superorganism-experiment: creating our own society

We are creating our own society. A place citizens have FULL control, have their own MONEY, have AI that serves THEM, and CONTROL together. Unstoppable by design, self-replicating, self-hosted, self-evolving, and human oversight with democratic governance. Well, that is our Utopian dream! It now runs and empowers a network of seedboxes.

Our work contains several novelties:

• ⏩ Streaming Torrents. Quality streaming in P2P, competitive to Tiktok/Youtube/Netflix {warning: still on separate branch. Streaming GUI requires QT, decoder, Javascript bloat 🥹}.
• 🪞 Self-replication. Servers that can buy other servers using Bitcoin. Fully automated cloning of servers.
• ⚡ Trust. First trust framework and true Peer-to-Peer agent communication fabric. No DNS, no central control.
• 🧑‍🚒 AI models in a real-time competition for survival of the fittest using multi-arm-bandit and model score gossip.
• 👓 Find information using decentralized relevance ranking
• 🥇 First decentralized voting system. Your place, your control, your vote.
• 🥼 User-driven self-evolution. The emergent voting behavior is that users drive the roadmap using democracy. No lawyer or company can stop the will of the people.

Download .EXE : https://github.com/Tribler/superorganism-experiment/releases

AI before it was cool

Together with Szeged University we deployed fully decentral machine learning in 2012. Completely useless, due to lack of compute, lack of data, and our lack of experience. Last year we got decentral learning for search operational plus the business model.

Towards recursive self-improving AI

Strong AI is a natural monopoly. Citizens should control AI. We want control of AI to be as simple as possible. Superorganism has a system where you can vote for new features. For example:

Our superorganism is an exploratory experiment for self-hosted and self-improving systems under direct human control. It is a decentralized system ability to create seedboxes, vote on systems changes, and donate GPU resources for decentralised AI. It runs on our machines, but it's complex stuff. Frontpage screenshot below with 1) torrent management, 2) self-replicating server fleet overview, 3) voting part, and 4) autoresearch (shown page). Our Eternal Autoresearch is inspired by "Darwin Gödel Machine: Open-Ended Evolution of Self-Improving Agents". But we rely our Bitcoin payments for real human developers, so more classical collective intelligence (until AI agents become better at coding).

Technical documentation

Superorganism is the joint work of professors, post-docs, phd students and master students (i.e. the ones doing the real work!) at Delft University of Technology and academic partners. Algorithms and key components are documented here.

Algorithm	Superorganism algorithm extensive documentation
SurvivalRank	Evolutionary survival of fittest algorithm using decentralized multi-arm bandits
FuSST	Social Capital Accounting for trust with Sybil resilience
TwoStepDemocracy	TwoStepDemocracy: prototyping of self-evolving, democratic, and decentralized systems {#8812}
SelfReplicate	Self-replicating seedbox servers using programmable money
DelftClaw	Infrastructure for decentralized OpenClaw semantic search agents: identity, trust, communication, and self-replication (5 bsc students, needs integration)

Features - Everything we built so far

Now that the basic components are all operational, we're working on integration.

1. A million URLs with creative commons content on Youtube + download tooling + Bittorrent autoseeding
2. Decentralized survival-of-fittest learn-to-rank algorithm.
3. Semantic overlay (ToDo integrate our decentralized semantic overlay)
4. Search engine (ToDo integrate our decentralized learn-to-rank AI algorithms)
5. Voting and use your Bitcoin wallet to stake your identity (ToDo move from testnet to mainnet)

Taking vibe ideas into production

⚠️Exploratory prototype, this is to guide our production-grade work. Our Tribler production network will never contain any AI vibe code. We never had a full-scale network outage in 21 years.⚠️ Superorganism-experiment vibe code is unstable and unmaintainable. The algorithms, interface, and architecture are purely proof-of-principles guiding our roadmap. With the new NWO funding until 2032 we will create the first fully decentralized Youtube/Tiktok/Netwflix alternative with our Sybil-tolerant MeritRank trust model. Currently we're taking the first steps to re-implement superorganism ideas into production-level code. First priority is human coding of decentralized vector database, unigrams-based vocabulary and SentencePiece integration. Performance evaluation is important. For instance, see this interesting looking architecture of decentralized semantic search, but it has surprisingly poor performance:

General

Qt UI resources are listed in the ui/resources/ directory, including icons, fonts, images, and the resource manifest. They must be converted using the PySide6 resource compiler. Run the following command after adding new UI resources:

pyside6-rcc ui/resources/resources.qrc -o ui/resources/resources_rc.py

Democracy

• Various overlay messages for vote exchange and object synchronization
• Software developers get paid for pull request using Bitcoin with special commitments flags ALL|ANYONECANPAY.
• local SQLite is used for simple local storage

Survivalrank

• Different learn-to-rank models are deployed in the network and performance score is determined (NDCG)
• decentralized multie-arm bandit ensures survival-of-the-fittest and cleaning of low-performance variants.

