AI Unlocks Chemical Traces of Earth’s Earliest Life in 3.3-Billion-Year-Old Rocks

Scientists have uncovered new chemical evidence showing that life existed on Earth at least 3.3 billion years ago, pushing back the detectable molecular record of biology far earlier than previously confirmed. Using a combination of high-resolution chemical analysis and machine learning, an international team led by the Carnegie Institution for Science has demonstrated that even heavily altered ancient rocks still hold faint molecular traces—chemical fingerprints—that reveal whether life was once present.

This research doesn’t rely on intact biomolecules, which almost never survive billions of years of geological heating, pressure, and deformation. Instead, it focuses on tiny molecular fragments created when rocks are thermally broken down in the lab. These fragments contain patterns of chemical distributions that differ depending on whether the original material came from life or from non-biological processes. By training an artificial intelligence system on hundreds of samples, the team taught the model to recognize these subtle differences with over 90% accuracy.

The results reshape our understanding of early Earth and significantly expand the time window in which scientists can detect ancient life.

Major Findings From the Study

The team analyzed rock samples dating back more than 3.3 billion years, representing some of the oldest preserved crust on Earth. Their AI-assisted analysis revealed that these rocks carry chemical signatures consistent with biological activity, even though no visible fossils or intact biomolecules remain.

Researchers also detected chemical evidence related to oxygen-producing photosynthesis in rocks at least 2.5 billion years old, suggesting that organisms capable of releasing oxygen existed almost a billion years earlier than the previously accepted molecular timeline. Prior to this work, reliable molecular traces of life were only confirmed in rocks younger than about 1.7 billion years.

To build and validate their detection method, the scientists compiled a huge dataset of 406 samples that included:

Modern plants
Animals
Fungi
Bacteria
Billion-year-old fossils
Sedimentary rocks
Meteorites
Synthetic organic compounds

This diverse dataset allowed the model to learn the chemical differences between biological and non-biological materials, as well as differences among various types of organisms. For example, the AI could distinguish plant-derived biosignatures from animal-derived ones and separate photosynthetic materials from non-photosynthetic ones.

The core analytical technique used was pyrolysis–gas chromatography–mass spectrometry (Py-GC–MS). Pyrolysis heats a sample in the absence of oxygen, causing long molecular chains to break into fragments. Gas chromatography then separates these fragments, and mass spectrometry identifies them. The resulting chemical “profiles” served as training data for the AI model.

The team’s machine-learning approach used a random forest classifier, which performed consistently well across different tasks, achieving accuracy rates between 93% and 98% on test datasets.

Why This Discovery Matters

The new findings reshape the timeline of life on Earth by nearly doubling the span of time over which molecular evidence can be reliably detected. This has major implications for fields such as geology, planetary science, and astrobiology.

Expanding the Biosignature Time Window

Until now, biosignatures detectable through chemical methods rarely extended beyond 1.7 billion years. This study extends that window to 3.3 billion years, giving researchers access to a far earlier chapter of Earth’s biological history. This period, known as the Archean Eon, was a time when Earth’s atmosphere lacked oxygen and life consisted primarily of microbes.

Reassessing Early Photosynthesis

Detecting photosynthesis-related chemical patterns in 2.5-billion-year-old rocks suggests that oxygenic photosynthesis may have evolved long before the Great Oxidation Event (around 2.4 billion years ago). This event dramatically increased oxygen levels in Earth’s atmosphere, enabling the rise of more complex life. The new evidence indicates that the biological mechanisms for producing oxygen were already in place hundreds of millions of years earlier.

Transforming How We Search for Life

The study shows that life leaves behind persistent chemical echoes, even if every visible structure, biomolecule, or fossil is gone. This has profound implications for the search for life beyond Earth.

Future missions to Mars, Europa, Enceladus, or Titan may not need pristine fossils to detect ancient life. Instead, scientists could analyze the chemical fingerprints in rock or ice samples, looking for the same types of patterns identified in this research. This approach could be applied to:

Martian rock samples collected by NASA’s Perseverance rover
Samples returned to Earth by future missions
Meteorites linked to ancient planetary crust

Because the technique works even on samples subjected to high heat and pressure, it could allow life detection in environments previously considered too altered or degraded.

Understanding Biosignatures: What Counts as Evidence of Life?

Biosignatures are measurable indicators that life was present at some point. These can include:

Shape-based fossils (e.g., stromatolites)
Biomolecules (e.g., lipids, pigments)
Isotopic patterns (e.g., carbon fractionation)
Chemical distributions (as in this study)

The challenge is that with increasing age, more of these indicators disappear. Rocks older than 3 billion years are often metamorphosed—altered by heat and pressure—making fossil preservation extremely rare.

This study highlights that even when the original biomolecules are gone, their degradation products and the relative abundances of certain fragments can still reveal their origin. The idea is similar to how a burned log still has identifiable chemical markers tied to the tree it once was.

The Role of AI in Geobiology

Machine learning is becoming increasingly important in Earth sciences because geological samples contain highly complex chemical mixtures. Traditional statistical approaches struggle to interpret these patterns, but AI can recognize extremely subtle relationships.

In this study, the AI was trained to detect:

Whether a sample was biotic or abiotic
Whether biotic samples were photosynthetic or not
Whether biotic samples came from plants, animals, fungi, or microbes
Whether fossil samples were consistent with microbial mats or with more complex organisms

These capabilities make AI a powerful tool for studying the earliest history of life.

Broader Context: How Old Is Life on Earth?

While this study demonstrates chemical evidence of life 3.3 billion years ago, other lines of research suggest that life might be even older.

Some of the oldest proposed indicators include:

3.5–3.48 billion-year-old stromatolites
3.7 billion-year-old isotopic signatures from Greenland
4.1 billion-year-old carbon inclusions in zircon crystals (highly debated)

However, these claims are controversial because the samples are extremely altered or ambiguous. The new AI-enabled method offers a more consistent way to study such ancient materials without requiring pristine preservation.

The Potential for Extraterrestrial Life Detection

Because the technique works on samples with no visible fossils and no intact biomolecules, it may guide future life-detection missions. For example:

Mars rocks that appear heavily altered could still reveal chemical fingerprints of past microbial life.
Icy moon samples containing organic compounds could be analyzed for biotic patterns even if the organisms decomposed long ago.
Meteorites from ancient planetary crusts could be re-examined for overlooked biological signals.

This opens the door to detecting life in environments once considered geochemically inaccessible.