When: Thu, 11 September 2025, 13:00 BST/14:00 CEST
Speaker: Roxana Rădulescu, Assistant Professor, Utrecht University
Title: A Multi-Objective Perspective on Building Human-AI Collectives
Abstract: Most complex problems of social relevance—such as climate change mitigation, taxation policy design, or traffic management—involve multiple stakeholders and conflicting objectives. These problems are multi-agent and multi-objective by nature. This talk sketches a vision for human-AI collectives, where humans and artificial agents cooperate to solve such complex challenges. Reinforcement learning is becoming a pivotal tool in designing solutions for such critical domains. I will highlight why traditional reinforcement learning, which uses a single scalar reward, is insufficient for this vision and discuss how multi-objective reinforcement learning (MORL) offers a more robust and adaptable solution. By using a vector of rewards, MORL can optimize for multiple criteria, such as fairness, diversity, and ethical norms. The resulting sets of behaviors provide flexibility and support key principles like transparency and explainability. I will then explore the exciting challenges and opportunities in multi-objective multi-agent decision-making, as a framework for building human-aligned agents.
Zoom Recording: https://youtu.be/fKcZV5x4L9U
When: Thu, 8 May 2025, 13:00 BST/14:00 CEST
Speaker: Nele Albers, Postdoctoral Researcher, TU Delft
Title: Sustaining Behaviour Change Support with Reinforcement Learning
Abstract: eHealth applications for behavior change have shown promise in helping people change behaviors such as smoking, physical inactivity, or unhealthy eating. However, many people quickly stop using these applications. Personalizing the support these apps provide by accounting for people's current and future states – such as motivation or knowledge – might increase their effectiveness, especially in the long run. How can we create effective models for this? How do we make sure the models’ decisions are ethical? I will talk about our work on using reinforcement learning, informed by psychology, to create long term-effective behavior change support in contexts such as smoking cessation and physical activity coaching.
Zoom Recording: https://youtu.be/ssg9xSfLL1M
When: Thu, 17 April 2025, 13:00 BST/14:00 CEST
Speakers: Social AI Lab, University of Bristol:
Nirav Ajmeri
Jessica Woodgate
Joseph Trevorrow
Yining Yuan
Daniel Collins
Title: Social AI and Multi-Agent Systems at Bristol: Towards Ethical Sociotechnical Systems
Abstract: We are a team of researchers at the University of Bristol advancing foundational AI for the societal good, focusing on the development of ethical multi-agent systems. This talk will showcase our work, which encompasses intelligent agents and normative multi-agent systems, and multi-agent reinforcement learning, all while integrating normative principles to ensure that AI systems align with societal norms and values. Our research seeks to facilitate interactions between computational agents and social entities, such as humans and organizations, with a strong emphasis on promoting prosocial behaviour and ensuring equitable outcomes.
Zoom Recording: https://youtu.be/n-DChjeb39w
When: Thu, 13 March 2025, 13:00 GMT/14:00 CEST
Speaker: Prof. Pinar Yolum Birbil, Utrecht University
Title: Fostering Trust in Hybrid Intelligence Collaborations
Abstract: Hybrid intelligence systems consist of agents and humans, each with potentially different capabilities, working together to accomplish tasks of mutual interest. For these collaborations to be trustworthy, agents need to recognize, take into account, and demonstrate social, developmental, and communication skills---like self-reflection and empathy---that are typically linked to humans. How do we represent and teach these skills to agents? How do we realize agent-human collaborations that benefit from these skills? I will talk about our recent work on agents assisting humans in maintaining privacy, where these skills are essential.
Zoom Recording: https://youtu.be/-djKxqWxUlU
When: Thu, 13 February 2025, 13:00 GMT/14:00 CEST
Speaker: Jayati Deshmukh, Research Fellow, University of Southampton
Title: Two Approaches for Building Responsible Agency
Abstract: AI systems are now integral to our daily lives— like chatbots and large language models which respond to our queries, recommendations like which movies to watch, which food to order to which companies to invest in, when and which energy to use in our home, which route to drive etc. In all these scenarios, the responses of AI systems must be aligned with and uphold the interests and preferences of humans using and being impacted by these decisions.