Decentralized trust

Our new graph-based trust algorithm is not integrated into the prototype. This is a Sybil detection reference implementation. The FuSST algorithm fuses a trust graph G_T (the platform follow or interaction graph) with a behavioural graph G_B (a mutual k-NN graph over five platform-agnostic behavioural features) and runs belief propagation (SybilSCAR and SybilHP variants) over the combined structure, optionally seeded with a logistic-regression prior. This repository contains the complete code to reproduce every table and figure in the thesis across four labelled datasets (Cresci-2015, Cresci-2017, TwiBot-20, TwiBot-22) plus the Twitter-270k engine-validation benchmark.

Mycelium

This is the self-replication part. Any server with a bitcoin wallet and our scripts can buy another server, clone itself, move money to clone, and then this clone can buy more servers. This has proven to be very difficult to move from lab prototype into production-level code. We have been working since 2016 on this. We got generational Q-learning running. See our 2017 running code in PyPi called CloudOmate where you can buy a VPS fully scripted. No account creation, no bank account validation, no passport upload, no wait period, simply a $100 deposit against abuse. Basis for our self-replication:

Autonomous VPS provisioning system using Bitcoin payments and SporeStack API. Deploys Mycelium a BitTorrent seedbox orchestrator for Creative Commons content. What it does:

• Seeds content via BitTorrent (libtorrent)
• Auto-updates from central server GitHub (yes, we want self-hosting! Apply for phd position on this) and restarts on code changes
• Broadcasts seeded content to IPV8 peers for health monitoring

Local Bitcoin regtest environment

This project uses a local Bitcoin Core regtest node for development and integration testing. Regtest is a private blockchain intended for testing. It does not connect to mainnet, and blocks can be mined on demand.

The script scripts/regtest.sh automates the local setup by:

• creating a dedicated regtest data directory inside the project
• generating a bitcoin.conf
• starting bitcoind
• creating or loading a wallet
• mining initial blocks so the wallet has spendable funds
• exposing simple commands for status, reset, mining, and demo transactions

Dependencies

The script requires:

• bash
• bitcoind (part of Bitcoin Core)
• bitcoin-cli (part of Bitcoin Core)
• jq (for JSON parsing)

Installing dependencies

macOS

brew install bitcoin

brew install jq

Project-local data directory

The script stores all regtest data inside the repository:

.bitcoin/regtest-demo/

Configuration

The script supports a few environment variables, but all have sensible defaults.

Variable	Default	Description
BITCOIND_BIN	bitcoind	Path to the bitcoind binary
BITCOINCLI_BIN	bitcoin-cli	Path to the bitcoin-cli binary
DATA_DIR	./.bitcoin/regtest-demo	Regtest data directory
WALLET_NAME	demo	Wallet name used for testing
RPC_USER	demo	RPC username
RPC_PASSWORD	superorganism	RPC password
RPC_PORT	18443	Regtest RPC port
P2P_PORT	18444	Regtest P2P port
HOST	127.0.0.1	Host interface for the node

Usage

The script supports the following commands:

scripts/regtest.sh start
scripts/regtest.sh stop
scripts/regtest.sh reset
scripts/regtest.sh status
scripts/regtest.sh mine [n]
scripts/regtest.sh send <address> [amount] [op_return_hex]
scripts/regtest.sh treasury-address
scripts/regtest.sh sign-psbt <psbt_base64>

Command	Description
start	Starts the regtest node, creates the config file if needed, loads or creates the wallet, and ensures the wallet is funded. On first startup, the script mines 101 blocks. This is necessary because coinbase rewards must mature before they become spendable.
stop	Stops the running regtest node.
reset	Deletes the local regtest blockchain and wallet state, then starts from a clean environment. This is useful for repeatable integration tests.
status	Prints basic blockchain and wallet state.
mine	Mines one or more new regtest blocks. This is especially useful for confirming transactions during testing.
send	Sends a demo transaction from the local wallet.
treasury-address	Prints the address of the treasury. This is useful for funding the wallet from external tools or for testing incoming transactions.
sign-psbt	Signs a base64 PSBT with the local demo wallet using ALL|ANYONECANPAY and prints the signed PSBT. This is intended for the local funding-pledge flow.

Background and disclaimer

DISCLAIMER: this is throw-away code⚠️ This is exploratory running code for inspiration and active roadmapping. We will throw away this vibe code❗Our lab is focussed on critical infrastructure and we have a strict zero-AI policy for our production code. With our obtained 4.3 Million Euro in NWO funding we will turn these ideas into real production-level maintainable code. Our scientific aim: foundations of trust in the age of AI.