Evidence for the earliest life on Earth has largely relied on finding signs of structures that may have been created during the Archaean Eon by micro-organisms. Actual fossils don’t turn up until the Proterozoic. The most distinctive and diverse of these are members of the Ediacaran fauna dated at around 635 Ma . The oldest widely accepted multi-celled eukaryote fossil was found in 2.1 billion-year old sediments from Gabon (see: The earliest multicelled life; July 2010). There have been a few claims for biogenic material, such as microscopic tubular structures in 3.5 billion-year (Ga) old pillow lavas and 3.2 Ga cherts from South Africa (see: Early biomarkers in South African pillow lavas; April 2004 and Believable Archaean fossils; March 2010) which some researchers dispute. Then there are Archaean stromatolites, which may be evidence for bacterial mats. The oldest of them have been claimed to occur in the famous, 3.77 Ga Isua metasediments of West Greenland. But such early fossils are chance finds, so geochemists have entered the arena with attempts to find irrefutable chemical signatures for life in ancient rocks.

One approach is isotope geochemistry. Carbon isotope data have been widely used, because life processes, such as photosynthesis, result in a deficiency of ¹³C relative to ¹²C. This was tried on graphite crystals trapped in sedimentary phosphate minerals from Isua. The results were at first acclaimed as a sign of life at around 3.8 Ga, but then refuted. In 2015 a similar approach was applied to graphite trapped in a 4.1 Ga detrital zircon, seemingly pushing back evidence for life into the Hadean. But zircon is a mineral produced by crystallisation of magma, so the fractionation of carbon isotopes in trapped graphite seem unlikely to shed light on the earliest life. The main drawback to using carbon isotopes is because metamorphism, Fischer-Tropsch mechanisms in hydrothermal environments, and volcanic processes may be responsible for enrichment of lighter carbon isotopes relative to ¹³C. The relative abundance of the different isotopes of iron in Archaean sediment may give clues to the transient availability of oxygen generated by bacterial photosynthesis that would oxidise soluble Fe²⁺ to insoluble Fe³⁺. Promising results were obtained in 2013 from 3.8 Ga banded ironstones at Isua. But doubt was again raised, so the only generally accepted evidence is that of the microfossils found in hydrothermal cherts in Palaeoarchaean pillow lavas from South Africa and Western Australia and the earliest stromatolites, all around 3.4 to 3.5 Ga old. However, recent research may have opened up a more convincing route to tracking down ancient life forms –actual organic molecules that make up or are produced by organisms.

Michael Wong and co-workers at the Carnegie Institution for Science in Washington, DC, USA together with other colleagues from the US, Austria, Canada, China, Belgium, Norway, Australia, the UK and France used artificial intelligence to wade through the results of geochemical analysis of over 400 ancient and modern carbon-bearing samples. (Wong, M.I. and 28 others 2025. Organic geochemical evidence for life in Archean rocks identified by pyrolysis–GC–MS and supervised machine learning. Proceedings of the National Academy of Sciences, v. 122, article e2514534122: DOI: 10.1073/pnas.2514534122). Their objective was to track the presence of organically derived molecules as far back as possible. Their approach bears a passing resemblance to that used to build genomes of ancient fossils from broken bits of DNA that reside in them. Like DNA, bio-molecules degrade over time, but leave fragments in rocks that can be detected using pyrolysis gas chromatography and mass spectrometry. In itself PGC-MS is not especially new, but using artificial intelligence (machine learning) on a massive date set certainly is: perhaps the first major trial of AI in geology.

Their samples were not just ancient rocks going back into the Archaean as far back as 3.5 Ga, but included modern biological material, meteorites presumed to have been devoid of life since their origin in pre-solar system times and synthetic samples. Wong et al divided 272 samples with known biological affinities into 9 groups to train the AI algorithm. The analytical method breaks down organic and inorganic carbonaceous materials into fragments of molecules: the opposite of DNA sequencing. When subjected to PGC-MS each type of living organism, from bacteria to animals produces a distinct pattern of molecular fragments. The AI analysis is based on a sophisticated statistical algorithm being trained to recognise ‘debris’ from organic and inorganic carbonaceous compounds according to each sample’s geochemical ‘fingerprint’. Part of the ‘training’ was based on sediments that contain irrefutable fossil samples from as far back in time as the Mesoproterozoic (1000 Ma). Another part was based on definitely inorganic materials, such as carbonaceous meteorites. AI proved able to distinguish biological from inorganic material with a probability up to 0.9 (90%). These results suggested that older, more biologically uncertain material could be assessed.

The AI was able to distinguish general biogenic affinities from inorganic ones in samples with decreasing success going back in time: as high as 0.93 in the Phanerozoic to 0.47 in the Archaean. The oldest samples that reached the probability threshold for this distinction (0.6) were 3.3 Ga cherts from the Barberton Greenstone Belt in South Africa. Another distinction between photosynthetic and and non-photosynthetic affinities among the samples that ‘passed’ as probably biotic reached the 0.6 probability threshold at 2.5 Ga for a sample from South Africa. Non-photosynthetic, but still probably biotic samples extend as far back as 3.5 Ga in South African and Western Australian Greenstone Belts.

Although Wong et al’s preliminary exploration with their novel approach doesn’t take us beyond the current 3.4 to 3.5 Ga age for the earliest tangible suggestions of life. However, they note ‘…our sample inventory is notably lacking in ancient abiogenic samples’. This is a good indication of the promise for further progress that the approach offers. Previous research has sought intact biogenic molecules, with not a great deal of luck, over several decades. Their final conclusion is ‘…information-rich attributes of ancient organic matter, even though highly degraded and with few if any surviving biomolecules, have much to reveal about the nature and evolution of life.’ They have opened a very important avenue in palaeobiological research , as their methodology seems capable of fine tuning to all manner of pro- and eukaryote biochemical distinctions. It could even be used with extraterrestrial material, should we ever get any …

See also: Walsh, E. 2025. Researchers report earliest molecular evidence of photosynthetic life. Chemical & Engineering News, 18 November 2025.