Page 73 - Socially Intelligent Agents Creating Relationships with Computers and Robots

56 Socially Intelligent Agents

User State Assessment Module. This module receives a variety of data
about the user and the task context, and from these data identifies the user’s
predominant affective state (e.g., high level of anxiety) and situation-relevant
beliefs (e.g., interpretation of ambiguous radar return as threat), and their po-
tential influence on task performance (e.g., firing a missile). Since no single
reliable method currently exists for affective assessment, the User Assessment
module provides facilities for the flexible combination of multiple methods.
These include: physiological assessment (e.g., heart rate); diagnostic tasks;
self-reports; and use of knowledge-based methods to derive likely affective
state based on factors from current task context (e.g., type, complexity, time
of day, length of task), personality (negative emotionality, aggressiveness, ob-
sessiveness, etc.), and individual history (past failures and successes, affective
state associated with current task, etc.). For the preliminary ABAIS prototype,
we focused on a knowledge-based assessment approach, applied to assessment
of anxiety levels, to demonstrate the feasibility of the overall adaptive method-
ology. The knowledge-based assessment approach assumes the existence of
multiple types of data (e.g., individual history, personality, task context, physi-
ological signals), and from these data derives the likely anxiety level. Anxiety
was selected both because it is the most prevalent affect during crisis situations,
and because its influence on cognition has been extensively studied and empir-
ical data exist to support specific impact prediction and adaptation strategies.

Impact Prediction Module. This module receives as input the identi-
fied affective states and associated task-relevant beliefs, and determines their
most likely influence on task performance. The goal of the impact predic-
tion module is to predict the influence of a particular affective state (e.g., high
anxiety) or belief state (e.g., “aircraft under attack”, “hostile aircraft approach-
ing”, etc.) on task performance. Impact prediction process uses rule-based
reasoning (RBR) and takes place in two stages. First, the generic effects of the
identified affective state are identified, using a knowledge-base that encodes
empirical evidence about the influence of specific affective states on cognition
and performance. Next, these generic effects are instantiated in the context
of the current task to identify task-specific effects, in terms of relevant do-
main entities and procedures (e.g., task prioritization, threat assessment). The
knowledge encoded in these rules is derived from a detailed cognitive affective
personality task analysis (CAPTA), which predicts the effects of different af-
fective states and personality traits on performance in the current task context.
The CAPTA process is described in detail in [6]. The separation of the generic
and specific knowledge enhances modularity and simplifies knowledge-based
adjustments.