Today there are more things fighting for our attention than ever before. We are constantly juggling work, family, and social lives. What we don’t realize is, try as we might to multitask, our attention span is quite limited. Recent findings suggest that we can hold about four items (plus or minus 1) in our short-term memory at the same time (Johnson 110). This means that we only have the capacity to focus on a few items at any given moment. Difficulties may arise when doing something that engages other parts of the brain such as driving a car. In our busy lives, with our email, phone, and navigation at our fingertips, using a mobile device while driving is common. A recent study for the European Commission “estimated that driver distraction is likely to be a factor in 10–30% of all road collisions in Europe each year” (IAM Roadsmart ii). In the past 10 years, hands-free phone laws have been established across the U.S. and car manufacturers have responded by making hands-free calling standard in most in-car “infotainment systems”.

In 2014, Apple launched Carplay, a seamless way to integrate your iPhone to your car’s infotainment system. The system not only included hands-free calling, but also voice control for other functions like music selection and navigation via Siri (Apple, Inc.). Studies have shown that human working memory is limited and it has been proven that using a mobile device while driving increases cognitive demand. With this implication in mind, and further development of voice control functions, Apple Carplay is on track to help drivers safely interact while driving.

Defining Working Memory

Two separate types of memory — long and short term — have been defined as early as 1883 by Francis Galton. In 2012, Alan Baddeley published his comprehensive findings on working memory as a way to more accurately portray what goes on in short-term memory. He postulated that short-term memory was the “simple storage of information” while working memory (WM) “implies a combination of storage and manipulation” (Baddeley 5). Baddeley proposed a system for working memory that consisted of four parts: the Central Executive, the Visio-spatial Sketchpad, the Phonological Loop and the Episodic Buffer. The Central Executive is the central controller. The Visio-sketch Pad was distinguished as a short-term visual information store where there was very little capacity. The Phonological Loop thought of as the store for verbal information. The Episodic Buffer being the link across all sections of WM. The Central Executive is the portion in charge of attention. He theorized that the prefrontal cortex is where the working memory functions (Baddeley 3).

Defining Cognitive Load

When discussing the limits of working memory and driving, one must consider what causes distraction. A combination of visual, manual, and cognitive components make up a distraction. Driving can be done safely when the task has the full attention of the working memory. Distracted driving is when the attention of the driver is diverted from the functions that constitute safe and effective driving towards a competing stimulus. This may result in a lack of attention to tasks critical for safe driving. For example, while driving a car, you have to monitor the road, read road signs (visual), and steer the wheel (manual). If the cognitive component, usually regarded as a “general withdrawal of attention”, or a wandering mind comes into play, the driver has become distracted. Often used interchangeably with cognitive distraction, cognitive load is defined as the amount of energy being demanded by an alternative task. The concept of cognitive load. factors heavily in studies conducted on the implications of working memory and driving (Reimer et al. 454–468).

Multitasking and the Bottleneck Effect

Meyer and Kieras (1997) completed research on executive cognitive processes and multiple-task performance. Based upon their work, they theorized that when presented with multiple tasks, the brain cannot fully focus. If multiple tasks are performed, the tasks are prone to error and take longer to complete, perhaps even doubling the time they take to complete (Meyer and Kieras 63–87). Their basis for this is that the brain must restart and refocus on each task and the time in the middle of the swap between tasks leaves gaps inefficiency. So, not only does one do each task less effectively but they lose time in the process.

A study completed by Ruthruff (2001) focused on a bottleneck effect that happens when people try to carry out two separate response tasks simultaneously. Similar to the findings by Meyer and Kieras, Ruthruff found that responses were delayed for one or both tasks. Additionally, he noted that the second task is often slower at first, then speeds up when the stimulus is reduced. The finding is known as the “psychological refractory period (PRP) effect” (Ruthruff 73).

Working Memory and Driving

What do the limits of the working memory mean for drivers? Research completed in 2012 for MIT, showed that as the cognitive demand on the driver increases, the awareness of their environment decreases. The study used recordings of the drivers’ gaze and driving performance from 108 drivers of equal genders across three age groups, 20–29, 40–49 and 60–69, recruited in the Boston, MA area. (Reimer et al. 454).

