Sound Intelligence's "aggression detection" algorithm deployed by schools reportedly contained high rates of false positive, misclassifying laughing, coughing, cheering, and loud discussions.