Neurodivergent behavior gives insight to neural noise

Human motion exhibits distinct statistical properties approximated with two principles. Fitt's law states that the standard deviation of a biological signal is proportional to its mean (Fitts PM 1954), and the variance minimization principle states that variance of human motion is minimized (Harris CM, Wolpert DM 1998). Individuals with autism spectrum disorder (ASD) exhibit larger deviation from their mean motion than typically developing (TD) individuals (Torres EB et al. 2013). Millisecond measurements of the velocity micro-statistics within directed pointing motions by participants qualify as a biomarker of ASD (Wu D et al. 2018). Per contra, motion models utilizing the minimum variance principle fail to encapsulate these properties in both TD and ASD participants. Our work builds a new class of motion models using a dynamic phase-space optimization principle of the discrepancy between planned and observed states that both better predict participants' motions for a wider range of scenarios. Based on the free energy principle, which hypothesizes that the brain works to minimize the surprisal of events (Friston K 2010) and using the dynamic programming principle for stochastic control, where the whole is the optimal solution of each of its sub parts given some reasonable conditions (Nisio M 2015), our model encapsulates ASD as a probability cloud of solutions to equations with parameters governing the internal brain dynamics.