Only 35% of seniors (over 65 years of age) in the U.S. use the Internet, reflecting an age-based digital divide when compared with the fact that 90% adults in the ages of 19-29 use the Internet. This project will address this divide through a novel approach that employs conversational agent technology to reduce the critical cognitive and social-psychological barriers hindering older users in web-based consumer environments. The goal of this project is to broaden the engagement of the growing yet underrepresented senior population in Internet technology, a medium that can dramatically improve their quality of life. This goal closely aligns with the human-centered computing objective of transforming the human-computer interaction experience through the development of systems that are aware of the abilities and special needs of people that use them. Research on animated pedagogical agents has revealed promising results on the effectiveness of agents in learning and motivation. However, no previous research program has investigated how agent technology can be designed to promote autonomy and empowerment in the older population, particularly in web-based consumer environments, involving multi-dimensional information processing and complex decision-making. To address this critical gap in the research, this project will systematically investigate, three aspects of agent interactions with older users: 1) locus of control (agent vs. user), 2) interactional style (functional vs. relational), and 3) modality of exchange (unimodal-voice vs. unimodal-text vs. dualmodal-voice + text). This research program will apply a multi-phase, mixed-methods approach involving qualitative studies in the first phase, and a series of controlled experiments with over 400 older users in subsequent phases. The purpose is to examine the effectiveness of the three aspects of agent-user interaction in: 1) reducing cognitive barriers (reducing information load, increasing navigation convenience, enhancing information search and retrieval ease), 2) reducing social-psychological barriers (enhancing control and efficacy, increasing trust, enhancing perception of social support), and 3) increasing Internet technology use intent. The experimental studies will further determine whether users' gender, visual or hearing impairments, and prior Internet competency interact with the three aspects of agent-user interaction to affect the desired outcomes. The findings from this project will generate new knowledge on how multimodal systems employing conversational agents can be designed for the abilities and special needs of older users leading to a potentially transformative and empowering experience for this underrepresented population in information technology.

Seniors are increasingly finding the necessity to engage in web-based consumer environments (e.g., online banking, shopping, trading, travel reservations). While the functionality of agents in these domains merits examination, the broader significance of this project lies in its ability to inform the development of agent-mediated interfaces for other applications such as websites that provide important health and medical information to seniors. Anecdotal evidence gathered by the project team from a prototype system has revealed the transformative potential of this technology for older users. This pilot research funded by the Office of Outreach at Auburn University, has extended the impact of this project to constituents in the state of Alabama. In addition to research, outreach will also serve as an important mission in the dissemination of future findings from this project to stakeholders at local and national levels. With the goal of enhancing diversity in research and education, this project will also involve a greater representation of African-American (AA) study participants from surrounding counties in Alabama, and actively engage AA graduate students, who are already part of the PI's lab in the educational goal of this project. This project will further enhance the infrastructure for research and education across two disciplines through an interdisciplinary seminar course will be offered to graduate students in computer science and consumer affairs to enhance the understanding of the future researchers on how humans perceive and use computing artifacts such as conversational agents.

Project Report

This project was the first to systematically investigate the effectiveness of three aspects of agent-user interaction – locus of control, interactional style, and modality of exchange (agent and user) in reducing cognitive barriers (information load, navigation convenience and information search and retrieval ease), social-psychological barriers (control and efficacy, trust, and perception of social support), and increasing Internet use intent through a series of controlled experiments with over 400 older adults. In study 1, through four focus group interviews, we identified six barriers to the adoption of web-based consumer environments (such as online shopping and banking) among older users (mean age, 73). These included barriers relating to perceived risks, trust, social support, familiarity, experience, and search. In study 2, a laboratory experiment with older users (mean age of 69 years) demonstrated that a conversational agent that serves search and navigational/procedural support functions in an e-commerce environment reduces the social-psychological barriers for senior users through increased social support, trust, and patronage intention for the website in comparison with a control website without an agent. Study 3, a laboratory experiment with older adults (mean age, 71.4), empirically examined the effects of agent modality of exchange and locus of control on seniors’ Internet experience. We found that information overload was perceived to be lower in the text-only (vs. text + voice) agent modality when the agent played a passive helper role (user-control); however, perceived information overload was lower in the text + voice agent modality when the agent played an active doer role (agent-control). Similarly, perceived efficacy was higher in the text-only condition when the user had control over the interaction, whereas efficacy was perceived to be higher in the text + voice condition when the agent had control. Differences were also evident based on the gender, Internet competence, and hearing impairment of the users. Female users perceived lower information load when the agent communicated via both text and voice than via text only, whereas male users perceived lower information load when the agent used only text than both text and voice. Older users with high Internet-competence showed higher perceived usefulness of the text-only agent than of the text + voice agent; whereas older users with low Internet competence perceived higher usefulness of the text + voice agent than the text-only agent. Older users who experienced hearing impairment preferred the agent who played an active doer role (agent-control) to the agent who played a passive helper role (user-control); however, older users who did not experience hearing impairment showed the opposite preferences with respect to perceptions of social support, perceived concentration on the task, and patronage intent. Study 4, a laboratory experiment with older adults (mean age, 71.2) revealed interesting results with regard to effects of agent interaction style and user modality of exchange on senior’s interaction experience. In general, social-oriented agent interaction produced greater social presence than did task-oriented agent interaction; however, differences were evident based on gender, Internet competence, and hearing impairment of the users. Male users benefitted more from social- than task-oriented agent interaction style with respect to navigational convenience, information overload, and perceived efficacy; whereas female users benefitted more from task- than social-oriented interaction style with respect to the above. Social-oriented agent interaction benefitted users with low Internet competence more than users with high Internet competence with respect to perceived efficacy, trust in website integrity, perceived usefulness, and reuse intent for the website; whereas, task-oriented interaction style benefitted users with high Internet competence with respect to the above. Lastly, social-oriented agent interaction benefitted users with (vs. without) hearing impairment with respect to perceived efficacy, information overload, and website ease of use; whereas, task-oriented interaction benefitted users without (vs. with) hearing impairment. Study 4 surprisingly revealed that older users perceived greater social presence when communicating with the agent using text rather than voice. However, differences were also evident based on the gender and hearing impairment of the user. Female users perceived greater trust in website benevolence when communicating using voice rather than text; whereas, male users perceived greater trust when communicating using text rather than voice. Voice recognition and voice user modality benefitted users with (vs. without) hearing impairment with respect to trust in website ability and online flow, whereas, text user modality benefitted users without (vs. with) hearing impairment. This project importantly generates new knowledge that the extent to which a conversational agent can provide social and cognitive assistance to older users depends on the agent design characteristics as well as diverse user characteristics, which indicates that agent design cannot be done with a one-size-fits-all strategy. The broader significance of this project lies in its ability to inform the development of agent-mediated interfaces for other applications such as websites that provide important health and medical information to seniors, which may directly benefit from the specific findings and implications of this project.

Agency
National Science Foundation (NSF)
Institute
Division of Information and Intelligent Systems (IIS)
Type
Standard Grant (Standard)
Application #
0955763
Program Officer
Ephraim P. Glinert
Project Start
Project End
Budget Start
2009-08-07
Budget End
2013-07-31
Support Year
Fiscal Year
2009
Total Cost
$583,814
Indirect Cost
Name
Clemson University
Department
Type
DUNS #
City
Clemson
State
SC
Country
United States
Zip Code
29634