Use of an automatic tracker as a function of its reliability

Abstract

The present paper reports two studies investigating the use and usefulness of an automated tracker as a function of its reliability. The participants' task was to update the position of several targets when new information was received about the current position of existing targets, new targets, and noise signals. They could update the position of each target manually or they could assign one or more targets to an automatic tracker (AT), which used the same information available to the participant to update the position of the targets it was responsible for. In the first study, the reliability of the AT was varied from totally unreliable to very reliable. Participants' use of the AT and system performance increased as the reliability of the AT increased. However, actual use of the tracker was not a simple function of its reliability. Instead, use appeared to be a function of both AT reliability and the participants' ability to do the task manually. The second study examined system performance and use of an AT as a function of task difficulty (number of targets that had to be tracked) and initial reliability of the AT (high versus moderate reliability). When the task was more difficult, most participants continuously assigned targets to the AT independent of its reliability. There was also a significant correlation between AT use and the percentage of targets tracked. Some participants in the lower task difficulty condition did make less use of the AT, if they received the less reliable AT first. The results of these studies differ somewhat from previous research with automatic controllers, which have found that participants do not tend to use an automated system unless it is extremely reliable. Possible reasons for the difference were task difficulty and the fact that, with this system, participants retained ultimate control of the tracking task. If the AT did not track a target, the participant had the opportunity to handle that target manually. One finding that was consistent with the literature was that when the automated tracker was reliable, some participants failed to recognize errors made by it even with feedback and sufficient time to correct the errors.

Keywords:

• Automated Controllers Human-computer Interaction Trust Reliability