AI can't tell if something is alive if it hasn't seen it before
What happened
New research shows that AI models trained to detect life are easily fooled by samples they haven't encountered in their training data. This means AI will likely give many false positives when searching for life on other planets.
Why it matters
Space agencies and astrobiologists have been exploring using AI to analyze samples from other planets for signs of life. This paper shows that current AI methods are not reliable for this task. It means that any AI-driven 'detection' of extraterrestrial life would need extensive human verification, slowing down discovery and increasing costs.
The signal
Watch for new research that focuses on training AI with more diverse, out-of-distribution data, or for hybrid human-AI systems that explicitly account for these false positives.