Asimov Press Designing AI for Disruptive Science
Mar 23, 2026
Alvin Djajadikerta, writer on tech and scientific progress, explores how AI could enable true paradigm shifts. He contrasts exhaustive data mapping with simpler, more useful schematics. He examines why current AIs replicate existing frameworks, how historical breakthroughs arose, and proposes deliberate AI designs and metascience experiments to foster disruptive science.
AI Snips
Chapters
Transcript
Episode notes
AI Risks Producing Hypernormal Science
- Current AI excels at prediction within existing frameworks but struggles to invent new categories.
- This creates risk of "hypernormal science": better predictions paired with reduced capacity to ask novel questions.
Design AI To Propose New Conceptual Vocabularies
- Build AI that invents new conceptual vocabularies rather than only predicting labels.
- Prioritize visionary machines that propose alternative schematics, not just more navigable maps.
How Einstein And Darwin Won With Simple Cores
- Einstein and Darwin succeeded by proposing simple core principles that explained diverse data despite missing details.
- Einstein's two postulates led to E=mc^2; Darwin lacked genetics yet natural selection still reshaped biology.