Professor, University of Texas Dallas, US

Talk title: Toward Social AI Systems that Recognize and Synthesize Human Behaviors

Talk slides

Social AI systems should replicate human capabilities of recognizing social signals from users and responding with appropriate human-like behaviors, mirroring the complex orchestration involved in human-human communications. A message conveys more than words, including emotions, cognitive state, intentions, and desires. Building systems that recognize these social signals has many benefits, including clarifying ambiguities in the message and inferring a better understanding of the user’s state. This seminar will discuss our effort to recognize social signals, focusing on emotions. If we intend to build truly interactive AI systems, recognizing social signals is just the beginning. These systems must also be able to communicate back to the user, conveying human-like behaviors. Nonverbal behaviors externalized through head, face, and body movements for socially interactive agents (SIAs) play an important role in human-computer interaction (HCI). Believable movements for SIAs have to be meaningful and natural. The advances in this area will lead to SIAs that can express meaningful human-like gestures that are timely synchronized with speech, enabling novel venues for artificial agents in human-machine interaction. This seminar will discuss our effort to address these challenges, presenting models that capture principled temporal relationships and dependencies between speech and gestures that are carefully considered.

Associate Professor, University of Colorado Boulder, US

Talk title: Toward Trustworthy AI for Mental Healthcare: Exploring Socio-Demographic Bias, Privacy Risks, and Collaborative Decision-Making

Talk slides (coming soon)

The rise in accessibility of smartphones and wearable devices has revolutionized the monitoring of human states beyond laboratory settings. This advancement has resulted in the collection of real-life, multimodal, temporal data, forming a valuable basis for developing machine learning (ML) algorithms that track an individual's internal and contextual states, holding great promise for enhancing mental healthcare. Simultaneously, the relationship between humans and artificial intelligence (AI) has evolved into a collaborative dynamic, where humans and AI systems work together towards common objectives. However, several challenges, both technical and societal, hinder the widespread adoption of such technologies. This talk will discuss the concept of trust in human-centered AI for mental healthcare. Specifically, it will explore concerns related to unintentional disclosure of personal information when using multi-modal bio-behavioral signals in ML systems, and investigate potential socio-demographic biases that arise from these systems. Additionally, the talk will present findings from decision-making tasks that involve human clinicians collaborating with explainable AI systems to estimate human states and mental health outcomes.

Assistant Professor, Politecnico di Milano, Italy

Talk title: When Social Artificial Intelligence Meets Human-Robot Interaction

Talk slides (coming soon)

This talk explores the intersection of human-robot interaction (HRI) and advancements in social artificial intelligence (AI), highlighting how these technologies can enhance the quality and effectiveness of interactions between humans and robots. The talk will start with an overview of recent HRI studies focused on mental well-being and empathy elicitation. I will then focus on  the capabilities of social AI algorithms, particularly in recognizing empathy, assessing mental well-being, and detecting irony. I will conclude with a comprehensive case study demonstrating the successful integration of social AI models to improve human-robot interaction.

Assistant Professor, University of Kansas, US

Talk title: Contextualizing Social Artificial Intelligence for Gender, Culture, and More

Talk slides

This talk explores the importance of enabling social artificial intelligence (SAI) systems to consider the demographic context of the behaviors it observes and tries to “understand.” Motivating examples will be drawn from large-scale observational studies conducted by the speaker to argue that same observed behavior (e.g., a certain frequency/intensity of smiling) may be entirely typical (e.g., around the 50th percentile) in one population (e.g., American women) but highly atypical (e.g., around the 90th percentile) in another population (e.g., Japanese men). Adopting a one-size-fits-all approach to interpreting such behaviors is very likely to mislead the SAI system. Methods for collecting normative data for relevant behaviors and building this contextual knowledge into SAI systems will be discussed. Finally, challenges and opportunities will be discussed related to large-scale observational studies, contextualization of SAI systems, and moving beyond demographics.