A mid-size sport car (Volvo XC90) was instrumented with sensors including cameras for capturing driver behavior, vehicle surroundings and audio recording. An eye-tracking system was also set up. The Experiment was conducted on a major highway where the speed limit was 65mph. The drivers were presented with low, moderate and high levels of cognitive workload. For the low level, the driver was given a set of numbers and was asked to repeat those numbers back out loud. At mid-level, the driver was asked to hold each number in memory and asked to speak out loud the previously given number. At the high level, the driver was asked to report verbally, the sum of the numbers given. Driving speed, steering wheel position and acceleration data were recorded directly from the car’s built-in sensors or CAN (Controller area network). The driver’s visual attention was determined by eye-tracking x and y coordinates of the eye position on a vertical plane and a distance as wide as the windshield. Gaze concentration was determined by watching the deviation from standard gaze coordinates. The study found that horizontal gaze concentration was prominent with increased cognitive demand. Meaning, there was a lack of dispersed gaze. “67% of the drivers narrowed the distribution of their eye fixations during the highest-demand period relative to the single-task driving period” (Reimer et al. 465). Researchers studied “gaze concentration” of drivers and found that their awareness of surroundings were affected by increases in cognitive demand. They concluded that developers of advanced driver support systems should consider gaze concentration as a measure of driver cognitive workload because gaze measurements detect visual demand (454).

Apple Carplay and Cognitive Performance Study

IAM Roadsmart, the UK’s largest independent road safety charity, released a report in January 2020 where gaze concentration was one of the criteria used for examining driver safety. This study provides more recent insight on the effect Apple Carplay has on the cognitive performance of drivers. The study used a driving simulator equipped to measure the driver’s “reaction time, eye gaze behavior and vehicle control” (IAM Roadsmart 3). Participants drove the vehicle through 3 trials, one “control drive” where they did not interact with the infotainment system, one “voice-enabled drive” where they performed tasks using only the voice-activated features, and one “touch-enabled drive” where they performed tasks using the touch screen. While driving they were asked to complete five system tasks, using either voice function or touch, including:

• Playing music on Spotify
• Playing BBC radio
• Navigating
• Receiving text messages
• Making a call

While they completed these tasks, they were tested on their reaction time by responding to red bars displayed on the screen. “The main purpose of the red bar was to measure reaction time to an external stimulus whilst interacting with the infotainment system” (IAM Roadsmart 43).

The text message that was sent to participants was “Hi. It’s Rosie. Looking forward to seeing you later. Can you please bring with you some apples, stamps, a thank you card, sun cream, onions, coffee and hair gel?” (IAM Roadsmart 7). The second task asked for the participant to make a phone call. During the call, they were asked to recall as many items on the list as possible. This list contained 7 items, which is pushing the limit of working memory.

The results of the study showed that out of 19 participants, and 76 reaction time events, cumulatively they failed to respond to the red bar 14 times during the touch task and 8 times during the voice task and twice in the control drive. This data shows that interacting with Apple Carplay using touch “interfered with the participants’ ability to respond to the red bars” (IAM Roadsmart 27). Reaction times for the Spotify (figure 1) and radio task were slowest when using touch, but the reaction times for the navigation task were slowest using voice (figure 2).

Overall the average driving speed during the touch task was significantly below the posted speed limit. The standard deviation was about 7mph (29–30). During the call and text portion of the drive, the participants had trouble maintaining lane positioning during both touch and voice tasks. When asked to recall the grocery list, 9 participants recalled 3 items during the voice task and 6 recalled 4 items during the touch task. This is consistent with the limits of working memory. Participants reported feeling less distracted overall during the voice portion of the drives. During the touch task, eye-tracking data showed the participants looking at the infotainment system the most during the Spotify task, navigation task and reading text message/make a call task. Time spent looking at the display was significantly lower during the voice portion.

