Drosophila detects negative visual evidence against self-motion

Sensory inputs available to animals are often open to multiple interpretations. When evaluating the likelihood of a certain interpretation, it is important that they consider not only evidence confirming that interpretation, but also evidence against it. Human reasoning often fails to appropriately weigh negative evidence, a phenomenon termed confirmation bias. However, since behaving based on false beliefs is costly, animals may detect negative evidence better in the context of innate, naturalistic behaviors. An example of such behaviors is visual course stabilization. Existing models assume that animals perform template matching over the vector field of local visual motion to detect optic flow and estimate self-motion. However, template matching overlooks an entire class of alternative scene interpretations: that the observer is stationary and external objects are moving consistently with optic flow templates. A potentially useful visual cue to detect the absence of self-rotation is visual patterns that are stationary on the retina. Because all visible features of the world move in the same direction during genuine observer rotation, stationary visual patterns strongly argue against the existence of self-rotation. Here, we demonstrate that the fruit fly Drosophila suppresses its rotational course stabilization behavior in the presence of stationary visual patterns, a manifestation of negative evidence detection. In parallel in silico experiments, we show artificial neural networks trained to distinguish self- and world-motion also learn to exploit stationary visual patterns, supporting the functional interpretation of the fly behavior. Using genetic manipulations and measurements of neural activity, we constrain microcircuit mechanisms of stationary pattern detection, which we find to include a visual neuron type called Mi4. Overall, our results exemplify how the compact brain of flies takes negative evidence into account to improve its heading stability, exploiting geometrical constraints of the visual world.