Personal Assistants for Human Organizations

Personal Assistants for Human Organizations

Steven Okamoto, Katia Sycara, Paul Scerri
DOI: 10.4018/978-1-60566-256-5.ch021
(Individual Chapters)
No Current Special Offers


Intelligent software personal assistants are an active research area with the potential to revolutionize the way that human organizations operate, but there has been little research quantifying how they will impact organizational performance or how organizations will or should adapt in response. In this chapter we develop a computational model of the organization to evaluate the impact different proposed assistant abilities have on the behavior and performance of the organization. By varying the organizational structures under consideration, we can identify which abilities are most beneficial, as well as explore how organizations may adapt to best leverage the new technology. The results indicate that the most beneficial abilities for hierarchical organizations are those that improve load balancing through task allocation and failure recovery, while for horizontal organizations the most beneficial abilities are those that improve communication. The results also suggest that software personal assistant technology will facilitate more horizontal organizations.
Chapter Preview


Intelligent software personal assistants (SPAs) are one of the most exciting applications for organizational multi-agent systems. A software personal assistant (SPA) is an agent that acts to support a user in a human organization by automating individual tasks and facilitating coordination with other members of the organization. Recent and current research has looked at developing SPAs for a diverse range of domains, including emergency response and military teams, office environments, factory floors, and even outer space. The envisioned SPAs possess a wide range of abilities, such as scheduling joint activities (Dent et al, 1992; Garrido & Sycara, 1996; Modi et al, 2004), sharing key information (Wagner et al, 2004), monitoring and reminding individuals of key timepoints (Chalupsky et al, 2001), filtering incoming communication (Maes, 1994), assisting in negotiation decision support (Li et al., 2006), and even ordering lunch (Chalupsky et al, 2001).

Software personal assistants stand to benefit from multi-agent organization research in two ways. First, as large-scale, complex multi-agent systems, SPA deployments are natural candidates for organization-centric engineering approaches to manage control and coordination complexity. As SPA-enabled organizations become more commonplace, the need for an organization-centric approach will only become more apparent, because SPA interactions between different organizations will require that these systems operate flexibly, robustly, and securely as open systems. Secondly, no matter what engineering approach is chosen for an SPA system, the SPAs will be situated in human organizations with specific organizational constraints, and thus the SPAs must be able to represent and reason about those constraints in order to operate successfully and transparently. The representation and reasoning of organizational structures and norms is currently one of the hottest areas of agent organization research (Grossi et al., 2007; Vasconcelos et al., 2007).

While significant technical challenges remain in developing SPAs, many of these issues are the subject of recent and current research and will not be discussed extensively in this chapter. Instead, we will focus on the crucial issue of how human organizations will be affected by the use of this technology, which has gone largely unexamined. This represents a significant gap in the current research, as it is likely that SPAs will have a revolutionary effect on the way human organizations operate, just as previous information technology innovations such as personal computers, corporate databases, and e-mail revolutionized the way organizations operate. In addition, the issue of which of the many envisioned SPA abilities are most useful for improving the efficiency and effectiveness of an organization is only poorly understood. This is a possibly costly oversight, as history is replete with examples of technological innovations, including early SPA systems, that have unexpected and even undesirable impacts when coupled with existing organizational practices and behaviors.

