How AI Scientist‑v2 Is Pioneering Fully Automated Scientific Discovery

AI Scientist‑v2 is an open‑source system that automates the entire scientific workflow—from hypothesis generation and experiment execution to data analysis and paper drafting—and has already produced a fully AI‑written paper accepted at the ICLR 2025 workshop.

From Template‑Based to Autonomous Exploration

Version 1 relied on human‑written templates, offering high success rates for well‑defined tasks but lacking true creativity. Version 2 replaces templates with a "progressive agent tree search" architecture, where an "experiment manager" agent explores a branching decision tree of research possibilities, accepting lower success rates in exchange for open‑ended discovery.

"v1 follows clearly defined templates, leading to high success rates, while v2 adopts a broader, more exploratory approach with lower success rates. v1 is best for clear‑goal tasks; v2 is designed for open scientific exploration," the project README explains.

Core Technology: Multi‑Agent Research Pipeline

The system’s pipeline consists of four stages:

Hypothesis generation : proposes novel, testable research ideas based on existing knowledge.

Experiment execution : automatically writes Python code, sets up the environment, and runs the experiment.

Data analysis : processes results, creates visualizations, and extracts insights.

Paper drafting : assembles the findings into a manuscript that follows academic conventions.

The experiment manager evaluates each research branch, decides whether to deepen investigation or prune it, and allocates compute resources efficiently.

Important Warning: Capabilities and Risks

The README warns that the system executes LLM‑generated code, which may import unsafe packages, perform uncontrolled network access, or spawn unexpected processes. Users are advised to run the software inside a sandboxed Docker container and assess risks themselves.

Quick Start: Setting Up on Linux

For developers who want to try the system, the project provides clear setup steps. It runs on Linux with an NVIDIA GPU.

conda create -n ai_scientist python=3.11

conda install pytorch torchvision torchaudio pytorch-cuda=12.4 -c pytorch -c nvidia

conda install anaconda::poppler conda-forge::chktex

pip install -r requirements.txt

The system supports OpenAI, Gemini, and Claude (via AWS Bedrock) APIs; users only need to set the appropriate API‑key environment variables.

Who Should Pay Attention

AI and scientific‑computing researchers : can use the platform to automate large‑scale hypothesis testing and study multi‑agent system design.

Open‑source developers and enthusiasts : the codebase offers a concrete example of autonomous agent architectures and workflow automation.

Academic publishers and educators : the project prompts discussion about peer review, the nature of innovation, and the evolving role of human scientists in an era of AI‑driven research.

AI Scientist‑v2 marks a transition from AI as a mere tool to AI as a research collaborator, opening a window onto the future of human‑machine co‑creative science.