Adaptive iNterfaces for accessible and Inclusive digiTal services for older Adults – ANITA

Access to digital services is conditioned by interaction with virtual assistants (VA). Few works have been dedicated to the design of VAs that are acceptable and able to adapt to the various interaction styles of older adults (OAs), in particular with cognitive disorders. The aim of ANITA is: a) a better understanding of the needs of IPs, as potential users of digital tools (acceptance, preferences, usefulness, feeling of social inclusion, ethical issues), and b) an objective characterization of OAs interaction styles with VAs through iterative testing, to help the development of more adaptive, acceptable, efficient, and inclusive VA-like interfaces for OAs. ANITA team brings together major academic and industrial players with complementary skills in communication and multilingual information processing (LIG, UMR 5217, GETALP), multimodal interaction and social signal processing (ISIR, UMR7222, PIRoS) , next-generation interface design (PME SPooN.ai), and robotics, assistive technologies and AI applied to geriatrics (Broca Hospital, APHP, *coordinator). In addition to coordination and management (WP1), the work plan will include 3 initial tasks (Broca): (1) a literature review (WP7) will focus on the accessibility needs of OAs regarding digital interfaces and barriers/ facilitating factors encountered when interacting with VAs, (2) OAs needs will be explored through a qualitative study (WP2), (3) a baseline assessment (WP6) will be performed using the current VA of SPooN with 3 test scenarios. The data collected to model user behavior will be the behavioral/interactional strategies of users and their cognitive, affective, sensory and socio-demographic profile. Baseline data collection will allow ISIR (WP3) and LIG (WP4) to identify, respectively, the parameters for defining and measuring user needs during the interaction task (including accessibility and usability) and the type of interface changes to be implemented in future versions of the system. The results of the tasks (WP7, WP2, WP6) will be used to design a version (V1) of the adaptive VA to be tested in wave 1 (WP6), using a Wizard of Oz (WoZ) technique (to avoid the risk of a malfunctioning fully automated system). The WoZ experimenter will see and hear the user behavior, interpret it and select the appropriate responses for the VA (visual and audio adaptations). Insights gained with the development and testing of (V1) will enable us to create an improved and more complex (V2) of the VA system, tested again in Wave 2 (WP6) for effectiveness. Iterative integration of modules for “user behavior” and “VA adaptive interface” (WP5) will allow to have in the last year of the project a fully automatic adaptive VA interface with automatic algorithms (trained on data collected in Waves 1-2) for user behavior analysis and dynamic adaptation of the interface. WP8 will focus on the dissemination and exploitation of results. Throughout the project, given our vulnerable population, we will pay a thorough attention to the analysis of ethical issues (risk of deception, attachment, replacement of human assistance, stigmatization) and legal (consent, confidentiality, security ) related to the use of artificial intelligence, the collection of data and the determination of user profiles for the implementation of VAs. A key aspect of the project will be to develop guidelines to frame the design and use of VAs, to inform users about how these systems work, and ensure they can give informed consent when using them.