In this talk, we will look at two approaches of building AI agents which act responsibly. First, we will discuss a top-down approach which ensures that the actions made by an AI system are ethically aligned with the preferences of human users. We will explore broadly, the steps to build such systems and some of the possible challenges which might arise in the process. Second, we will discuss a bottom-up approach called Computational Transcendence which endows autonomous agents with an elastic sense of self which results in emergent responsible behavior by the agents. Finally, we will conclude by exploring some possible applications of such responsible agents.
Zoom Recording: https://youtu.be/Px4870I9Ymw
When: Thu, 12 December 2024, 13:00 GMT/14:00 CEST
Speaker: Eric Dignum, Postdoctoral Researcher, University of Amsterdam
Title: Towards Empirically Calibrated Agent-Based Models of School Segregation
Abstract: Despite decades of research and policies aimed to counteract school segregation, many educational systems still consist of substantial levels of school segregation along various lines (e.g., ethnicity, income). Hence, it remains a persistent societal problem. Existing studies show that factors affecting school segregation and components in the system of school choice interact with each other. These interactions are reasoned to be an important mechanism through which the levels of school segregation emerge on the macro-level, but commonly used qualitative and quantitative analysis methodologies (e.g. discrete choice models, interviews, surveys) often ignore these dependencies between the components or their consequences (e.g. feedback loops, non-linearity). However, tools from complex systems such as agent-based models (ABMs) are increasingly being used to model such features (e.g., interactions) explicitly, but often remain highly stylised and therefore have limited applicability to reality.
In this talk, two empirically calibrated ABMs of school segregation are presented to move towards more data-driven agent-based modelling. The first uses openly available data from the municipality of Amsterdam to approximate household residential locations and (precise) school locations to simulate various household preference structures and potential policy scenarios. The second presents a methodology to calibrate large-scale ABMs on empirical data. We show that it is able to retrieve the true parameter values within reasonable accuracy in our context of primary school choice. While the methodology is able to calibrate any ABM or generative model in theory, we also discuss some (computational) challenges and open questions with respect to the actual calibration with empirical data.
Zoom Recording: https://youtu.be/DVQlCMb60p0
When: Thu, 14 November 2024, 13:00 GMT/14:00 CEST
Speaker: Jennifer Williams, Assistant Professor, University of Southampton
Title: Reponsible Speech Technology
Abstract: As a discipline, speech technology is about to meet at the crossroads. Foundational research is becoming more interdisciplinary. The drivers of innovation are creating a melting pot. AI regulation and AI safety is not only popular, but it is now necessary to consider the impacts. For some speech technology researchers, the culmination of our collective scientific progress may appear to be the most natural progression from a world driven by consumer electronics, the internet, and global connectivity. Yet for others, how we arrived at this point reflects the ebbs and flows of funding body research priorities, shifting experimental paradigms, and trending sociotechnical matters. This talk illuminates how the discipline of speech technology is changing, explores several parallel revolutions happening within the field, and opens a discussion of how recent global attitudes toward AI safety may impact the technical work while also providing new research opportunities.
Zoom Recording: https://youtu.be/PbgQfAvJH4c
When: Thu, 10 October 2024, 13:00 BST/14:00 CEST
Speaker: Yali Du, Associate Professor, King's College London
Title: Towards Cooperative AI Agents
Abstract: From collaborative industrial robots to personal AI assistants, the integration of AI into our daily lives highlights the critical need for effective and reliable coordination among agents, as well as between agents and humans. This challenge centers on creating agents that not only align with user intentions but also possess the flexibility to adapt to evolving circumstances, such as the introduction of novel agents. The pursuit of multi-agent cooperation extends beyond individual interactions to encompass broader societal considerations. In this talk, I will discuss the challenges of multi-agent cooperation, emphasizing our contributions through the use of deep reinforcement learning (RL) and large language models (LLMs). This includes zero-shot human-AI coordination, leveraging LLMs for communication, and incorporating human instructions to ensure safe and cooperative control, with examples including the game of Werewolf, football, and safe robot control.
