Overview
The Generative Research Team (GRT) is a synthesis of human, computational, and informational entities that employs Active Inference, systems engineering, and cognitive security to explore research topics. Roles within the GRT are modular and composable, allowing for flexible resource and attention allocation. The GRT can be designed to address various areas of concern such as Research, Peer Review, Funding, Communication, Cryptography, and more, using implementations that blend human and computational capacities. Tools like Active Blockference and cadCAD, along with Cognitive Security concepts like Narrative Information Management (NIM) and Verifiable Information Ecosystem (VIE), ensure situational awareness and reliability in research trajectories. Here we provide both a worked and a sketched example of a GRT with specific roles and functions. Through collaborative environments, innovation, resilient architectures, and meta-prompting, the GRT adapts to unexpected changes. Effective communication strategies facilitate wide dissemination of research findings. The primary novel contributions here include the exploration of augmented architectures, the integration of Active Inference as a cognitive kernel into GRTs with shared intelligence, and the application of cognitive models for enhanced research processes. Additionally lists of related work and tools are provided, reflecting some of the contemporary work in this space. In the future, advanced GRT will be able to navigate uncertain landscapes and produce impactful outcomes. Today we are using GRT to study Active Inference; tomorrow we will use Active Inference to study GRT.
Overview source: Curated paper metadata.
Methods and contributions
Read the source for the full argument, qualifications, and evidence.
Findings and contributions
- The authors found it preferable to split browsing/gathering and curation across two agents rather than one agent doing all tasks.
- In the worked example the agents executed tasks such as literature review on active inference, summarizing key points, and preparing a report; the authors note the approach is currently limited.
- Proposes that GRT roles be flexible: any role may be non-existent, singular, plural, or autonomous (e.g. DeSci DAOs).
- Cognitive security concepts (NIM, VIE) are currently integrated only via meta-prompts, with fuller implementation left to future work.
- States the primary contributions as augmented architectures, Active Inference as a cognitive kernel in GRTs, and cognitive models for research processes.
Methods
- Two-agent AutoGPT literature-review GRT (Professor Karl, RA Joe) — A worked example built on open-source AutoGPT runs two LLM agents that forage for Active Inference literature and write a report.
- Shared-folder file protocol for asynchronous agent coordination — Agents communicated through text files (project description, requests.txt, research_data.txt, report.txt) in a shared folder.
- LangChain tools with FAISS vector memory per agent — Setup code gives each agent file read/write, CSV and webpage-QA tools (plus DuckDuckGo search for the assistant) and FAISS-backed OpenAI-embedding memory.
- Sketched eight-role GRT for a project proposal using AEOS — A larger GRT is sketched with roles such as Research Director, Statistical Modeler and Grant Forager, using AEOS Active/Informational Entities.
- Active Blockference and cadCAD as implementation layer — GRT work is hosted in Active Blockference, which combines Active Inference modeling with the cadCAD complex systems framework.
Summary sources: Paper metadata and evidence · Extracted source text.
Document downloads
PDFs are hosted here; supporting sources may use pinned archives.
- 2023_GenerativeResearchTeams.pdf PDF · 0.36 MiB · extracted-text source
Citation
Citation metadata follows the unified bibliography.
Catalog details and resources
Related works
Ranked by recorded methods, topics, contributors and associated software, with contributor roles and domain-only matches labelled. Metadata overlap is not a citation or a scientific equivalence.