How Artificial Intelligence Shapes Science: Evidence from AlphaFold
This National Bureau of Economic Research working paper analyzes the impact of the AlphaFold2 AI model on scientific discovery, specifically within structural biology. Released in July 2021, AlphaFold provided instant access to hundreds of thousands of predicted protein structures, potentially reducing the need for resource-intensive experimental determination. However, the study finds that the rate of experimental structure determination has remained largely unchanged, suggesting researchers use AI predictions to complement rather than replace traditional methods. The analysis reveals a significant shift in basic research focus, with a 15 to 40% increase in studies targeting proteins that previously lacked structural data. This indicates that AI is directing scientific inquiry toward less-studied areas. Despite this expansion in basic knowledge, the authors find no current evidence that early-stage applied drug development is increasingly targeting these newly characterized proteins. The findings highlight how frontier AI models influence the direction of academic research while noting that translational applications in pharmaceuticals may lag behind basic scientific advancements.
Wire timeline
How Artificial Intelligence Shapes Science: Evidence from AlphaFold
This National Bureau of Economic Research working paper analyzes the impact of the AlphaFold2 AI model on scientific discovery, specifically within structural biology. Released in July 2021, AlphaFold provided instant access to hundreds of thousands of predicted protein structures, potentially reducing the need for resource-intensive experimental determination. However, the study finds that the rate of experimental structure determination has remained largely unchanged, suggesting researchers use AI predictions to complement rather than replace traditional methods. The analysis reveals a significant shift in basic research focus, with a 15 to 40% increase in studies targeting proteins that previously lacked structural data. This indicates that AI is directing scientific inquiry toward less-studied areas. Despite this expansion in basic knowledge, the authors find no current evidence that early-stage applied drug development is increasingly targeting these newly characterized proteins. The findings highlight how frontier AI models influence the direction of academic research while noting that translational applications in pharmaceuticals may lag behind basic scientific advancements.
National Bureau of Economic Research Working Papers