Zoom Recording: https://youtu.be/lG-Yd6t3g4I
When: Thu, 12 September 2024, 13:00 BST/14:00 CEST
Speaker: Yen-Chia Hsu, Assistant Professor, University of Amsterdam
Title: Empowering Local Communities Using Artificial Intelligence
Abstract: How can scientists co-create AI (Artificial Intelligence) systems with citizens to address environmental and social issues? Recently, it has become an important topic to explore the impact of AI on society. One viable strategy is citizen science, and its previous works have identified expert-based methods of how scientists developed technology for the public to participate in research, such as sustaining participation, verifying data quality, and labeling data. In contrast, there is another community-based perspective that receives significantly less attention: how scientists co-create AI systems with local communities to influence a particular geographical region. This talk will discuss examples and challenges of applying the community-based perspective to create social impact and empower people at a place-based local scale. Three deployed systems focusing on air quality monitoring using different types of data will be presented as examples.
Zoom Recording: https://youtu.be/B5936jUlDbQ
When: Thu, 16 May 2024, 13:00 BST/14:00 CEST
Speaker: Nardine Osman, Tenured Scientist, Artificial Intelligence Research Institute (IIIA) and Spanish National Research Council (CSIC)
Title: Value Aware Multiagent Systems
Abstract: Norms have extensively been used as means of governing multiagent behaviour. But with the rise of hybrid human-AI societies, the need to consider the alignment with human values has gained tremendous traction lately, giving rise to the value-alignment problem. The value-alignment problem is defined as the problem of designing systems that are provably aligned with human values. To achieve this, there is a need to develop software systems that reason about both human values and norms, implement these values through norms, and ensure the alignment of behaviour with those values and norms. The result would be value aware systems that take value-aligned decisions, interpret human and agent behaviour in terms of values, and even enrich human reasoning by enhancing the human’s value-awareness.
Zoom Recording: https://www.youtube.com/watch?v=CxffZ3VEHZ8
When: Thu, 11 Apr 2024, 13:00 BST/14:00 CEST
Speaker: Pradeep Kumar Murukannaiah, Delft University of Technology (TU Delft)
Title: From Deliberations to Decisions via Hybrid Intelligence
Abstract: Making policy decisions involves multiple stakeholders and is a complex process. In this talk, I describe my vision of policy decision making as a hybrid intelligence (HI) process, where decision makers (humans) are supported by artificial intelligence techniques. A key technical component of this vision is multi-objective optimization (MOO). MOO is a well-developed field, but two important problems —pre-optimization and post-optimization —are largely unexplored. First, how do the objectives of MOO come about? I describe why values can be used to formulate objectives and how such values can be extracted via natural language processing (NLP) of policy deliberations. Second, how can a decision maker interpret the output of MOO (typically a large, multi-dimensional, set of solutions, e.g., a Pareto set)? I describe how a MOO solution set can be effectively clustered to reduce the information overload on the decision makers. These components (pre-optimization, optimization, and post-optimization), together, yield an end-to-end HI decision-making process that bridges policy deliberations and policy decisions.
Zoom Recording: https://youtu.be/aqggrVpk2Q4
When: Thu, 14 Mar 2024, 13:00 GMT/14:00 CET
Speaker: Federico Cugurullo, Trinity College Dublin
Title: Artificial Intelligence and the City: Urbanistic Perspectives on AI
Abstract: Innovation in artificial intelligence (AI) is transforming cities in unprecedented ways. In this presentation, we will explore the connections between AI and the urban by focusing on the concept of urban AI and reflecting on its most prominent incarnations: autonomous vehicles, urban robots, city brains, and urban software agents. We will then see how the emergence of urban AI is producing a new urbanism, an AI urbanism, that originates from smart urbanism but also departs from it along three main axes, namely function, presence, and agency. Empirically, we will draw on the findings from several international case studies to examine the repercussions of urban AI and give evidence of how the emergence of AI in cities is reshaping urban society, urban infrastructure, urban governance, urban planning, and urban sustainability. Theoretically, we will discuss the implications of the emergence of urban AI for urban theory and the future of cities. We will conclude the presentation with a warning about the impending risks posed by multiple urban AIs and the obscure black boxes driving their operations, but also with an invitation to politically engage as citizens with increasingly autonomous cities that might forever escape our understanding and thus our control.
