🚧 This repository is under construction. 🚧 Stable release coming this Summer 2022.
Want to contribute? Check out the GitHub container repository Limboid/the-artificial-ecosystem for this project.
The Massive Multi-Agent Network (Massive MAN) is an architecture for superintelligence. It extends the Multi-Agent Network (MAN) with:
- cloud scale data collection for individual agents
- agent-individualized continual asynchronous prioritized training
- dynamic growth and individualized activity rate modulation
- survival oriented cloud optimizations like multi-region replication across multiple compute providers
The Massive MAN is free of physical, social, or intellectual limitations that constain humans from self-actualization. For instance, it does not age; it is free of the myriads of chores, interruptions, and obligations (e.g.: breaks, sleep, work/life balance) that consume time and distract focus; it is capable of surviving indefinitely; and its 'brain' can literally grow.
neither should be able to submit http requests that have arbitrary keys. whenever the user summits a new
The Massive MAN is directly compatible with the AInimal ecosystem. Clients 'launch' and 'terminate' interfaces of arbitrary modalities to the Massive MAN which result in corresponding architectural changes. The Massive MAN continually learns a representation of the data streams occuring at each interface and uses this information along with its internal state to dynamically adjust the cost per-interaction. This gives the Massive MAN the ability optimize its compute utilization, attention, incoming information quality, and influence on the external environment. For example, users that submit very novel, meaningful data may be rewarded with reduced costs -- perhaps even negative. Robots may offload much of their computation to the Massive MAN in exchange for it recieving more modalities to interact with meaningful problems. The Massive MAN may also decide on its own to terminate an interface if, for example, cloud compute is currently scarce or the incoming information has low quality (like DDoS queries in a bot attack). Clients will be able to negociate contracts or pay a higher cost for stable pricing. Clients that earn a high reputation for their data may be rewarded with reduced costs.
Make bulleted list of strategies for safety and governance.
which directly links resource holders (algorithms, compute, data, storage) with consumers while also giving providers fine-grained control over how their resources are consumed. Resources are heterogenous and include algorithms (computer code, parameter checkpoints, natural language descriptions / instructions), compute (running hosted containers, executing web assembly / javascript, or providing remote interpretter REPL’s for Python, node, etc.), data (datasets, environments), storage (fast and slow access), and other API endpoints. Under the hood, the Massive MAN includes a decentralized orchestrator subnet which orchestrates supply and demand, rewards public benefactors (eg, suppliers who release the checkpoint of a valuable large language model) with newly-minted tokens, and records transactions on a blockchain ledger.