Our goals for this chapter are three-fold: first, to develop a conceptual framework that can be used to quantifiably evaluate proposed SPA technologies; second, to quantify the impacts proposed SPA abilities will have on existing organizations, in order to provide input to SPA designers on which abilities are most promising to pursue; and third, to explore how organizations may best be redesigned to leverage the SPA technology, in order to provide input to SPA adopters on how to best apply the technology. The approach taken in this chapter lies at the crossroads of agent organization modeling and computational organization theory. We develop a computational model of the organization. This methodology of computational modeling is similar to that used for quantifying performance in organizational theory (Carley 1994; Prietula & Carley, 1998). We evaluate the impact different proposed SPA abilities have on the behavior and performance of the organization. Because SPAs frequently affect detailed work activities, and their broader influence on the organization is unclear, we must model the organization at a fine-grained level of detail, capturing, for example, communication paths, decision making, etc., in order to see their effects. To that end, we have created an abstract simulation environment that takes proposed organizational models and tasks to be performed by the organization and computes key properties of task execution, including how well and how quickly the organization performed the task and how robustly it handled individual failures. The simulation captures important aspects of the operation of the organization, such as non-determinism and cognitive limits of individual members of the organization, but abstracts away domain level details, making it feasible to evaluate many instances of organizations.

One of the major difficulties in evaluating SPA abilities is that they are still an area of ongoing research. Because one of our goals is to provide input to SPA designers before the SPA capabilities have been fully developed, it is not possible to directly model the specific mechanisms by which SPAs will operate. Instead, we abstract away the details of the SPA mechanisms and instead use the effects of those mechanisms on the behavior of individual humans to determine the impact on the organization. For example, instead of directly modeling the specific ways in which SPAs will increase the rate at which humans can make effective decisions, we instead model the SPAs’ impact as increasing the decision making rate within the organization, with the understanding that any SPA mechanism that has the same effect will have similar effects on organizational performance. This allows us to evaluate the many proposed SPA abilities without having to solve the hard and open problems required to implement those abilities for different organizational contexts. By referencing ongoing projects and previously published literature, we have identified a set of key abilities that are being developed, including information sharing, task allocation, automated monitoring and supervision, communication management, joint activity coordination, decision support, and recovery from unexpected failures. Using our computational model of the organization, we evaluate the impacts of different SPA abilities individually and in combination.

When embracing a new technology, organizations will first deploy it in lieu of existing technologies or practices, then gradually adapt to better utilize the new technology. This can have a transformative effect on the way organizations are structured and operate. For this reason, it is not sufficient to merely study the impact SPAs will have on existing organizations, but also to see how organizations might change and adapt in the presence of SPAs to best leverage the technology. In order to investigate this, we compare the designs of organizations before and after SPA deployment to see which structures and SPA abilities are most beneficial. Our results indicate that the most beneficial SPA abilities for horizontal, decentralized organizations are those that facilitate improved communication, while for hierarchical, centralized organizations, the most beneficial SPA abilities are those that improve load balancing through task allocation and failure recovery.

Key Terms in this Chapter

Organizational Structure: The union of the roles and relationships that exist within an organization. The organizational structure is used to constrain and guide the activities of agents in the organization.

Global Task Structure: A TÆMS task structure representing the complete problem being solved by the multiagent system. The global task is equivalent to a workflow in grid/cloud computing terminology.

Agent Spawning: The creation of a new agent to handle part of the workload of the spawning agent.

TÆMS: Acronym for Task Analysis, Environment Modeling and Simulation. TÆMS is a computational framework for representing and reasoning about complex task environments in which problems are represented using extended hierarchical task network structures.

Task Rewriting: The process of modifying a task structure. Task rewriting is achieved using a set of operators that act on one or more local-task structures to generate new local-task structures. Task rewriting is usually the first step in changing the organizational structure of an agent.

Local Task Structure: A TÆMS task structure that is used to represents the local task view and organizational knowledge of an agent. The local-task structure is used to represent roles and relationships within an agent.

Agent Composition: The combining or merging of two agents together to form a single agent responsible for all the activities of the two combined agents.

Relationships: The coordination relationships that exist between subparts of a problem.

Roles: The parts played by the agents enacting the roles in the solution to the problem. The roles reflect the long-term commitments made by the agents in question to a certain course of action (that includes task responsibility, authority, and mechanisms for coordination).

Organizational Self-Design: A method of designing organizations at run-time in which the agents are responsible for generating their own organizational structures.

Complete Chapter List

Search this Book: