Our brain uses past experiences, integrated over multiple timescales, to shape how it makes decisions in our uncertain and dynamic world. These adaptive processes can take many forms, across both conditions and individuals, with different combinations of costs (like processing time) and benefits (like flexibility) that can make them difficult to compare and benchmark. Here I will describe our recent efforts to characterize the effectiveness of decision processes with respect to the complexity of model they use to convert past observations into useful predictions that can guide choices. I will show that this approach: 1) has a solid theoretical foundation using concepts drawn from physics and other fields; 2) can account for substantial individual variability of human subjects performing certain decision tasks; and 3) leads to quantitative predictions about the most efficient and effective solutions to a host of decision problems according to a fundamental “law of diminishing returns” relating accuracy to complexity. I will then show that these notions of complexity can be encoded in pupil-linked arousal systems that, in turn, may reflect the influence of neuromodulatory systems like the locus coeruleus-norepinephrine system on coordinated neural dynamics that can affect how information is integrated over time.
The brain can be viewed as a probabilistic estimator, where sensory statistics bias judgments. We addressed this question in the neural pathway supporting the owl’s sound localization. In the owl’s auditory midbrain sensory evidence on sound direction is weighted by its reliability to generate an adaptive motor command for head-orientation. This coding can emerge by convergent projections from a map of space onto premotor neurons controlling behavior. Thus, the topographic sensory representation of auditory space can be read out to adjust behavioral responses by statistics of the sensory input. These experimental results indicate that sensory statistics are both represented and anticipated in the owl’s brain, and built into premotor signals. Experimental results on human subjects yields consistent results, providing evidence towards convergent coding strategies underlying orienting responses across species.