AI's Impact on Scientific Research Grows Stronger

Modern artificial intelligence is the product of decades of scientific research. Now, it’s paying back that effort by accelerating progress across academia. Since the advent of deep learning in the 2010s, researchers have been using AI to analyze data, conduct literature reviews, and model complex processes in every scientific discipline. The scope of problems they can tackle is expanding by the day.

Google DeepMind’s Alphafold has become the poster boy for AI’s use in science, with its inventors winning the 2024 Nobel Prize in Chemistry. Alphafold uses advances in transformers to solve the protein folding problem that had puzzled scientists for decades. Previously, the only way to discover a protein’s shape was with complex imaging techniques like X-ray crystallography and cryo-electron microscopy. Alphafold can predict the shape of a protein from nothing more than the series of amino acids making it up. This made it possible to predict the shape of every known protein in just two years, with transformative impacts on biomedical research.

Alphafold 3, released in 2024, goes even further by predicting both the structure and interactions of proteins, DNA, RNA, and other biomolecules. Google has also used AI to map the most detailed human brain connections to date, providing vital tools for exploring neuronal architecture and connectivity. This could boost our understanding of neurological disorders and core cognitive processes like learning and memory.

AI is also revolutionizing materials science. Google DeepMind’s GnoME predicted 2.2 million novel inorganic crystal structures, including 380,000 stable ones that could form the basis of new technologies. Meta has released and open-sourced its own transformer-based materials discovery models, along with a dataset of over 110 million materials simulations. Microsoft’s MatterGen uses a diffusion model to produce novel inorganic crystals with specific properties.

One of AI’s biggest strengths is its ability to model complex systems, making it a natural fit for weather forecasting and climate modeling. Google DeepMind’s GraphCast model uses graph neural networks to generate 10-day forecasts in one minute, with higher accuracy than existing methods. The European Center for Medium-Range Weather Forecasts has deployed an AI forecasting system that is faster, 1,000 times more energy efficient, and 20 percent more accurate. Microsoft’s Aurora, a foundation model for the Earth system, outperforms existing approaches in predicting air quality, ocean waves, and tropical cyclones.

AI is also contributing to fundamental discoveries in physics. The Large Hadron Collider uses AI to sift through data from millions of particle collisions per second. Researchers in Germany are using AI to detect neutron star mergers, helping scientists observe these events in real-time. The promise of AI taking on the role of scientist itself is exciting. Combining lab automation, robotics, and machine learning, it’s becoming possible to create “self-driving labs” that autonomously run experiments to achieve specific goals.

Carnegie Mellon University researchers showed that their AI “Coscientist,” powered by OpenAI’s GPT-4, could plan and carry out the chemical synthesis of known compounds. Google’s multi-agent system, powered by its Gemini 2.0 reasoning model, helps scientists generate hypotheses and propose new research projects. Another “AI scientist” developed by Sakana AI wrote a machine learning paper that passed peer review for a prestigious AI conference.

Despite the excitement, AI’s role in science has potential downsides. Neural networks are black boxes, making results challenging to interpret. Many researchers are not familiar enough with the technology to catch common pitfalls. Nonetheless, the incredible power of these models to crunch through data and model things at scales beyond human comprehension remains a vital tool. With judicious application, AI could massively accelerate progress in a wide range of fields.