Younho Seong James Llinas Colin G. Drury Ann M. Bisantz Center for Multisource Information Fusion State University of New York at Buffalo
The study of trust has a long history in sociological literature: however, that history is not rich in empirical studies. Some representative, recent studies of human trust in automation have been performed on a continuous chemical process control simulation ( Lee & Moray, 1992, 1994; Muir & Moray, 1996). This means that the objects which human operators dealt with were machines or displays representing the behavior of machines. Thus, the induced characteristics of human trust in automation were essentially concerned with predictability, dependability, and faith, whose attributes could be easily captured from the behavior of machines. Current automation, which in the context of this study is technology using a data fusion process, produces estimates of situational conditions which are, ideally, of reasonable but imperfect quality, i.e., there is some uncertainty in the estimates. However, this study extends the studies of trust to situations where, additionally, the automation system is open to deliberate manipulation by adversaries. Having another person or a group of people at the other end changes the central scheme of the problem. In this environment, called Information Warfare (IW), human operators must deal with faults resulting from both mechanistic automation failure (imperfect data fusion) and premeditated deception or misguidance manipulated by an adversary. Human operators should have the ability to distinguish the faults perpetrated by the foe, to calibrate their trust in decision-making aids and to eventually accomplish their mission efficiently. Without knowing the system's vulnerability from an adversary, human operators may regard faults of corrupted information by an adversary as automation failures in the data fusion process or malfunction of displays. In this paradigm, some characteristics of trust, which were identified in previous studies, such as fiduciary responsibility, may not be applicable because of the hostile environment. Simultaneously, the new paradigm should be able to include the simple relationship between human operators and decision-making aids.
Among the sociological studies defining trust in the interpersonal relationships, Rempel, Holmes, & Zanna ( 1985) characterized trust as a multi-faceted construct having three dimensions. While this represents one classification of trust characteristics, Sheridan ( 1980) suggested a more comprehensive set of seven possible characteristics of human trust in the human-machine systems. As we are dealing in IW with trust of machines, and the data fusion processes, Sheridan's classification is a better starting point. We will consider the applicability of Sheridan's characteristics in turn where the domain is IW.
The first of Sheridan's aspect of trust in automation seems to relate to notions of reliability. This implies a system of reliable, predictable, and consistent functioning. In other words, a sense of reliability in a decision aid (DA) is established when it can be observed to create the same output repeatedly under a particular set of circumstances. There are three fundamental points of vulnerability from the real world situation to the display for operators. These are true input, observable input and displayed output. True input is susceptible to deception and corruption of operations and actions by artificially creating a false circumstance. Observable input is susceptible to parametric or algorithmic corruption. This is creating false output from internal processes in spite of a correct circumstance in the real world. Displayed output is susceptible to algorithmic corruption. This is creating a false display in spite of both (a) an unperturbed real