Our Approach

The Problem

The brain is an organ that generates actions based on an organism’s goals, past experiences, and current sensory information. How does this evolutionarily designed machine work? What algorithms does it use to produce successful behavior? These questions are difficult because all of these elements, goals, experiences, and sensory stimuli, come in many different varieties. One simplifying approach is to reformulate behavior as a series of choices between possible options. 

Saccadic Choices

In mammals, and especially primates, one particular type of choice is made all the time, at a rate of about 16,000 choices per hour: the choice of where to look next. Our eyes explore the visual world mostly by making rapid, stereotyped movements, called saccades, from one fixation location to another. This is because our visual system can analyze the information around the fixation point with much higher resolution than that in the periphery. Although we typically generate saccades without much thought, each one of them is preceded by a brief yet complex competition process whereby various potentially informative locations in space are considered until the next target to look at is selected. This occurs in areas of the brain dedicated to oculomotor control. 

The goal of our research is to understand how saccadic choices are made, and because this process varies not only according to the visual information that is available at each point in space, but also depending on the individual’s current goals and past experiences, this is likely to provide valuable intuition about how brains operate in general. In our lab we use novel choice tasks, electrophysiological recordings from oculomotor circuits, and computer models and simulations to determine how the activity of single neurons relates to the visually guided choices made by a participant. 

These famous eye tracking data from Yarbus (1967) provided early evidence that goals have a strong influence on eye movements. White traces correspond to eye position from a single observer viewing a painting (They Did Not Expect Him, by IP Repin). Dots correspond to fixations and lines to saccades. The conditions were "remember the clothes worn by the people" (left), "estimate how long the visitor had been away from the family" (middle), and free viewing (right).

Studying Urgent Choices in the Laboratory

In a typical task in our laboratory, two gray circles appear on a computer monitor, one turns green and the other red, and the participant is instructed to look at, say, the red one — in which case red is target and green is distracter. Across trials, the location of the target varies unpredictably, and we measure the outcome (correct/incorrect) and how rapid the response is (i.e., the reaction time). Although similar tasks are used by many labs, what makes our approach unique is urgency: participants must respond very quickly, often before knowing which circle is target and which distracter. This mimics what often happens in real life, when relevant visual information arrives spontaneously (say, a stoplight turns yellow) while the oculomotor system is busy planning the next saccade (say, to a car on an adjacent lane). 

This dynamic allows us to determine exactly when the relevant visual event, the circles turning red and green, informs the participant’s choice (see dashed line in panel b of the figure below), and whether an eye movement was a guess driven by the initial ongoing plan or an informed choice guided by the new visual information. This way, depending on when and how a particular neuron responds during the task, we can circumscribe its potential contribution to the observed behavior. For instance, some neurons may respond to the colored circles early enough to inform the target selection process, whereas others may respond too late to causally influence the choice. Finally, by simulating in a computer the activity of populations of neurons we can test whether their hypothesized contributions can explain in detail the behavior of the participant. 

The compelled saccade task is our prototypical urgent-choice task. (a) Task schematic. Each trial starts with the participant fixating on the central spot (Fixation), whose color indicates the color of the eventual target (red, in this example). Crucially, the go signal (offset of the fixation spot ; Go) is given first, before the identities of the target and distracter are revealed (Cue), and the subject has a limited time to respond; the reaction time (RT), measured from the go signal to saccade onset, must be < 450 ms. Task difficulty varies randomly across trials depending on the time gap between the go and the cue (Gap; 0–250 ms), but the key quantity that determines performance is the processing time (PT, computed as RT – gap), which is the maximum amount of time available for viewing the cue in each trial. (b) The tachometric curve, which characterizes performance in the task, describes how the percentage of correct responses depends on processing time. When the color cue is seen for less than 120 ms or so, most responses are guesses, so performance is at chance (50% correct). However, the success rate increases rapidly thereafter, approaching 100% after about 200 ms of cue viewing time. The dashed line marks the rPT that corresponds to 75% correct. It approximately divides guesses from informed discriminations.

Real-Life Urgent Choices

We think that our urgent laboratory tasks are good models for real-life situations in which responding quickly (and accurately) is essential, as during a sword fight, or when a cat is hunting a mouse, but sports provide many examples of the rapid sensory-motor interactions that characterize urgent behavior. To perform successfully, a batter trying to hit a 100 mph fastball, a tennis player trying to return a 140 mph serve, or a soccer goalie trying to stop a 70 mph penalty must all master a similar skill: being able to process the relevant visual information about a ball's trajectory in a very short amount of time (about 400 ms in all three cases) while preparing to move, or while already moving. Soccer goalies illustrate this best because they try to anticipate which side the ball will go to, and their commitment is often obvious. Although an incorrect guess makes them look silly, a correct guess gives them the opportunity to adjust their movement on the fly to block the ball. To have a chance of stopping the ball, the early commitment is a necessity. Similarly, a participant performing the compelled saccade task must prepare to move as soon as the go signal is detected; otherwise, if the response takes too long, it is considered an error, regardless of which side was chosen.

Professional coaches often articulate such strategies quite explicitly. For example, Richard Millman, a fantastic squash coach, constantly emphasizes movement preparation ahead of the next shot: "go to where the ball is going to be!" That is what allows players like "The Artist," Ramy Ashour, to make such otherworldly shots as the one captured below.