Origin of life
How non-living chemistry became self-copying, evolving cells is still unknown.
Open in the interactive tree →Life on Earth arose within a few hundred million years of the planet becoming habitable, but nobody knows the route from simple molecules to the first cell. Leading ideas are an 'RNA world', metabolism-first scenarios and lipid protocells, and proposed settings range from hot vents to ponds.
As of October 2026
Geological evidence suggests life emerged between 4.32 and 3.48 billion years ago, and a 2024 study placed the last universal common ancestor around 4.2 billion years ago. Nucleobases and amino acids have been found in meteorites and comets, so the building blocks are plentiful. Researchers have built simple self-assembling vesicles, but a functional protocell that undergoes Darwinian evolution has not been made in the laboratory.
What is missing
- Knowledge of early Earth conditions: no rocks survive from the right time and place
- A chemical route in which RNA, lipids and metabolism arise together and cooperate
- Ribozymes that can copy themselves accurately
- A second, independent example of life for comparison (for example on Mars or Europa)
Becomes possible once solved
- Creating living systems from non-living chemistry in the lab
- Estimating how common life is in the universe
- New design principles for synthetic cells
Open steps
- A self-copying RNA enzyme Medium AI leverageMake an RNA enzyme that copies RNA, itself included, accurately and in useful yield; the best small one has 94% fidelity but about 0.2% yield.
- RNA, lipids and metabolism together Medium AI leverageFind conditions in which RNA, membranes and metabolism-like reaction cycles arise and cooperate from simple starting chemicals.
- Protocells that evolve Low AI leverageA compartment that copies its own information and undergoes Darwinian evolution has never been built in the lab.
- Reading early Earth's record Medium AI leverageRead the few surviving rocks and meteorites older than 3 billion years for chemical traces of life and for the conditions of the right time and place.
- A second example of life Low AI leverageSearch for independent life on Mars or Europa as a comparison for how life starts; this needs samples and agreed biosignature tests.
Where AI could help
Low AI leverage. The gaps are lost geological evidence and an unproven chemical route; AI can search reaction space but cannot recover rocks that no longer exist.
- Autonomous labs that explore prebiotic reaction networks far faster than manual chemistry
- Machine-learning models of ribozyme sequence space to look for self-copying RNAs
- Pattern recognition in ancient rocks, meteorites and Mars samples to test for chemical biosignatures
- Agents that check competing scenarios against geochemical data
Shown so far
- In November 2025 PNAS reported that supervised machine learning on mass-spectrometry data from 406 samples found biogenic molecular signals in 3.33-billion-year-old rocks and signs of photosynthesis by 2.52 billion years ago. source
- In June 2021 Nature Communications reported a Glasgow robot chemist that used machine learning to pick reagents and run multi-week prebiotic reaction experiments that produced complex molecule mixtures. source
Prerequisites
- Evolution by natural selection1859
- Periodic Table of Elements1869
- Krebs cycle1937Metabolism-first theories look to a reverse Krebs cycle as an early self-sustaining chemistry
- The genetic code1966
- Catalytic RNA and the RNA world1982-1983The RNA world idea rests on RNA that can both store information and catalyse