Demis Hassabis, CEO of Google DeepMind, has proclaimed that we are on the cusp of the singularity, a future moment when AI surpasses human intelligence and transforms the world. This statement was made in the context of scientific AI, highlighting Google’s weather prediction software, WeatherNext, which successfully predicted Hurricane Melissa’s catastrophic landfall in Jamaica last year.

The contrast between Hassabis’ ambitious vision and the practical achievements of WeatherNext highlights the tension between two approaches to AI in science. The first approach involves creating AI tools designed to solve specific scientific problems, while the second approach focuses on developing agentic, LLM-based systems that can execute cutting-edge research projects independently.

The latter approach is driving AI enthusiasm, with recent excitement around recursive self-improvement, where AI systems could drive their own advancement. Agentic systems are making significant research contributions, sometimes with minimal human guidance, and autonomous AI scientists are emerging on the horizon.

This shift heralds a new era for science, where humans and AI systems collaborate as equals or AI makes scientific progress on its own. While Google continues to work on specialized AI tools, the rise of autonomous AI researchers may reduce the need for super-specialized tools like AlphaFold or WeatherNext.

Nevertheless, such tools remain widely popular among scientists, with over three million researchers worldwide using protein structure predictions from AlphaFold. The future of AI-driven science is uncertain, but one thing is clear: the path forward is evolving, and the role of humans in scientific research is changing.

Photo by GEORGE DESIPRIS on Pexels
Photos provided by Pexels