This article was first published on The Economist Now & Next hub on 21 November 2024.
With AI turbocharging research—even powering robot scientists—a new era of faster, cheaper scientific discovery is under way
While most of the world has just started fixating on Artificial Intelligence (AI), it has actually been part of the scientific toolkit since the 1960s. For most of its life, the use of AI has been confined to disciplines like particle physics and mathematics, where practitioners are well versed in computer code.
But now, just as AI is transforming almost every industry, it is also reshaping other scientific disciplines. The technology is a lot easier to use and more people—graduates in engineering and the sciences—know how to use it. According to CSIRO, Australia’s science agency, more than 99% of research fields were producing AI-related results by 2023.
There are high expectations for AI to supercharge research and actually reimagine the scientific process. Scientists say that the technology is enabling them to look at problems in a different way—unleashing creativity, as well as enhancing efficiency and accuracy. For example, literature-based discovery uses AI to search through millions of research papers to find patterns and connections, and then suggest new hypotheses for scientists to investigate. It can even matchmake collaborators from different fields.
This is leading to ambitious ideas for how AI can address critical societal issues such as improving healthcare and combating climate change.
Finding the needle in the haystack
In the pharmaceutical industry, the protracted timelines and high costs of developing new drugs mean that new therapeutics for rare diseases are often considered uneconomic. However, AI tools are revolutionising the drug-discovery process, making it faster and cheaper to develop new treatments. Scientists predict that using AI in the preclinical stage of drug development could bring a time and cost saving of up to 50%. This has the potential to reshape the economics of the industry, and help people with rare and incurable conditions.
For example, AI-driven biotech company Insilico Medicine used AI to find a new treatment for idiopathic pulmonary fibrosis, a rare progressive illness of the respiratory system. Developed in 2020, the drug is now in phase 2 clinical trials, meaning it is being tested on patients with the condition. It took just 18 months and $3m to reach that stage—a fraction of the time and money normally spent.
Insilico was able to develop the treatment so quickly by using its Pharma.AI platform, which was built on years of modelling large biological, chemical and textual data sets. Similarly, Google DeepMind has created a resource for medical researchers by building a database of more than 200m protein structures. The database powers AlphaFold, an AI system that predicts a protein’s 3D structure from its amino-acid sequence. The system was used to find—in just 30 days—the structure of a protein that influences how a type of liver cancer spreads, paving the way for a new targeted treatment.
The scientist who never sleeps
Just as AI is used to automate the search for new drugs and speed the process, it is also being deployed to automate long sequences of high-frequency tasks in laboratories—something that humans cannot execute with such high levels of reproducibility.
The first fully autonomous mobile robotic chemist was created in 2020 at the University of Liverpool. Using AI to analyse data and make decisions on what experiments to do next, it navigates the lab by touch sensors and light detection and ranging. Operating on its own for eight days, the robot completed nearly 700 experiments—more than a PhD candidate would do in the four years it takes to earn their degree. It produced a photocatalyst (materials that change the rate of a chemical reaction on exposure to light—in this case, for hydrogen production from water) six times better than what human researchers had devised.
The team that developed the robotic chemist now has two mobile robots, one working on catalysis research in the area of clean energy and the other on organic chemistry relevant to pharmaceuticals.
Keeping a (human) hand on the controls
AI promises to transform scientific research but, as in other spaces, it is not a silver bullet and must be used responsibly. Just as schoolchildren are using ChatGPT to write their essays, AI is being used by some scientists to write bogus research papers that are being published in scientific journals. This compromises the solid foundation of trustworthy findings that scientific research is based on. “Across all professions, technological progress can make us more productive, but it doesn't absolve us of the need to ensure that our output is accurate and will be used responsibly,” says Gareth Dickson, a partner at Mishcon de Reya. “That is especially true in research, where rigorous, informed peer review takes time but must be preserved given its critical role in generating trust in progress.”
“One of the key takeaways from the pandemic is that even some of those who stand to benefit from medical interventions can be quite easily led to believe that science cannot be trusted, that unusually rapid progress should be feared and that big pharma is spying on them,” notes Mr Dickson. “A careful balance must be maintained to ensure the unprecedented expansion of scientific discovery continues and is embraced by its intended beneficiaries, without stirring up greater suspicions."
AI may be the tool but, in this case, it is the human editors and peer reviewers at journals who are not doing their jobs properly. More broadly, AI relies on humans choosing the data it learns from and creating the rules by which it operates. While it may take on more and more scientific roles, the humans remain in charge.