My research lies at the intersection of computer science and psychology. My work primarily focuses on multitasking and interruption in complex domains, such as driver distraction and human-computer interaction, using computational cognitive models to simulate human behavior. Please click on the tabs below for more information about each topic.

What is a cognitive model?
A cognitive model is a computational process (akin to a computer program) that aims to think and behave like a person. While many flavors of cognitive models exist today, the most popular are those developed in the context of a cognitive architecture—something like a computer programming language that incorporates the abilities and limitations of the human system (e.g., accounting for how people forget information, or the limits of their hand or eye movements). Cognitive models developed in a cognitive architecture (such as ACT-R, Soar, and others) then inherit the predictions of the architecture and thus more closely represent human thought and behavior.
How do cognitive models simulate thoughts and behaviors?
Modern cognitive models, especially those developed in cognitive architectures, generally run as computer simulations along with a simulated task environment. For example, our own models of driver behavior interact with a driving simulator and thus must deal with all the intricacies of the driving task (like handling the steering wheel and pedals in the face of realistic vehicle dynamics). Some cognitive models even interact directly with the external world using computer vision and robotic movements.
What kinds of tasks can cognitive models perform?
Cognitive models have been developed to perform all kinds of tasks ranging from simple psychological experiments to complex real-world tasks. Some models have accounted for basic psychological phenomena, such as list memory, time perception, visual search, and analogy. Other models have focused on applied tasks including driving, mathematical learning, complex decision-making, and air-traffic control. For example, the ACT-R web site lists applications of the ACT-R architecture for a variety of tasks. Our own laboratory has focused on developing cognitive models of tasks related in some way to human multitasking, as described in our recent book.
What's so interesting about multitasking?
Multitasking comes in a variety of flavors, but generally speaking can be thought of in two ways. Concurrent multitasking occurs when these tasks occur simultaneously, like the classic example of tapping your head while rubbing your belly. Sequential multitasking involves doing multiple tasks one after another, usually when one task is interrupted for the other task to proceed, like dealing with a phone call or email in the midst of writing an essay. From a scientific standpoint, multitasking is fascinating in that it can be very easy in some situations (like walking and talking) but extremely difficult in other situations (like texting while driving, or listening to two voices at the same time). These ideas are discussed in detail in our book The Multitasking Mind.
How do people do two things at the same time?
We have developed a theory called threaded cognition that aims to explain how people multitask. The theory says that each task can be represented as a thread that weaves its way through the brain's processing resources. Several threads can run independently, especially when there is little to no overlap in terms of the type of processing; for example, if a reading thread is using vision and a movement thread is typing, these threads can largely run without interference from each other. However, there is also a central bottleneck (called the procedural resource) that is needed by all threads, and thus limits the amount of independence between threads. This interplay between independent threads and a central bottleneck allows the theory to account for both our multitasking abilities and the limitations of those abilities.
How do people manage and recover from interruptions?
A recent theory called memory for goals says that, when interrupted, people rehearse the current mental context so that it can later be retrieved from memory when returning from an interruption. Viewed under the lens of threaded cognition, people actually multitask during an interruption: while performing the interrupting task (email, chat, whatever), they maintain a concurrent cognitive thread that performs this rehearsal of mental context (see the book). Thus, interruptions are not only disruptive for the original task, they can even be disruptive for the interrupting task due to interference from this concurrent rehearsal.
When are interruptions most disruptive?
There have been several studies in the past decade reporting that interruptions are most disruptive in the middle of a subtask, when people are trying to hold information mentally while doing the task. Building on these studies, we have recently shown that in the context of deferrable interruptions, people show a strong tendency to defer dealing with interruptions while they are mentally holding information, instead waiting for a point of lower workload when this information is no longer needed.
How do people drive, cognitively speaking?
When steering a car, drivers rapidly scan two distinct visual areas: the lane directly in front of them to keep the car centered, and the lane in the distance to guide smooth steering especially around curves. This information is used to adjust the steering wheel and, when another car is in front, adjust the car's speed as well. We have developed a computational model of steering and lane changing that demonstrates how people can adjust steering roughly 4-5 times per second -- typically adequate for normal driving, but when distracting tasks interrupt these adjustments, performance can quickly degrade.
How does cell-phone dialing affect driving?
There have been a number of studies showing the negative effects of cell-phone dialing (and conversation) on driving performance. In our own study, we found that manually dialing a phone (by pressing keys) was significantly more distracting than voice dialing, leading to an decreased ability to keep the car centered in the lane. Dialing using an address book-style menu also degrades driving. These effects are present for hands-free devices mounted on the dashboard; thus, our studies have agreed with previous studies in that hands-free and handheld phones can both be distracting.
How does iPod use affect driving?
A recent GMAC survey reported that 20% of drivers age 18-24 have used an iPod while driving. Knowing how distracting cell phones can be, it may not be surprising that using an iPod while driving can also be distracting. More surprising is the size of the effect. Our study of iPod distraction found that selecting a song on an iPod can degrade performance almost twice as much as dialing a cell phone. Even more surprisingly, selecting a song can degrade performance twice as much as watching a video on the iPod. The level of distraction for iPod use is severe and warrants further consideration as an important source of driver distraction.
Can just thinking about something affect driving?
Yes! We ran a laboratory study in which people simply memorized a list of numbers, similar to trying to remember items to buy at a grocery store or directions to a new location. Drivers only had to think about this list while driving, and yet they exhibited a 50-millisecond slowdown in their brake reaction times (for a 9-item list). While 50 milliseconds may not seem like much, consider that a car on a highway can travel about 10 feet in that short time, potentially being the difference between a crash and a near-miss.
How can we predict the distraction potential of new devices?
Many areas of engineering have tools at their disposal for making predictions about new devices and machines, such as design systems for automobiles or planes that can predict wind resistance, drag, lift, etc. before the vehicle is actually built. Such tools have been difficult to come by for predictions of cognition and behavior. However, we have developed an initial system called Distract-R in which a design can specify a prototype of a new in-vehicle device, like a new radio interface or cell-phone dialing technique. The system uses a model of driver behavior to predict the distraction potential of this new device based on common measures of driver performance. Our hope is that this system can facilitate rapid prototyping of many possible devices and pare down the number to a few (say 2-4) devices that would then be developed and rigorously tested within real vehicles.