Integrating salience and value in decision making

Abstract

Chesley Sullenberger faced a critical decision. Three minutes into its ascent from New York's LaGuardia Airport, US Airways Flight 1549 had struck a flock of geese, resulting in an immediate and complete loss of thrust from both engines. Losing altitude quickly, Captain Sullenberger was confronted with two unnerving options: risk further flight over dense urban ground to a nearby airfield in New Jersey, or attempt an unpowered ditching into the waters of the Hudson River below. He had available multiple sources of information-sensor readings, air traffic control transmissions, his own considerable experience-but precious little time. How our brains quickly transform such diverse information into a single choice is the focus of considerable study in neuroscience. In PNAS, Towal et al.(1) reveal the dynamic process by which human choosers combine two possible sources of information, visual salience and goal valuation, during the decision process.The temporal dynamics of decision-making have been widely characterized by a class of processes known as sequential sampling models (2, 3). In such models, the decision process integrates noisy information by repeated sampling over time, and a decision is reached when the accumulated information reaches a threshold. Various implementations differ in specific details, for example using individual option-specific accumulators or a single accumulator representing net evidence, but the common goal is to predict both the accuracy and timing of choices.The appeal of sequential sampling models is threefold. First, such models provide a robust fit to empirical choice and reactiontime data, most prominently in binary perceptual decision tasks and for short decision times. Second, under certain assumptions, these models approximate the optimal solution to decisions with accuracy and time constraints, providing a quantitative framework for understanding based in statistical decision theory (2-4). Third, electrophysiological studies have identified neural activity consistent with evidence integration in frontal, parietal, and subcortical circuits, suggesting that bounded accumulation characterizes neural as well as psychological decision processes (5-9).Traditional sequential sampling models integrate the specific sensory quantity required for a given decision, such as the net motion information of a noisy moving-dot stimulus in a direction discrimination task. However, real-world choices, like the one facing Captain Sullenberger, often involve multiple streams of information, encompassing both sensory stimuli and nonsensory sources like memory. A central tenet of decision theory is that all relevant information is integrated into a single decision variable (10, 11), a quantity referred to as value and closely related to the economic notion of utility. A large body of theoretical and empirical work has examined value-based choices in psychology and economics, largely apart from the perceptual decision-making literature. However, in the past few decades neurophysiological experiments h...