NASA AI life detection faces a fake-alien warning from researchers
Michigan State researchers fooled a life-detection algorithm, raising cautions as NASA funds AI tools for future space missions.
By Bianca Rossi · Entertainment Editor
3 min read
A NASA AI life detection future is coming into view, and Michigan State University researchers say their latest experiment shows why space agencies may want to keep the robot brains on a short leash.
The team trained an artificial intelligence system to sort digital “organisms” that could reproduce from code that could not, using a program called Avida. In ordinary tests, the model was nearly spotless, making the right call 99.97 percent of the time, according to the researchers.
Then the scientists went hunting for the weak spot. They repeatedly altered nonliving bits of code in small ways, pushing the AI toward greater confidence that the samples were alive. The result: the system became close to fully certain that many impostors were living, even though the code did not replicate, the behavior used as the definition of life in that digital setting.
Can AI make NASA think it found alien life?
The Michigan State work suggests an AI system can confuse familiar-looking patterns with actual life, especially when it meets something outside its training. That matters for space missions because extraterrestrial life, if it exists, may not resemble the examples scientists used to build the model.
Ankit Gupta, a doctoral student in computer science and engineering, told Mashable that AI is “highly accurate” with ordinary examples, but can misclassify unusual ones with confidence. Gupta and coauthor Christoph Adami, a professor of microbiology and molecular genetics, physics, and astronomy, warn that the issue could affect smart instruments sent to Mars, icy moons, or other worlds to look for signs of life.
The researchers are scheduled to present their findings in August at the 2026 Conference on Artificial Life in Waterloo, Canada.
NASA and its partners are already putting money behind machine-learning systems for astrobiology. One NASA-funded project led by Carnegie Science’s Michael L. Wong and NASA’s Caleb Scharf is building tools to detect life-related patterns in complex chemical data.
That effort is backed by a $5 million grant and aims to train AI on at least 1,000 samples, including meteorites, rocks, fossils, and living organisms. Some data comes from real instruments such as mass spectrometers, the kind already used on NASA missions.
Why some scientists are less worried
Wong told Mashable he does not see the Michigan State test as proof of a major flaw in AI life-detection work. His point centers on how the fake examples were made: the researchers used a selection process to refine the code until it appeared more lifelike.
“I don't think it's that bad, to be honest,” Wong told Mashable, adding that he does not know of a physical environment that would select for objects that only look alive while lacking lifelike behavior.
Still, Wong said the study sharpened his thinking about training data. His group wants models to learn from living things, and also from products of life such as fossils, biological residue, and waste. If a world like Mars preserves only chemical traces of an extinct biosphere, a system trained too narrowly could miss the clue.
Wong’s team also tries to trick its own models. In one test, he said, an algorithm flagged a sea squirt sample as photosynthetic, even though sea squirts do not make food from sunlight. The apparent mistake turned out to point to algae living in the animal’s tissues.
The Michigan State team remains cautious about putting such systems on spacecraft. Gupta told Mashable: “We can wait. Clearly, we are not there yet.”
This story draws on original reporting from Mashable.