The Memory Glasses: Subliminal vs.Overt Memory Support with Imperfect Information Richard W. DeVaul, Alex “Sandy” Pentland Massachusetts Institute of Technology Media Laboratory {rich,sandy}@media.mit.edu Vicka R. Corey Harvard Medical School A. A. Martinos Center for Biomedical Imaging vicka@nmr.mgh.harvard.edu Abstract Wearables are frequently designed to support users en- gaged in complex “real world” activities, ranging from food inspection to ground combat. Unfortunately, wearables also have the potential to interfere with the very tasks they are designed to support, either by distracting the user or providing them with misleading information. In 2002 we published a pilot study suggesting that a subliminal visual cuing system might be an effective low- attention interaction strategy for just-in-time memory sup- port. In this paper we present the results of a larger study demonstrating that not only is wearable subliminal cuing significantly effective (increasing performance by a factor of approximately 1.5,p =0.02), but even incorrect sublim- inal cues can actually improve performance. By contrast, consciously-visible incorrect cues caused performance to degrade. 1. Introduction An HCI designer expecting a desktop environment can assume that the user is probably sitting down someplace safe, with few extraneous stimuli competing for their at- tention. By contrast, interaction design in the wearable- computing domain must take far more into account. The user may be driving a car or engaging in combat, and such other tasks will divide their attention and limit their re- sources for interacting with the computer. Wearable computing applications are frequently in- tended to provide support for complex real-world activities, from food inspection[14] to service and maintenance[24] to military command and control (SAAB WISE project, US Army Objective Force Warrior program). However, the use of such wearable task-support applications can interfere with the performance of the very task they are designed to support. They can be distracting, taking too much of the user’s attention away from the task at hand. Or they can ac- tively misdirect the user, giving wrong information or bad advice. The risk of misleading the user is one task-support ap- plication designers should not ignore. Real-world environ- ments are complex, and the information available to wear- able task-support applications is likely to be limited and im- perfect. Unsurprisingly, our research suggests that presenting the user with incorrect information can result in a decrease in task performance. However, in at least some cases this effect may actually be inverted — improving user perfor- mance even when the computer’s cue is incorrect — if the cue is presented subliminally, rather than consciously visible. (We discuss this surprising result further in Sec- tion 4.7.1 below.) More complex than the problem of misdirection is the problem of divided attention. Contemporary theories of attention[4, 25] are based on finite resources of perception and cognition. Simultaneous tasks that require conscious attention compete for a person’s limited attentive resources. The tasks interfere with one another, and the person’s per- formance suffers on all of the tasks. Interference issues are particularly problematic for wear- able task-support applications. As the complexity of the real-world task increases, so does the need for task support. But at the same time, the user’s capacity for attending to a task-support application decreases. Effective support is hardest when the user needs it most. Task interference by a support application can be mit- igated by reducing its perceptual or cognitive demands. One important technique for reducing interaction complex- ity is context awareness. This is the use of non-explicit user and environmental input, typically acquired through sensors[16, 3, 22]. Context awareness can reduce, or in some cases eliminate, the need for explicit user input for some applications. However, context awareness can only directly address half of the interaction problem — the part about getting in- Proceedings of the Seventh IEEE International Symposium on Wearable Computers (ISWC’03) 1530-0811/03 $ 17.00 © 2003 IEEE