Can AI be trusted in the search for extraterrestrial life?

Artificial intelligence can easily make mistakes, mistaking non-living objects for signs of life on other planets. This is the conclusion reached by two researchers from the University of Michigan who tested AI on simulated life forms in a computer program, reports Space.com.

What the problem is

Lead researcher Christoph Adami noted that AI has already been shown to have a serious Achilles' heel when it tries to classify objects that are unlike those on which it was trained.

The scientists used the Avida program — a digital life simulator where organisms are represented as code capable of reproducing and competing for resources. This made it possible to create a large set of examples of "life" and "non-life" for testing.

In the experiment, the AI was tricked after just 15 changes in the sequence of "molecules." The model would begin to assert with absolute confidence that it was seeing life, when in fact it was a non-living sequence. And this occurred with any initial dataset.

Why this is critical for the search for life

AI is trained on Earth-based samples of life. However, extraterrestrial life may differ significantly in chemical composition and processes. Such "out-of-distribution" data can lead to false positive results.

This is particularly important for future missions. If a rover or probe to Europa, Venus, or exoplanets relies on AI to analyze spectrometric data, the risk of error increases.

Adami warned that if AI is trained on Earth's biotic and abiotic samples, it may return a positive verdict when no life actually exists.

What's next

The next step is testing on real data, not just in digital simulation. The scientists plan to present their results at a conference on artificial life in August 2026.

In brief

Researchers from the University of Michigan have shown that AI can easily be tricked into mistaking non-living objects for signs of life. This happens when data is "out-of-distribution" — differing from what the model was trained on. For the search for extraterrestrial life, this creates a serious risk of false positive results. Scientists warn that trusting AI in such missions without careful verification is premature at this stage. The results will be presented at an artificial life conference in August 2026.