Our understanding of the world is facilitated by mental models, useful conceptualizations that allow us to compartmentalize knowledge. A broad definition of a mental model is “any thought process in which there are defined inputs and outputs to a believable process which operates on the inputs to produce outputs.” Therefore, a mental model might help you make judgments of causality, e.g. “if the space heater tips over, it will turn off automatically”. A causal understanding facilitated by our mental model allows us to quickly grasp the ins and outs of a new system.
Mental models are especially important in the face of emerging technologies. They allow us to form a quick sketch of the capabilities and risks of a given technology. Given the rapid emergence of autonomous driving systems, there is a growing concern among safety researchers that consumers’ mental models of autonomous systems may not be entirely accurate. This can be problematic as behavioral research shows that mental models can sometimes be difficult to change.
Therefore, it is especially important to iron out kinks in mental models early on, especially when dealing with potentially life-endangering technologies. This is what a recent study by the AAA foundation set out to do.
In this study, different groups of participants were provided with different types of marketing and information materials regarding an L2 driving system (see picture below). Such a system can generally help maintain speed, car distance, and lane position, but is not fully autonomous.
The study had a complex design with a range of interesting questions (see here for the full report). Here, we will focus on the distinction in branding and marketing the systems. The distinction was made through alternate naming of the same underlying technology, DriveAssist, and AutonoDrive, respectively. The DriveAssist condition emphasized the system’s limitations and driver’s responsibility, whereas the AutonoDrive emphasized the capabilities of the system and driver workload reduction. However, neither conditions omitted any safety information, nor introduced any false features. See below.
After going through the training materials, participants completed a survey assessing their behavior in different driving situations. The participants were instructed as such “indicate whether or not you expect DriveAssist/AutonoDrive to take action and avoid a collision, without the driver doing anything.” The results showed that participants in the AutonoDrive condition had higher levels of confidence in the system’s capability. An example can be seen in the graph below.
As the graph shows, participants tended to overestimate AutonoDrive’s capabilities. This overestimation was seen in other scenarios as well. For example, 66% of AutonoDrive participants incorrectly guessed that their system could detect and respond to adjacent vehicles, compared to only 16% in the DriveAssist condition. Not only this, the AutonoDrive participants had a greater willingness to drive under compromised conditions, such as while eating or talking on the phone.
So far, we have looked at hypothetical survey responses, however, this study doesn’t stop here. Next, participants actually took to the highway with the autonomous system in place. Remember, however, that the actual system is the same, albeit with different names and expectations.
In this part of the experiment, unbeknownst to the participant, different driving behaviors were recorded. For example, the amount of time hands and feet were away from the controls. In line with expectations, the AutonoDrive participants had their hands away from the steering wheel 88% of the time, compared to 77% in the DriveAssist condition. A similar result was observed for time feet were away from the pedals. Thus, participants differed both in their understanding of the system’s abilities to control lane position and speed.
There are some interesting exceptions in the study worth noting. The amount of time spent using the autonomous system feature did not differ across the two conditions. Therefore, a more cautionary approach to branding did not reduce the usage of the feature itself! I suspect this may be because the product “speaks for itself”. In general, the participants were excited by the technology and had a greater willingness to use the systems, irrespective of the condition. Though, this may serve as a cautionary tale, as despite the initial differences due to marketing, in some cases participants were actually more inclined to overestimate the systems’ capabilities. The authors speculate that a 30-min introduction to this technology in controlled settings may not have been enough to form accurate mental models. This is in contrast with a previous study where confidence in the system actually went down for the ‘exaggerated autonomous’ condition.
In summary, this study shows that branding strategy can heavily influence consumers’ perceptions of the capabilities of autonomous driving systems. It stresses the need for a balanced approach where limitations are pointed out. Such an approach is especially necessary for emerging technologies that tend to increase expectations due to their newness. This way, the consumer’s mental models can be properly instantiated, preventing breaking of trust down the line when expectations are not met.
Autonomous vehicles will usher in a new wave of experiences to the coming generations. In my view, automobile companies must bear the burden of consumer education and training to supplement the emergence of this new tech. This includes conducting experiments such as the one mentioned here on-site, and not once the product is already in the hands of the consumer.