardcoded logic, instead relying solely on mutable DNA codons and selective ATP allocation to drive digital ontogenesis. Neurocrypt seeks to elucidate the computational properties of evolutionary substrates and probe the limits of bottom-up intelligence.

Microcrypt Protocol and Evolutionary Roadmap

The most refined implementation of Neurocrypt to date is the Microcrypt protocol, characterized by its lightweight genome schema, parallel interpreter design, and deterministic mutation interface. Forthcoming iterations will expand the DNA schema to define topologies of dynamic tensor networks, embedding architectural plasticity into the genome. This fusion of neurogenetics and differentiable computation is envisioned to produce elastic, neuromorphic agents capable of self-modifying inductive structures.

Core System Architecture

Genomic Encoding and Instruction Set

Digital agents ("cells") in Neurocrypt are defined by a linear genome composed of codons, each codon being a fixed-size byte triplet ([u8; 3]). These encode low-level executable instructions parsed by a virtual machine-like interpreter. The Microcrypt instruction set includes:

Load = 0 // regs[arg1] = mem[arg2]  [ATP: 1]

Store = 1  // mem[arg2] = regs[arg1]  [ATP: 1]

Set = 2  // regs[arg1] = arg2 [ATP: 1]

Add = 3  // regs[arg1] += arg2  [ATP: 1]

Sub = 4  // regs[arg1] -= arg2  [ATP: 1]

JmpIfEq = 5  // if regs[arg1] == arg2: PC+=2  [ATP: 2]

JmpIfNeq = 6 // if regs[arg1] != arg2: PC+=2  [ATP: 2]

Figure: Mircocrypt's Genome instruction set

Codons are executed sequentially by the dna_interpreter, a parallelized execution engine leveraging the Rayon library to enable multithreaded evaluation across cell populations. The interpreter increments the program counter (PC) linearly, except when conditional jumps alter flow.

Metabolic Model and Selective Pressure

Each cell maintains a bounded ATP budget representing computational energy. Instruction execution consumes ATP proportional to complexity. Upon exhaustion, the cell is culled from the simulation. Viable cells, defined by their productive interaction with the environment, accumulate ATP rewards and are eligible to reproduce.

Reproduction is asexual and entails codon-level genome duplication with stochastic mutation. This mechanism operationalizes Darwinian selection, favoring genotypes encoding adaptive behaviors. ATP thresholds restrict reproductive access to sufficiently performant agents.

Register and Memory Architecture

To enable complex behavior and inter-agent coordination, each cell possesses:

• Registers: A 4-element array of 8-bit integers ([u8; 4]) for temporary data storage and computational state.

• Dish Memory: A shared addressable memory space consisting of Slot structs containing a value: u8 and an owner_id. The memory pool is globally bounded by mem_max, constraining state persistence and facilitating generalization.

This dual-memory structure enables both ephemeral internal computation and semi-persistent environmental signaling, which are prerequisites for emergent cooperation and distributed problem-solving.

Serialization and Reproducibility

Each cell’s complete state, including genome, registers, ATP, and metadata, can be serialized in either JSON (for interpretability) or binary (for storage efficiency). This facilitates longitudinal experimentation, snapshot analysis, and reproducible digital evolution studies.

Mutation Engine: Controlled Genotypic Perturbation

Genomic mutation is governed by two tunable hyperparameters:

• mut_prob: the stochastic probability of mutation per reproductive event.

• mut_rate: the number of discrete mutations applied if mutation occurs.

The mutation operator draws from a set of defined transformations:

• Byte Flip: Bitwise inversion of a codon element

• Increment/Decrement: Arithmetic mutation of codon values

• Swap: Intra-genomic codon position exchange

• Insert: Addition of a novel random codon

• Delete: Removal of a codon

These operations are applied pseudo-randomly and can be restricted to high-ATP cells, ensuring that only fit agents are eligible to mutate, thereby preserving lineage viability while maintaining genetic diversity.

Example: Neurocrypt in Its Simplest Instantiation

(Zero-State Benchmark via Microcrypt)

This example illustrates Neurocrypt's foundational operating mode, specifically the Zero‑State Benchmark implemented within the Microcrypt protocol. The demonstration, accessible at Z-Cab, exemplifies the setup, execution, and evolutionary dynamics in the system’s most stripped-down configuration.

Highlights of this benchmark:

• Cells start from nonsense DNA

• Survival and replication are solely guided by ATP-based reward signals.

• This zero-state paradigm captures emergent computation free from scaffolded behaviors, allowing intelligence to arise bottom-up under Darwinian pressure.

Entropic Limitations and Proposed Extensions

A principal limitation of the current implementation is its dependence on pseudo-random number generators (PRNGs) for mutation entropy. Over extensive runs, PRNGs exhibit finite periodicity and insufficient entropy gradients. To address this, we propose future integration of analog entropy sources such as:

• Ambient radiation sampling (e.g., Geiger counters)

• Photonic variability (e.g., UV photodiodes)

• Atmospheric noise (e.g., radio wave fluctuation)

These additions would enhance mutation unpredictability and provide a more biologically plausible stochastic substrate, critical for simulating long-term open-ended evolution.

---

Citation: Lazic, Hadrian. Neurocrypt Genesis: A Platform for Artificial Abiogenesis and Emergent Computation. 2025. GitHub repository: Github Repo

---