Use-dependent Biases as Optimal Action under Information Bottleneck

TL;DR

This work reframes use-dependent biases in sensorimotor control as a consequence of information-processing limits, formalized through an information-bottleneck framework with $I(S;A) \le C$ and an optimal stochastic policy $p(a|s) \propto p(a) e^{-\frac{1}{\beta} J(s,a)}$. It shows that, under a quadratic cost and Gaussian action prior, the action distribution biases toward recently repeated actions with a mean $\mu=\frac{\beta s+\epsilon s_0}{\epsilon+\beta}$, and that the bias magnitude scales with $\beta$, the size of the information budget. The authors derive two testable predictions: (i) dominated vs non-dominated hands differ in information capacity, predicting smaller biases for the dominant hand, and (ii) higher movement speeds reduce use-dependent biases due to larger initial-direction costs. Experimental data from a handedness study and reanalysis of a speed-manipulation dataset validate these predictions, supporting a normative, information-processing-based account of motor biases and their relation to handedness.

Abstract

Use-dependent bias is a phenomenon in human sensorimotor behavior whereby movements become biased towards previously repeated actions. Despite being well-documented, the reason why this phenomenon occurs is not yet clearly understood. Here, we propose that use-dependent biases can be understood as a rational strategy for movement under limitations on the capacity to process sensory information to guide motor output. We adopt an information-theoretic approach to characterize sensorimotor information processing and determine how behavior should be optimized given limitations to this capacity. We show that this theory naturally predicts the existence of use-dependent biases. Our framework also generates two further predictions. The first prediction relates to handedness. The dominant hand is associated with enhanced dexterity and reduced movement variability compared to the non-dominant hand, which we propose relates to a greater capacity for information processing in regions that control movement of the dominant hand. Consequently, the dominant hand should exhibit smaller use-dependent biases compared to the non-dominant hand. The second prediction relates to how use-dependent biases are affected by movement speed. When moving faster, it is more challenging to correct for initial movement errors online during the movement. This should exacerbate costs associated with initial directional error and, according to our theory, reduce the extent of use-dependent biases compared to slower movements, and vice versa. We show that these two empirical predictions, the handedness effect and the speed-dependent effect, are confirmed by experimental data.

Figures (3)

• Figure 1: Illustration of an information bottleneck in sensorimotor control. The sensorimotor system can be considered, in abstract terms as an information channel transforming sensory states into motor actions. Limited computational resources of the sensorimotor system can be understood as an information bottleneck between sensory states and motor actions. Mathematically speaking, the mutual information $I(S;A)$ between states $S$ and actions $A$ must not exceed some bound $C$.
• Figure 2: A. Use-dependent biases in reaching movements from a representative participant. B. If we assume that the cost is a quadratic function of reach error, and that the overall distribution of movement directions $p(a)$ is Gaussian, this theory predicts that actual movement direction will follow a Gaussian distribution whose mean is biased towards previously generated actions – exactly predicting the previously reported phenomenon of use-dependent biases. C. Illustration of experimental setup in which participants made planar arm movements while viewing a display.
• Figure 3: A. Bootstrap analysis: histogram plot of the magnitude of the use-dependent biases in 10,000 bootstrapped samples; green area indicates $99\%$ confidence interval of [0.0037, 0.0639] B. Use-dependent biases at Slow and Fast movement speeds. Left, data from wong_motor_2017 showing the effect of movement speed on the magnitude of use-dependent biases. Right, theory predictions. We fit the model to data for fast movements (red) and predicted the change in use-dependent biases when moving more quickly, based on increased directional error costs predicted by optimal control theory.