Artificial intelligence could be throwing off the search for alien life just as much as humans’ own cognitive biases, according to a new paper published in the scientific journal Acta Astronautica.

When a neural network was shown an image from a crater on the dwarf planet Ceres, it identified curious patterns, including both a square (which people also saw) and a triangle.

After the neural net detected the triangular shape in the images, people in the study also began to see it, even though they hadn’t previously. It’s an example of how false positives from AI could trip up extraterrestrial studies.

Ever since the Dawn spacecraft picked up images of what look to be a vast network of bright spots in the Occator crater on Ceres — a dwarf planet in the asteroid belt — there’s been conjecture over whether the whiteish spots are made up of ice, or some kind of volcanic salt deposits. Meanwhile, another controversy has been brewing over them: What exactly are those shapes seen in the bright spots, called Vinalia Faculae? Are they squares or triangles? Did extraterrestrials create them?