>"If bees can perceive zero with a brain of less than a million neurons, it suggests there are simple efficient ways to teach AI new tricks." Why are they just

>"If bees can perceive zero with a brain of less than a million neurons, it suggests there are simple efficient ways to teach AI new tricks." Why are they just speculating about this? Do they think it would be hard to train an image classifier to say "left" or "right" depending on whether that half of the image had more geometric objects in it, then show it one vs none objects to test it? I don't think this would be at all difficult to do, the training data could even be procedurally generated. EDIT: It looks like the test accuracy was only ~60% but from the supplements it took them only 30-50 training examples over the course of 2-4 hrs to reach the criteria of 80% correct: >When presented with the unfamiliar numbers of one versus zero, bees chose the lower number of zero (63.0 ± 2.9%; z score = 4.23, P < 0.001; Fig. 1D), showing an understanding that an empty set is lower than one,which is challenging for some other animals (5, 6, 18). http://science.sciencemag.org/content/360/6393/1124 http://science.sciencemag.org/content/360/6393/1124 [1] http://science.sciencemag.org/content/suppl/2018/06/06/360.6393.1124.DC1 http://science.sciencemag.org/content/suppl/2018/06/06/360.6...