AI in Science and Discovery

How AI is accelerating scientific research, from predicting protein structures to discovering materials, and why this could matter for long-run growth.

One of the most promising uses of AI may be in science itself. If AI can speed up discovery, it could help reverse the trend that ideas are getting harder to find and boost long-run economic growth.

AlphaFold

Proteins are the building blocks of life, and their three-dimensional shapes determine what they do. Predicting a protein’s shape from its sequence was a problem scientists had struggled with for decades. In 2020, AlphaFold, developed by Google DeepMind, predicted protein structures with accuracy comparable to experimental methods. DeepMind later released predicted structures for over 200 million proteins, freely available to researchers.

In 2024, the Nobel Prize in Chemistry was shared by Demis Hassabis and John Jumper of Google DeepMind for protein structure prediction, and David Baker for computational protein design.

Other uses

Drug discovery: AI helps identify promising drug candidates and predict how molecules behave, potentially cutting the time and cost of developing medicines.
Materials science: AI models have predicted many new stable materials, useful for batteries and electronics.
Weather forecasting: AI weather models can produce forecasts quickly and, in some cases, more accurately than traditional methods.
Mathematics and coding: AI assists researchers in proofs and software.

Economic significance

Economists have long seen new ideas as the engine of long-run growth. If AI raises research productivity, allowing more discoveries per researcher, it could have effects far beyond individual industries. Some economists speculate AI could significantly speed up economic growth; others are more cautious, noting that experiments, regulation and real-world testing still take time.

Limits

AI predictions must still be tested in laboratories and clinical trials.
Scientific data can be limited or biased.
The benefits may concentrate among well-funded institutions.

Years of work in minutes

Determining one protein's structure through experiments could take a researcher months or years. AlphaFold can predict many structures in minutes. Researchers studying diseases, from malaria to neglected tropical diseases, can now start with predicted structures, saving large amounts of time and money.

Thinking AI replaces scientists

AI speeds up parts of research, such as prediction and screening, but scientists still design experiments, interpret results and test findings in the real world.

Key takeaways

AI is accelerating scientific research in biology, chemistry, materials and weather.
AlphaFold predicted structures for over 200 million proteins.
The 2024 Nobel Prize in Chemistry recognised AI-driven protein research.
Raising research productivity could boost long-run growth, though testing still takes time.