Human scientists trounce the best AI agents on complex tasks (www.nature.com)

A recent report from the Stanford Institute for Human-Centered AI highlights a dramatic surge in the integration of artificial intelligence (AI) within the natural sciences, with AI-related publications increasing almost thirty-fold from 2010 to 2025. By 2025, over 80,000 papers mentioned AI, indicating a significant embrace of AI technologies by scientists. Despite this rise, the report reveals that current AI agents struggle with complex multistep workflows, scoring only about half as high as human experts. This raises questions about the effectiveness and reliability of AI in scientific research, as noted by researchers like Yolanda Gil and Arvind Narayanan. The emergence of foundation models tailored for specific scientific domains, such as the AION-1 model for astronomy, showcases advancements in AI capabilities. However, doubts remain regarding whether the rapid adoption of AI enhances scientific productivity or quality. Gil acknowledges the transformative role of AI in research but emphasizes the need for more evidence to evaluate its true impact. This ongoing dialogue is crucial for guiding the use of AI tools in the scientific community, as researchers navigate the balance between innovation and maintaining rigorous scientific standards.