Funding for Autonomous Research Discovery (ARD) System: PsyBEE

Summary

This proposal seeks approval to fund the development of an Autonomous Research Discovery (ARD) System for PsyDAO. The ARD System aims to create a multi-layer, multi-agent scientific research discovery platform focused on psychedelic science. This project will enhance PsyDAO's research capabilities, revolutionizing how we approach and fund scientific research in the psychedelic field, and generating value that can be productized into IP distributed throughout the bioDAO ecosystem: ARD for bioDAOs-at-large.

Background

The ARD System, proposed by community member @stw2_, is designed to leverage large language models (LLMs) and autonomous agents to create a comprehensive research discovery pipeline. This system will cover various aspects of the research process, from data scraping to research proposal development, enabling continuous learning and optimization of research discovery efforts.

Proposal Details

We propose to allocate 10 ETH from the PsyDAO treasury to fund the development of the ARD System. The funding will be distributed across 5 project milestones, with 2 ETH released at each milestone.

Project Milestones and Funding Schedule

1. Project Initialization and Data Scraping Module (2 ETH)
2. Knowledge Graph Development (2 ETH)
3. Lead Generation and PsyBEE Implementation (2 ETH)
4. Research Proposal Generation Module (2 ETH)
5. System Integration and Optimization (2 ETH)

Expected Outcomes

1. PsyBEE: A psyDAO bot providing psychedelic news and research alerts
2. Psychedelic Science Knowledge Graph
3. PsyDATASET: A curated dataset for psychedelic science
4. PsyGPT: A specialized GPT-like model for psychedelic research
5. Autonomous lead and research proposal generation system

Governance and Oversight

• The Psychedelic Science Fund multisig will be responsible for releasing funds upon commencement of each milestone.
• The project lead (@stw2_) is an ai researcher at a top university ai lab in Germany and will provide bi-weekly updates to the community on the project's progress.
• The community will have the opportunity to review and provide feedback at each milestone.

Budget

The total budget for this project is 10 ETH, to be paid in installments over 5 project milestones each with a corresponding payment of 2 ETH at the commencement of the milestone work, plus pass-through operational expenses for compute, servers (hosting bot, storing scraped data, embeddings, knowledge base), proxy services, LLMs, API calls, etc.

1. Project Initialization and Data Scraping Module (2 ETH)

   • Set up project infrastructure
   • Develop and implement the data scraping module for retrospective and continuous scraping
   • Deliver initial dataset of scraped psychedelic research data

2. Knowledge Graph Development (2 ETH)

   • Design and implement the Psychedelic Science Knowledge Graph structure
   • Populate the knowledge graph with scraped data
   • Develop query mechanisms for efficient data retrieval

3. Lead Generation and PsyBEE Implementation (2 ETH)

   • Develop lead generation module using the knowledge graph
   • Implement PsyBEE bot for delivering news and research alerts
   • Integrate PsyBEE with Discord/Telegram for community access

4. Research Proposal Generation Module (2 ETH)

   • Develop autonomous research proposal generation system
   • Implement integration with knowledge graph and lead generation module
   • Deliver initial set of autonomously generated research proposals

5. System Integration and Optimization (2 ETH)

   • Integrate all modules into a cohesive system
   • Implement learning and optimization mechanisms
   • Integrate Human-in-the-loop (HITL) for the proposal generation in order to generate agentic workflow data for future fine-tuning of base models

The next phase of the project will focus on training custom base models in a self-supervised manner (PsyGPT). Additionally, we will improve agent performance through supervised learning based on agentic workflows, which are heavily dependent on the quality of data gathered during the Human-in-the-loop phase. The quality and volume of this HITL data will play a critical role in the fine-tuning and optimization of the models and workflows.

Each milestone will be considered complete upon delivery of the specified outputs and approval by the Science Fund multisig.

Pass-through operational expenses are estimated at 1 ETH for compute, servers (hosting bot, storing scraped data, embeddings, knowledge base, etc.), proxy services, LLMs, API calls, and additional operational funds will be approved on a rolling basis with upon request from the project lead.

Tokenization

The PsyBEE project IP will be registered onchain using a Molecule IP tokens. PsyDAO will mint a PsyBEE IP-NFT and fungible PsyBEE IP tokens, with the fungible IP tokens distributed to PSYC and PSY holders for free via yield farming.

Voting

PSYC holders can vote on this proposal with the following options:

• Yes: Approve the funding allocation for the ARD System project
• Comment: See comments in Discord