Teaching AI How Science Works (ifp.org)

A new initiative aims to revolutionize AI's role in scientific research by establishing "Unstructured Data Generation Labs" that capture the often-overlooked tacit knowledge essential for hands-on experimentation. Current AI applications in science heavily rely on structured datasets, but the complexity of real-world scientific processes — like troubleshooting material synthesis — requires understanding nuances beyond what is documented in academic literature. This ambitious project seeks to generate multimodal datasets by integrating data collection methods, such as body cameras and system logs, during actual research activities across various fields, including biotechnology and advanced materials. With a proposed budget of $2 billion over eight years, the initiative will unfold in three phases, starting with pilot programs that will evolve into larger, focused research efforts. This approach not only promises to enhance research productivity by 10-100 times but also paves the way for creating AI that can effectively accelerate scientific discoveries and solve critical challenges. By bridging the gap between AI capabilities and practical scientific processes, these labs could transform the landscape of research, enabling faster advancements in areas like drug development and materials science while ensuring safety through stringent oversight.