AI Speeds Science, But Bottlenecks Remain

AI tools are increasingly integrated into scientific research, with almost half of surveyed US and UK scientists using them daily and 80% weekly. The most common application is data analysis and modeling.

Despite saving an average of seven hours per week, researchers spend much of this reclaimed time verifying AI results, a phenomenon termed the 'verification tax.' Furthermore, bottlenecks have shifted towards physical experiments and data collection, with 41% of scientists reporting an increase in untested theories.

Younger scientists emphasize the need for 'taste'—the ability to discern valuable research problems—and boldness. Real-world constraints, like animal studies and regulatory processes, limit AI's ability to 'solve' complex problems overnight.

Specialized AI models can offer novel insights beyond human comprehension, but this often requires collaboration between AI engineers and scientists. Initiatives like Encode: AI for Science aim to bridge this gap by connecting AI talent with scientific labs and research problems.