Agenda

Title: “Social and algorithmic choices in the transmission of (mis)information”.

Abstract: The circulation of information requires prior exposure: we cannot disseminate what we do not see. In online platforms, information exposure results from a complex interaction between social and algorithmic forms of curation that shapes what people see and what they engage with. In this talk, I will discuss recent research investigating the social and technological factors that shape what information we encounter online and how it diffuses from user to user. I will also discuss why it is important we adopt a systematic approach to mapping the information environment as a whole – this approach is needed to capture aggregate characteristics that go unnoticed if we only focus on individuals. Networks, I will argue, offer measurement instruments that can help us map that landscape and identify pockets of problematic content as well as the types of audiences more likely to engage with it.

Bio: Sandra González-Bailón is the Carolyn Marvin Professor of Communication at the Annenberg School for Communication, and Director of the Center for Information Networks and Democracy (cind.asc.upenn.edu). She also has a secondary appointment in the Department of Sociology at Penn. Prior to joining Penn, she was a Research Fellow at the Oxford Internet Institute (2008-2013). She completed her doctoral degree in Nuffield College (University of Oxford) and her undergraduate studies at the University of Barcelona. Her research agenda lies at the intersection of computational social science and political communication. Her applied research looks at how online networks shape exposure to information, with implications for how we think about political engagement, mobilization dynamics, information diffusion, and the consumption of news. Her articles have appeared in journals like PNAS, Science, Nature, Political Communication, the Journal of Communication, and Social Networks, among others. She is the author of the book Decoding the Social World (MIT Press, 2017) and co-editor of the Oxford Handbook of Networked Communication (OUP, 2020).

Para entrar en la reunión, pulse el siguiente enlace:
https://oficinavirtual.ugr.es/redes/SOR/SALVEUGR/accesosala.jsp?IDSALA=22997696
- Contraseña de la reunión: 867285

Title: Large Language Models for AI Planning: Efficiency and correctness through code generation

Abstract: Among the most important properties of algorithms investigated in computer science are soundness, completeness, and complexity. These properties, however, are rarely analyzed for the vast collection of recently proposed methods for planning with large language models. In this work, we alleviate this gap. We analyze these properties of using LLMs for planning and highlight that recent trends abandon both soundness and completeness for the sake of inefficiency.

We propose a significantly more efficient approach, Thought of Search (ToS), that can, at the same time, maintain both soundness and completeness. We exemplify on four representative search problems, comparing to the LLM-based solutions from the literature that attempt to solve these problems. We show that by using LLMs to produce the code for the search components we can solve the entire datasets with 100% accuracy with only a few calls to the LLM. The caveat is that ToS requires a human in the loop, collaboratively producing a sound successor function and goal test. To overcome this limitation, in this work, we make a first major step towards automating ToS (AutoToS), completely taking the human out of the loop of interactions with the language model. AutoToS guides the language model step by step towards the generation of sound and complete search components, through feedback from both generic and domain specific unit tests. AutoToS achieves 100% accuracy, with a small number of feedback iterations, using LLMs of various sizes on all evaluated domains.

Bio: Michael Katz is a Principal Research Scientist at IBM T.J. Watson Research Center. His primary research interests are in domain independent planning and in the integration of planning and RL. He is a AAAI Senior Member. He was a program chair of ICAPS 2021 and currently serves as an ICAPS executive council board member and a Competition Liaison (second term). He received numerous awards for his research, including Influential Paper Award in 2023 and Best Dissertation Award in 2011 and for his domain-independent planning solvers. He was a co-organizer of seven editions of the Heuristics and Search for Domain-independent Planning (HSDIP) workshops, as well as of five of the seven editions of the Bridging the Gap Between AI Planning and Reinforcement Learning (PRL) workshop in 2020-2024, which he co-created. He is the creator and a workshop chair of the first edition of the workshop on Planning in the Era of LLMs (LM4Plan@AAAI 2025). He was a panelist at workshops/bridge program at AAAI/ICAPS. He has given tutorials on AI Planning at AAAI/IJCAI/ICAPS. He frequently serves as an AC/SPC for AAAI/IJCAI/ICAPS.

Para entrar en la reunión, pulse el siguiente enlace:
https://oficinavirtual.ugr.es/redes/SOR/SALVEUGR/accesosala.jsp?IDSALA=22997792
- Contraseña de la reunión: 432510