Can AI Detect Life? Lessons from Artificial Life
This work warns astrobiologists and AI researchers that current machine learning approaches for life detection are unreliable due to out-of-distribution generalization failures, potentially leading to significant false positives in extraterrestrial sample analysis.
The paper demonstrates that modern machine learning methods for detecting life in extraterrestrial samples are easily fooled into producing near 100% confidence false positives when analyzing out-of-distribution samples, such as those from artificial life, highlighting a fundamental flaw in applying AI to life detection.
Modern machine learning methods have been proposed to detect life in extraterrestrial samples, drawing on their ability to distinguish biotic from abiotic samples based on training models using natural and synthetic organic molecular mixtures. Here we show using Artificial Life that such methods are easily fooled into detecting life with near 100% confidence even if the analyzed sample is not capable of life. This is due to modern machine learning methods' propensity to be easily fooled by out-of-distribution samples. Because extra-terrestrial samples are very likely out of the distribution provided by terrestrial biotic and abiotic samples, using AI methods for life detection is bound to yield significant false positives.