AI Models and New Drugs

Syllabus: Awareness in the fields of IT, Space, Computers, Robotics, Nano-technology, Bio-technology and issues relating to intellectual property rights.

Background

• In 2020, Google DeepMind declared that AlphaFold had solved the long-standing protein-folding problem in biology.
• This breakthrough was expected to revolutionise drug discovery, yet five years later, no major wave of new drugs has emerged.
• This contradiction reflects Eroom’s Law, which shows declining efficiency in drug discovery despite technological progress.

Understanding Eroom’s Law

• Eroom’s Law (Moore spelled backwards) observes that drug discovery becomes slower and more expensive with time.
• Unlike computing progress predicted by Moore’s Law, pharmaceutical innovation does not accelerate with increased investment.

Why AI Falls Short in Drug Discovery

• AI Expands Quantity, Not Quality
  • Drug development begins with a hypothesis about how a molecule affects disease.
  • Even before AI, scientists generated millions of hypotheses, most failing in trials.
  • AI now produces billions of hypotheses, but cannot judge quality or biological relevance.
  • Intuition, creativity, and scientific insight—crucial for good hypotheses—remain human-driven.
• AI Works on Bounded Problems, Not Open-Ended Exploration
  • Protein folding suited AI because it had:
    • A defined question,
    • A vast dataset (~1.5 lakh known structures),
    • A clear sense of a correct answer.
  • AlphaFold excelled in this structured environment, similar to a student cracking a standardised exam.
• Drug Discovery Is Not a Bounded Puzzle
  • Drug discovery resembles open-ended exploration, not a predictable exam.
  • It lacks fixed data, clear rules, or predictable outcomes.
  • Historically, major medicines—penicillin, insulin, paracetamol, metformin—emerged from serendipity and human curiosity, not algorithmic prediction.
  • AI can replicate known patterns but cannot imagine novel mechanisms or therapeutic directions.

Conclusion

• AI accelerates data processing but cannot replace human creativity at the core of drug discovery.
• Discovering transformative medicines remains an uncertain, human-driven, intuition-based process—far beyond the current reach of algorithmic logic.