Sakana AI Scientist performs research by itself
ChatGPT can write your research paper, but what if an AI can take care of the research, too? Japan-based Sakana AI partnered with two British universities to create “The AI Scientist,” the world’s first AI system for automating scientific research.
With help from the University of Oxford and University of British Columbia, it can perform every scientific method step. This ranges from forming research to summarizing results.
The official Sakana AI website also shared sample papers that The AI Scientist generated, such as “DualScale Diffusion: Adaptive Feature Balancing for Low-Dimensional Generative Models.”
Article continues after this advertisementHow does the Sakana AI Scientist work?
Introducing The AI Scientist: The world’s first AI system for automating scientific research and open-ended discovery!https://t.co/8wVqIXVpZJ
— Sakana AI (@SakanaAILabs) August 13, 2024
From ideation, writing code, running experiments and summarizing results, to writing entire papers and conducting peer-review, The AI… pic.twitter.com/SJuat9a2Uw
Sakana says The AI Scientist enables other AI models like Large Language Models (LLMs) to perform research independently. LLMs are the AI programs that follow popular chatbots like ChatGPT.
READ: Sakana AI lets you turn pics into traditional Japanese art
The Sakana AI Scientist follows four main processes:
Article continues after this advertisement- Idea Generation: The research AI “brainstorms” a diverse set of novel approaches based on the user’s desired topic. Then, the Scientist program will explore these research directions until it finds the most suitable one.
- Experimental Iteration: The Sakana AI Scientist executes experiments based on its chosen research approach and the user’s topic. Next, it plots results and notes important details for writing a research paper
- Paper Write-Up: Sakana’s latest AI program writes a research paper and uses the AI-powered research tool Semantic Scholar to cite relevant sources.
- Automated Paper Reviewing: The AI Scientist evaluates its previous papers with near-human feedback to guide future research projects.
READ: How to make your own AI assistant
Of course, Sakana’s AI Scientist has limitations, so it listed them:
- The AI Scientist cannot see, so it can’t fix visual issues with a paper or read its charts and graphs.
- This AI program can mistakenly implement its ideas or make unfair comparisons to baselines.
- Sometimes, the AI Scientist can make critical errors when writing and evaluating results. For example, it struggles to compare the magnitude of two numbers.
Sakana AI does not aim to replace human scientists. Instead, it wants to realize the full potential of artificial intelligence in scientific research for further innovation and problem-solving.