Other
How AI is revolutionising science
The EconomistBear, Andy Davis, Alex Zarankov, Emily Shuckborough, Adi Rakhoum, Andy Cooper, Alok Jha
- Transformative Scope: Generative AI is accelerating drug discovery for incurable conditions like idiopathic pulmonary fibrosis (IPF) and driving leaps in climate science, material science, and animal communication research.
- IPF Case Study & Statistics:
- Insilico Medicine utilized generative AI to develop an IPF treatment in just 18 months for $3 million, a fraction of traditional timelines and costs.
- The drug is currently in phase two clinical trials.
- Standard preclinical development savings via AI are projected to range from 25% to 50% in time and cost.
- Current IPF mortality rate: 50% of patients die within five years of diagnosis.
- Protein Structure Breakthroughs:
- Google DeepMind's AlphaFold algorithm predicts protein shapes from amino acid sequences, having built a database of over 200 million proteins.
- The tool is used by more than 2 million researchers, though predictions remain imperfect and occasionally erroneous.
- Investment Trends:
- AI drug discovery investment in China reached $1.26 billion in 2021.
- The "democratization" of drug discovery is attracting significant capital from diverse players beyond traditional pharmaceutical giants.
- Multidisciplinary Applications:
- Cambridge Climate Science: Super-resolution AI models enhance low-resolution electron microscope images, while literature-based discovery algorithms identify patterns across millions of papers to suggest new hypotheses.
- Tel Aviv Bat Research: AI algorithms correlate specific vocalizations with behaviors (e.g., fighting over food, mating), revealing dialects and "baby talk" in Egyptian fruit bats.
- Animal Communication: Researchers are also detecting regional accents in wolves and beginning to decode sperm whale sounds, though concerns regarding human bias in dataset collection persist.
- Emerging Risks & Fraud:
- Analysis indicates a rise in AI-generated content in scientific journals, including identical fake images submitted by different labs.
- Specific markers of AI plagiarism include phrases like "regenerate response" and nonsensical terminology (e.g., "random value swapped for irregular esteem").
- Over 1,000 papers have been flagged for containing identical AI-produced images, raising concerns about the integrity of the publishing process.
- Robot Scientists & Automation:
- University of Liverpool's Andy Cooper deployed a roving robot that conducted 700 experiments in eight days, equivalent to the work a human PhD student would typically take four years to complete.
- New initiatives involve a "double act" of two mobile robots: one for organic chemistry/pharmaceuticals and another for catalysis in clean energy.
- Reproducibility Crisis: Autonomous labs can address the "reproducibility crisis" by systematically repeating experiments and publishing negative results, which human scientists often avoid.
- Future Outlook & Requirements:
- Experts predict a "golden age of discovery" similar to the introduction of the laboratory, contingent on solving specific bottlenecks.
- Critical requirements for success include access to right data, appropriate regulations, and monetization strategies.
- Current limitations include the need for human intervention in decision-making and the challenge of encoding reasoning into large language models.
- Potential breakthrough is expected when AI systems discover findings that are "simply not realistically conceivable by humans alone."