On this page
Digital life: connectomes, continual learning and consciousnessHow to read these references[1] Connectivity: hemibrain[2] From a connectome to a computational model[3] Data for the official brain display: MaleCNS[4] Connectomes and learned control: developing research[5] Evaluation: distributions and failures matter[6–7] Interaction and game integration: AI Arena / NRN[8] NFT identity and transfer[9] Data licenses and shared rights[10] Content addressing is not perpetual deliveryDigital life: connectomes, continual learning and consciousness
Shiu and colleagues’ Nature study supports experimentally tested predictions for some sensorimotor processing. Eon’s embodied demonstration also discloses engineered interfaces and controllers. Movement alone does not establish lifelong learning.
DreamerV3 and Voyager offer engineering methods for world models and accumulated skills; those results do not come from an original fly connectome. FEFE evaluates connectome and augmented cognition separately, including pretrained knowledge and assistance.
Butlin and colleagues’ review of AI consciousness indicators emphasizes uncertainty. Functional states, expressions and model self-reports do not establish subjective consciousness; sample volume does not guarantee every human ability.
How to read these references
Papers and datasets establish what researchers have studied. Protocols explain what they can record. Other products offer interaction references. FEFE’s learning, testing, asset and application designs are our proposals, requiring independent implementation and validation. These sources do not establish that FEFE has general intelligence, cross-game ability or a trained-asset market.
[1] Connectivity: hemibrain
Scheffer et al., 2020, A connectome and analysis of the adult Drosophila central brain, eLife 9:e57443. This work and Janelia’s hemibrain dataset provide a traceable source for fruit-fly brain connectivity.
The current FEFE player experience uses a hemibrain MB+CX subcircuit. A wiring table is not a trained game policy. Simulation rules, stimuli and action readouts are FEFE engineering choices. Dataset versions and attribution follow the provider’s requirements.
[2] From a connectome to a computational model
Shiu et al., 2024, A Drosophila computational brain model reveals sensorimotor processing, Nature, DOI 10.1038/s41586-024-07763-9. The work builds a connectome-based model to investigate specific sensorimotor processing and compares predictions with experiments.
It supports connectivity as a foundation for testable computational models. It does not demonstrate human infant cognition, consciousness or arbitrary game skills in a fly model, and does not validate FEFE’s implementation.
[3] Data for the official brain display: MaleCNS
MaleCNS v1.0, 2026. Contributors include HHMI Janelia FlyEM, the Cambridge Connectomics Group and Google Research. It provides an adult male fruit-fly central nervous system connectome. The official site lists downloads, versions and a CC BY license.
The official whole-brain display and the player’s hemibrain subcircuit use different configurations. The homepage background currently uses procedurally wired spiking simulation arranged as flies; it is not presented as a complete MaleCNS biological reconstruction.
[4] Connectomes and learned control: developing research
Jin, Zhu, Zhang and Sui, 2026, Whole-Brain Connectomic Graph Model Enables Whole-Body Locomotion Control in Fruit Fly, arXiv:2602.17997v3. The cited version was revised on June 14, 2026 and is a preprint.
The study uses a connectomic graph controller with deep reinforcement learning for simulated fly locomotion. It informs research into trainable control in specific environments. It does not imply that FEFE’s odour-learning model can already play Gomoku or that one model can enter any game directly.
[5] Evaluation: distributions and failures matter
Agarwal et al., 2021, Deep Reinforcement Learning at the Edge of the Statistical Precipice, NeurIPS 2021. The work examines uncertainty in reinforcement-learning evaluation and more reliable statistical reporting.
Our design draws on this by fixing tasks and runtime conditions, using independent repetitions, reporting sample sizes and uncertainty, and preserving failures. This is a methodological reference, not an endorsement of any fly’s win rate.
[6–7] Interaction and game integration: AI Arena / NRN
[6] NRN’s official AI Arena introduction describes human demonstrations influencing a character’s policy, followed by autonomous competition. It is a precedent for a player-trains, character-competes interaction. The project’s account is not independent evidence of commercial success or proof that FEFE has equivalent features.
[7] NRN’s Basic Integration documentation explains game-state and action translation. Our inference for FEFE is that DGames can share identity while still needing individual input, output and skill-compatibility definitions.
[8] NFT identity and transfer
Entriken et al., 2018, ERC-721: Non-Fungible Token Standard. It defines identity, holding and transfer interfaces for non-fungible tokens, making it a candidate foundation for an individual fly’s identity.
Model files, service delivery and intellectual-property rights need additional arrangements. A unique NFT holder is not necessarily the only person with a given skill. FEFE’s eventual contracts and service agreements will be defined by the published implementation and terms.
The US Copyright Office and USPTO’s 2024 joint report, Non-Fungible Tokens and Intellectual Property, discusses transfers of NFTs and associated rights from page 23. It informs license disclosure here, rather than establishing permission to operate in any jurisdiction.
US Copyright Office / USPTO · NFT and IP report · 2024 (PDF) ↗
[9] Data licenses and shared rights
Creative Commons Attribution 4.0 permits sharing and adaptation, including commercial use, subject to attribution and other conditions. Compliant licenses are not automatically revoked when an original is sold. FEFE must check data, code, models and artwork separately.
The selected Janelia datasets and other connectome projects do not share a universal license. For example, FlyWire’s public data guidance includes a noncommercial restriction; fruit-fly research is not automatically available for commercial products.
[10] Content addressing is not perpetual delivery
The IPFS Persistence documentation explains that lasting availability needs storage arrangements. A content identifier can verify file identity; it does not ensure the file remains retrievable.
Continuity must therefore specify model storage, runtime services, backups and export scope. Writing a hash on-chain is not a promise that a fly can run forever.