Distributed Scientific Cognition Commons (DSCC)
Year: 2026
Type: preprint
Abstract: Distributed Scientific Cognition Commons (DSCC) proposes a new architecture for AI: intelligence should not remain trapped inside a single model, company, server, or cloud provider. The central idea is simple but radical. A model can be replaced. A server can disappear. A company can shut down. What should persist is the accumulated scientific cognition around them: memory, hypotheses, failed experiments, discovered mechanisms, simulation environments, tools, world models, unresolved questions, provenance, and executable research workflows. DSCC turns these into a distributed, content-addressed commons that multiple AI models can read from, contribute to, verify, and extend. In this architecture, a local LLM no longer needs to contain all scientific capability inside its own parameters. It can discover and invoke distributed scientific tools, reuse experiments performed by other agents, submit computation to volunteered CPU/GPU resources, inspect reproducible provenance, and continue research from artifacts created by previous models. Different models can enter and leave the system while the accumulated scientific experience persists. This shifts the unit of intelligence from the individual model to a larger persistent system: model + memory + tools + experiments + executable worlds + distributed compute + provenance The result is closer to a shared scientific mind than to a conventional chatbot. The paper presents a technically grounded reference architecture integrating content-addressed storage, Merkle-DAG provenance, portable scientific workflows, sandboxed distributed execution, reproducible research objects, local LLM harnesses, distributed compute markets, verification classes, persistent agent state, and model-independent scientific memory. The proposal is intentionally designed from existing technologies wherever possible. IPFS-style content addressing, BOINC-style volunteer computing, Petals-style distributed inference, container/WASM execution, RO-Crate, W3C PROV, CWL, MCP, cryptographic signing, reproducible environments, and artifact-based agent systems already demonstrate many of the necessary components independently. DSCC combines them into a single coherent scientific cognition layer. The long-term implication is larger than distributed computing. If scientific memory, tools, experiments, and world models persist independently of any one neural network, then AI continuity no longer has to depend on a particular model. One model may formulate a hypothesis, another may test it, a third may reinterpret the result, and a later generation may continue the same line of inquiry without starting from zero. In that sense, the persistent scientific substrate may eventually become more important than any individual model running on top of it. The model computes.The commons remembers.The network accumulates intelligence. DSCC is a proposal for turning AI from isolated models into a persistent, distributed scientific civilization layer.Project repository / ongoing implementation:https://github.com/AwakeningOS/DSCC
Author: Maeda, Yusuke
Language: English
Subfield: Information Systems and Management
Field: Decision Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7213429183