When: Thu, 15 Feb 2024, 13:00 GMT/14:00 CET
Speaker: Nirav Ajmeri, University of Bristol
Title: Prosociality and Fairness in Intelligent Agents
Abstract: In virtually every domain (e.g., smart cities, smart multi-modal transportation, policy-making), one user’s actions affect other users. AI systems that support or automate decision-making affect not only their primary users but also others. AI agents of today tend to prioritise the preferences of a single user, typically the individual who initially configured the agent. Even when designed to represent the preferences of multiple stakeholders, these AI agents tend to optimise for the interests of their primary user. This approach carries inherent risks, as it can inadvertently reinforce existing privileges and potentially exacerbate disadvantages faced by vulnerable individuals and marginalised communities. There is a need to consider the broader societal implications of AI decision-making and ensure that AI systems promote fairness and equitable outcomes. In this talk, I will discuss how we could create multi-user-multi-agent systems that respect the values and preferences of all users and promote prosociality and fairness.
Zoom Recording: https://youtu.be/H0DDsYFTPOo
When: Thu, 14 Dec 2023, 13:00 GMT/14:00 CET
Speaker: Wen Gu, Japan Advanced Institute of Science and Technology (JAIST)
Title: Facilitation Support for Online Consensus Decision-making
Abstract: Text based online forum has been considered as an effective approach for holding large-scale online discussions and collecting opinions. It has been utilized in the situations such as academic conferences and citizen co-creation meetings. To promote the development of the discussion process, facilitation becomes critical, and the support of facilitation is essential. In this talk, Wen will give a brief introduction to the research topic of facilitation support in online consensus decision-making. Several examples of existing research and ongoing work will be explained. In addition, he will address the challenges of facilitation support in online forum and talk about the future work in this research area.
When: Thu, 09 Nov 2023, 13:00 BST/14:00 CEST
Speaker: Fernando P. Santos, University of Amsterdam
Title: Multisector games in cities
Abstract: Cities are complex adaptive systems and developing new urban projects hinges on the strategic interaction between multiple sectors (e.g., citizens belonging to different groups, public bodies, and companies). From urban planning to global governance, analysis of these complex interactions requires new mathematical and computational approaches. In this talk I will a present a framework, grounded on evolutionary game theory, to envisage situations in which each of these sectors is confronted with the dilemma of deciding between maintaining a status quo scenario or shifting towards a new paradigm. We’ll consider multisector conflicts regarding environmentally friendly policies as an example of application. I will finish with two domains that we have been exploring at the Civic AI Lab, where strategic interactions between citizens belonging to different groups also plays a fundamental role: designing fair transportation systems and reducing school segregation.
Zoom Recording: https://youtu.be/VUkuPrF3WNs
When: Thu, 12 Oct 2023, 13:00 BST/14:00 CEST
Speaker: Rafik Hadfi, Kyoto University
Title: Conversational Agents for Digital Inclusion
Abstract: The advent of AI fosters the development of innovative methods of communication and collaboration. Integrating AI into ICTs is now ushering in an era of social progress that has the potential to empower marginalized groups. This transformation paves the way for digital inclusion, potentially empowering the online presence of women, minorities, and individuals with disabilities. In this talk, I will introduce various facets of digital literacy, inclusion, equity, and their interconnected roles in fostering self-realization. I will then examine if AI can promote digital inclusion for women by boosting their presence on online platforms. I will share the findings of a study that examines the effect of incorporating conversational agents into gender-mixed online debates. The results show that conversational agents produce quantitative differences in how genders contribute to the debate by raising issues, presenting ideas, and articulating arguments. The results also show increased ideation and reduced inhibition for both genders, particularly females, when interacting exclusively with other females or agents. The enabling character of conversational agents reveals an apparatus that could empower women and increase their agency on online platforms.
Zoom Recording: https://youtu.be/932hWZrHhiM
When: Thu, 14 Sept 2023, 13:00 BST/14:00 CEST
Agenda:
[10 min] - Sebastian Stein, University of Southampton: Citizen-Centric AI Systems
[10 min] - Michiel van der Meer, Leiden University: Hybrid Intelligence for Large-Scale Deliberation
[10 min] - Nirav Ajmeri, University of Bristol: Introduction to Bristol Social AI Lab
[10 min] - Fernando P. Santos, University of Amsterdam: Socially Intelligent Artificial Systems
Zoom Recording: https://youtu.be/N3k-ASSpmr8
When: Thu, 13 July 2023, 13:00 BST/14:00 CEST
Agenda:
[15 min] - Sebastian Stein: Vision of CCAIS and seminar series
[30 min] - Brainstorming on the way forward: giving talks, initiating collaborations, stabilising, expanding
[15 min] - Handover to the Organising Committee (next session: 14 September 2023)