Title: Ethical Issues in affective computing – the nudge of virtual affective agents
Presenter: Prof.Laurence Devillers ( Sorbonne University and CNRS-LIMSI )
Laurence Devillers is a full Professor of Artificial Intelligence at Sorbonne University (PIV) and head the team of research “Affective and social dimensions in Spoken interaction with(ro)bots: ethical issues”(since 2004) at CNRS-LIMSI. Her HDR (habilitation dissertation) in Computer Science in 2006 was about “Emotion in interaction: Perception, detection and generation” at University Paris-Orsay, France. Since 2020, she heads the interdisciplinary Chair on Artificial Intelligence HUMAAINE: HUman-MAchine Affective INteraction & Ethics (2020-24) at CNRS (http://humaaine-chaireia.fr).
Scientific results in affective computing and the first products such as speech analytics (emotion detection) in call centers, sentiment analysis in website, or affective robot for elderly people inspire questions around the ethics, the goals and the deployment of innovative products that can change our lives and consequently, the society. The emergence of such systems that keep us more and more connected to machine will modify the way we socialize, our reasoning capabilities and our behavior. This technology promises new forms of affective relations and interactions, as well as new market opportunities. Affective computing systems have a large field of applications: conversational agent, robot, e-bot, etc. Such systems are envisioned to interact with humans (with children, adults and frail people such as for example very young children, autistic or elderly) in a seamless non verbal and verbal dialogue in a variety of real-life contexts such as at home, at the hospital, on your phone, in your car, at the classroom, in public transports with different roles such as assistive, companion or still seller systems. Such systems are also envisioned to survey humans for safety reasons (for example in car). The deployment of such affective technology will lead to profound modifications of the way people interact with systems. Achieving seamless multimodal interaction with multiple people and planning for executing system’ speech, movement, expressions or still gestures in response to observed and interpreted user behavior requires an inherently multidisciplinary approach. We need a new interdisciplinary mix of computer science, social/psychological sciences and engineering to understand such affective Interaction, and the substantial impact of affective computing systems will have in terms of new applications. As researchers in the affective computing community, it is important for us to have these discussions and to enlarge our community.
The tutorial will be divided in three parts:
1. Main ethical issues in affective computing
2. Affective Machines and nudge: the HUMAAINE chair (L. Devillers)
3. Questionnaires and ethical committees
Title: Research methods and basic statistical analysis for affective computing
Presenter: Prof. Gale M. Lucas (University of Southern California)
Gale Lucas is a research assistant professor at the University of Southern California in the Viterbi School of Engineering and works at the USC Institute for Creative Technologies (ICT). She obtained her BA from Willamette University in 2005 and her PhD from Northwestern University in 2010. After teaching for a couple of years at small liberal arts colleges, she went back for a post-doc. She completed her post-doc with Dr. Jon Gratch at ICT, and then stayed on at ICT as a senior research associate. She works in the areas of human-computer interaction, affective computing, and trust-in-automation. Her research focuses on rapport, disclosure, trust, persuasion, and negotiation with virtual agents and social robots.
The proposed tutorial will cover a number of topics relevant to research methods and basic statistical analysis for running user studies in affective computing. Outlined below are key topics that will be covered during the tutorial:
Formulating a research question
Operationalization of variables
Independent vs. dependent variables
Measurement of different variables
Experiments vs. correlational research (vs. mixed)
Between-subjects vs. within-subjects designs (vs. mixed)
Null hypothesis statistical testing
Choosing the appropriate statistical test
Moderation vs. mediation
There will be subtopics under several of these broad topics, as well. To teach these topics, participants will be engaged using lecture, call-and-answer, voting, as well as participation in a “toy” research study (for use as a hands-on example for design and analysis).
Title: Sensing Affective States through Wearable Devices
Presenter: Prof. Vivian Motti (George Mason University) and Prof. Vânia Neris (Federal University of Sao Carlos)
Dr. Motti is an Assistant Professor on Human Computer Interaction in the Department of Information Sciences and Technology at George Mason University where she leads the Human Centric Design Lab. Her research interests lie on wearable computing, assistive technologies, and emotion regulation.
Dr. Neris is an Associate Professor on Human Computer Interaction in the Department of Computing at Federal University of Sao Carlos, in Brazil, where she leads the Flexible and Sustainable Interaction Lab. Her research interests lie on computer support for mental health and well-being, and adaptive and adaptable user interfaces.
The recognition of affective states using wearable sensors has become increasingly important to address mental health crises. The development and deployment of affective computing applications aim to address a growing need for interventions that facilitate the delivery of mental health services, reducing their costs, increasing access, and reducing stigma as well. Mobile sensing through wearables, smartphones, and pervasive computing enables applications to sense, monitor, and recognize users’ affective states, besides also delivering personalized interventions on demand. In this tutorial, we present attendees an overview of wearable sensing. First, we will contextualize the research and practice on wearable sensing for affect recognition. Then, we will describe data sources, signals and sensors used in affect recognition. We will explain the machine learning, models, tools and frameworks employed to classify data from users and their environments aiming to recognize affective states. Lastly, we will conclude the tutorial with a discussion about ethics, fairness, bias as well as the trade-offs and evaluation criteria involved in each approach. Knowing that open questions remain in the domain, for instance to ensure that recognition models are accurate in the classification process, and that potential risks for users’ privacy and safety are reduced, we will give attendees the opportunity to discuss and reflect about current applications, incentivizing a critical debate over the implementation and deployment of services that are optimized to collect essential data to preserve users’ privacy, to provide accurate classifications, to minimize risks, and to personalize interventions to meet individual’s needs.