NHTSA (National Highway Traffic Safety Administration) guidelines state that the sum of all visual fixations away from the road should be less than 12 seconds. The three tasks listed above — Spotify task, navigation task and reading text message/make a call tasks — during the touch task portion of the drive were above this guideline. The study also found people tend to significantly reduce their speed when distracted. This points to the “psychological refractory period (PRP) effect” mentioned earlier, in that when the human brain attempts to multi-task, one of the tasks is likely to be done slower (Ruthruff 73–80). It is clear, working memory has a limited capacity to take in information. With driving taking up a lot of attentional resources, there is little room for additional information to process. When the cognitive load is too high, this results in poor driving performance.

Implications in Design for Apple Carplay

The limits of working memory and cognitive load while driving outlined here have clear implications for Apple CarPlay. There are items that Apple has done to address the concerns of driver’s safety. First, the implementation of voice control via Siri to do any task on CarPlay is notable. This ensures that the driver can maintain focus on the roadway without needing to look down at or touch their iPhone. The IAM Roadsmart study found that voice control was less likely to cause a major distraction than touch control (IAM Roadsmart). Second, the developer guidelines released by Apple make it clear that the driver’s safety is top of mind. For example, they will only allow apps that have Siri capabilities to appear in CarPlay and won’t allow any instruction to pick up the phone to complete a task. Third, with the release of IOS 14, Apple widened the scope of apps it would allow appearing in CarPlay to include EV Charging, Parking and Quick Food Ordering (Apple, Inc.). The increased options have the potential to help users with complex tasks while driving.

It is clear that Apple’s guidelines direct the developers to design for the driver, but is that enough? The studies presented here are evidence that the strength of CarPlay lies in Siri. Even still, based on current research, using voice assistants while driving increases the cognitive load on the driver. If this is to improve, Siri is going to have to be smarter and more seamless.

The first step in the process of improving CarPlay is to compel drivers to use Siri instead of the touch interface. The participants in the IAM Roadsmart study were given a survey in which the majority self-reported their personal preferences were to use the touch interface over the Siri function (see table 5).

A 2019 study done by Raluca Budiu for the Nielsen Norman Group says, “the limited usage [of virtual assistants] reflects the poor usability of these assistants, which are still far from addressing real user needs” (Budiu). This means the Mental Models drivers have of Siri need to be improved, which is not an easy task. According to Budiu, once the user has a mental model of an IA (Intelligent Assistant), they will not often try to initiate new tasks. Users can be taught a new mental model, but they must be driven or inspired to do so. This is where machine learning could come into play to make personalized suggestions based on time and location while driving. This technology already exist in the latest IOS but is still fairly new. If Siri were able to learn when the driver was on a certain route at a certain time that they usually make the same text or call, then it would start to suggest that action verbally or with a simple on-screen notification. Learning the user’s everyday patterns could help mitigate some of the most common tasks they use CarPlay for (e.g. making a call, sending a text, navigating, playing music, etc.), It would also help with preemptive tasks, such as prompting the driver to set up their map destination and music selection before they pull out of the parking spot. In Budiu’s study, one user said of their desires for Siri, “I kind of want it to be more of a personal assistant instead of like an ask-type situation. I want it to come to me more than I come to it” (Budiu).

Future Implications

A study done by the NHTSA found, “five seconds is the average time a drivers’ eyes are off the road while texting. When traveling at 55mph, that’s enough time to drive the entire length of a football field blindfolded” (Lemaster-Sandbank et al. 2). It is clear, there needs to be a more seamless way to use voice activation while driving. If and when self-driving cars come to life, there will be less of a need for fully voice-activated experiences. What will that transition look like? What will happen to the older, human-controlled cars? Perhaps CarPlay would be something that would be regulated in any manual car. These are topics of further investigation. Until then, Siri and other IAs must be improved, to build trust with the user and become a more seamless integration for safety in driving. With these advancements, Apple Carplay has the potential to be a safe and effective way for drivers to operate a vehicle while interacting with their iPhone with minimal distraction.