The Social AI Group and the Social AI Centre for Doctoral Training (CDT) are organising the Fourth Workshop on Artificial Social Intelligence on 25 September 2026 in Studio 2 of the Advanced Research Centre, University of Glasgow, UK. 

Social AI involves developing an AI domain aimed at endowing artificial agents with social intelligence, the ability to deal appropriately with users’ attitudes, intentions, feelings, personality and expectations. This full day workshop (9am to 4pm) will host a series of invited talks by renowned experts in Social AI. Our goal is to bring together academic experts, students and industry professionals to encourage dialogs around the progress, challenges and opportunities in Social AI as AI continues to permeate all aspects of our social presence. 

Keynote Speakers include:

Dr Yiming Wang, Fondazione Bruno Kessler

"Embodied perception and reasoning in human-centric environments"

Recent advances in vision-language models and multimodal large language models have transformed embodied AI, enabling agents to move beyond specialist systems toward generalist models that can interpret visual observations and natural-language instructions to generate executable actions. While these models have demonstrated impressive capabilities in controlled laboratory settings, they continue to struggle in the human-centric environments where embodied agents will ultimately need to operate.

Unlike curated benchmarks, human-centric environments are visually diverse, spatially complex, and constantly evolving. Homes, for example, are highly personalized spaces whose objects are not only functional but also reflect individual preferences through subtle visual characteristics. Their layouts reflect personal styles being tidy or messy, and change with everyday activities: a cup may be neatly stored in a cupboard, sitting beside a sofa while in use, or left among dirty dishes in the sink. Human communication is equally variable, with instructions that may be brief, ambiguous, or even erroneous depending on the context and the individual. Addressing this complexity and variability is a key step toward deploying embodied AI in everyday settings.

In this talk, I will present a series of our recent works that enable embodied agents to better understand complex scenes and natural-language instructions, with applications to language-guided navigation and object grasping. Together, these works illustrate how advances in multimodal perception and reasoning can bring embodied agents closer to robust operation in the real world.

Dr Fabio Celli, Maggiolo Group

"From Personality to Societies: the Micro-Macro link"

Since about 20 years, researchers have leveraged the field of personality computing to extract psychological traits from digital footprints. Specifically, algorithms for personality recognition from social media can map individual psychological profiles (a micro-level) at a massive scale. 

At the macro-level, the emerging interdisciplinary field of cliodynamics can model large-scale social dynamics evolving through time. By applying historical phase recognition algorithms to historical databases, we can identify macro-structural transitions derived from collective behaviors—such as periods of political instability, economic stagnation, or state collapse. 

Connecting these two data-driven fields allows scientists to computationally analyze what sociologists call the "Micro-Macro link" between individuals and societies: how millions of micro-level individual expressions aggregate over time to trigger macro-historical phases, and conversely, how shifting societal eras alter cultural norms, shaping the behavior of individuals.

In this talk we will see an overview of personality computing in the last 20 years, the task of Historical Phase Recognition, the data to perform it, and potential impacts of LLM usage on collective memory.

Prof Viviana Patti, University of Turin

"Absit iniuria verbis - Monitoring and countering toxic speech online and beyond: Challenges and promises from a Computational Linguistics perspective"

In recent years, abusive language and in particular the phenomenon of online hatred against marginalized and vulnerable groups are exponentially increasing in social media platforms becoming a relevant social problem that needs to be monitored. Computational linguistics techniques have been applied to monitoring and counteracting toxic speech online, with particular emphasis on the development of linguistic resources and automatic tools for analyzing, detecting, and contrasting various forms of abusive language, ranging from anti-immigrants discourse to sexist and misogynistic behaviors. The research field encompasses several challenges that will be discussed. The targeted nature of online hatred and the multilingual environment pose challenges related to the development of robust approaches for abusive language detection in multidomain and multilingual settings. Pragmatic aspects associated with the use of profanity must also be addressed. Moreover, abusive language is often expressed and to be interpreted in the context of widespread social phenomena of stereotyping and gender/ethnic discrimination, and we are interested in recognizing implicit forms of abusive language. A recent challenge is overcoming the risks of both over-moderation and under-moderation. This has highlighted the need to better understand what is perceived as hateful and what is not by communities target of abusive language.

Additionally, there has been a recent interest towards developing more inclusive approaches that actively involve target groups. This shift includes a growing focus on underrepresented languages and communities, encouraging researchers to more actively consider ethical issues, and to implement community-based approaches for developing and benchmarking NLP models in specific languages which also implies the need to envision and design new evaluation frameworks. Two key lines of reflections will be proposed. The first line addresses the challenges related to the need of applying semantic grids of finer-grained analysis, encompassing the examination of intersectional hatred expressions and the identification of underlying phenomena, such as stereotypes, prejudices, unintended biases, and subtle forms of linguistic behaviour that contribute to perpetrating discrimination and injustice, including sarcasm and irony, key to detecting implicit toxic speech. In this context, the importance of inclusive design in corpora development will be discussed, aiming to incorporate the perceptions and perspectives of marginalized groups, in accordance with new approaches to NLP, such as perspectivism and FATA (First Ask Then Act!). On the other hand, we will delve into the potential of leveraging computational linguistics methods and language technologies to actively facilitate interventions against online hate speech. This involves, on the one hand, the creation of positive counter-narratives, dedicated to raising awareness about toxic discourse online and empowering individuals facing discrimination to express themselves both in virtual and real-world settings, where irony itself can serve as a strategy of counter-speech; on the other hand, the creation of tools for supporting the use of gender-fair language for promoting gender equality by using terms and expressions that include all identities and avoid reinforcing gender stereotypes.