WP6: Integrative-AI for human-computer interaction
Task Lead: UNITN, PI: Giuseppe Riccardi; Co-PI: Massimo Zancanaro
The WP addresses basic research topics in human interaction by adopting a multidisciplinary perspective which includes machine learning, computational linguistics, cognitive & social psychology. The fundamental goal is to identify and design the mental models for AI systems to be accepted and made adaptable to different user groups.
Effort 24p/m + cascading funds.
Description of work
The research in this WP will focus on an interdisciplinary approach to human-computer interaction with AI systems, including intelligent interfaces, affective computing, multimodal & multisensory interaction, by means of multidisciplinary aspects of cognitive & social psychology. Basic research topics in the human-interaction models for AI will be addressed. The fundamental aspects are to understand the impact of different interaction modalities, verbal and non-verbal signals, to identify and design the mental models for AI systems to be accepted and made adaptable to different user groups with demographic, geographic and cultural diversity and in diverse contexts of use.
T2.6.1 Core HCI for reliable and transparent Integrative AI Task leader Massimo Zancanaro
The task will explore core concepts of HCI like mental models, gulf of execution/evaluation, affordances/signifiers in the perspective of interaction with AI-based systems. The wider notions of acceptance, adoption and appropriation will also be tackled by investigating how users in natural settings conceive the interaction with realistic AI tools. The notions of explainability and transparency will be investigated from both a technical and a socio-technical perspective with emphasis on an integrated approach to AI. The main outcome of the task is an extended model of AI-human interaction as an extension of the Human-centered AI framework.
T2.6.2 Personal Conversations and Interactions in AI Task leader Giuseppe Riccardi
The task will investigate one of the most challenging questions in Human-Machine interaction. Are machines able to establish a personal connection with users and engage them in personal conversations and interactions? Most dialog systems are generally designed and trained with deep neural models to carry out general-purpose or simple task-based interactions. Last but not least, the models adapt poorly to new tasks. The core question is how can we create an agent with mixed initiative control that can engage in personal interactions and conversations and benefit users in the short and mid-term . We will investigate which are the requirements for designing such systems and make them sustainable over time. We will investigate how we can learn users’ personal space of events, persons, objects and associate emotions and ultimately benefit them. Our interdisciplinary approach is driven by insights from psychology, computational linguistics and human-interaction research.
T2.6.3 Evaluation of AI systems Task leaders Giuseppe Riccardi and Massimo Zancanaro
In this task we plan to address the fundamental problems of how Integrative AI systems can be evaluated. This is a fundamental problem to be able to make progress in science and engineering. The evaluation aspects are multifaceted. The most relevant aspects are the reliability, safety, trustworthiness and explainability as perceived and assessed by the users and/or third parties. A fundamental element is the balance between the system's autonomy and the users' control. This balance should be realized by diverse means and embodied in several distinct aspects of the interaction. Furthermore, different types of cooperative behavior should be considered and how they may be more or less appropriate to a given context of use. We plan to devise an unified framework by identifying and mapping the different aspects mentioned above to provide a general perspective on how to evaluate Integrative AI systems by considering the complexity of human-computer interaction