The "New Work on ToM 2: interacting with large language models" student conference is organized in partnership between the Faculty of Philosophy at the University of Bucharest (through its Department for Theoretical Philosophy) and the Faculty of Psychology and Education Science, (through its students from the Department for Cognitive Science). The conference aims to bring together students and early researchers in order advance our understanding of how we come to interact with Large Language Models (how do LLMs interpret our communicative behaviour and how do we interpret LLMs communicative behaviour), with a focus on (but not limited to) the role of speech acts in social interactions. We aim to explore how this unfolds in our interactions with large language models.

Generally, it is assumed that human communication is special in the sense that it allows for grasping a speaker’s intention beyond of what is simply verbally said. Recent researchers have claimed that LLMs may exhibit behaviour similar enough to ours, when it comes to theory of mind tasks, such as the False Belief Task (Strachan, 2024; Kosinski 2024). Kosinski argues that ToM- like abilities in LLMs may have emerged as a by-product of language ability (2024:2). The question that arises is if LLMs ability to perform well on some linguistic tasks actually entails real capacities in the social-cognition domain. Can LLMs infer or grasp what could have been said? What about what it’s left unsaid? Can they grasp the speaker’s intentions beyond what is written in the chat, or verbally said, and can they grasp deception? Are LLMs capable of assertion, and if so, does it entail they are capable of expressing beliefs? Are they capable of mindreading?

The conference will take place 20-21 March, between 9- 5 PM, local time for Bucharest, Romania. Regular presentations will be 30 minutes long, followed by 10 minutes long Q&A.

It will have a mixed format, in that speakers may choose whether they present online only or face to face at the event's location (if so, their session will enjoy a live audience, but it will also be streamed to remote participants).

Thursday, March 20th 2025

Chair: Sebastian Andrei Ioniţă

*The time slots correspond to Bucharest time zone, CET +1

11:00-11 40 Merel Semeijn (University of Groningen). Chatbots as fictional bullshitters.

11:40-12 20 Yan Xu (Durham University). Phenomenological analysis of speech acts: Eliciting the listener's responses’.

12:20 -13:00 Marco Fenici (Università Telematica Pegaso). Do LLMs Exhibit Theory of Mind? Insights from Conversational Approaches to False-Belief Task Success.

13:00 -14:00 Lunch Break

14:00 Keynote address: Alexandru Nicolae (University of Bucharest, Faculty of Letters and Institute of Linguistics „Iorgu Iordan - Al. Rosetti”). Some notes on the competence of LLMs.

15:00-15:40 Luca di Vincenzo (Sapienza University of Rome). Can Large Language Models Read Minds? A Multimodal Constraint Hypothesis?

15:40 – 16:00 Break

16:00-16:40 Bianca Savu (University of Bucharest, Faculty of Philosophy). Compliments and AI.

16:40-17:40 Keynote address: Gheorghe Ştefanov(University of Bucharest, Faculty of Philosophy).Propositional attitudes and speech acts.

Cross-referenced with the “How to Safely Interact with LLMs” peer learning workshop at UB

18:00 Keynote address: Iwan Williams (University of Copenhagen). Chatting with bots (Co-authored with Tim Bayne).

19:00 Keynote address: Neri Marsili (Universidad Nacional de Educación a Distancia). Commitment, Deniability, and Indirectness?

Friday, March 21st 2025

Chair: Selahattin-Can Petre-Akbulut

11:00- 11:40 Shaoyu Han (Durham University) Can LLMs Understand? AI Hate Speech, Intention, and the Extended Mind

11:40-12:20 Rareş Mihnea Iosifescu (University of Bucharest, Faculty of Psychology and Educational Sciences). Talking Your Way into Theory of Mind. Can Mental-State Talk Pave the Road for Theory of Mind Development?

12:20: -14:00 Lunch Break

14:00-14:40 Yikunoamlak Mesfin (University of Szeged) The Origins of Ethics: Insights from Evolutionary Ethics and Pragmatism for Addressing Digital Age Challenges.

14:40-15:20 Jeremie Garceau (University of Quebec in Trois-Rivières). Defining artificial agents’ ability to understand.

15:20-16:00 Corina Stavila (University of Bucharest, Faculty of Philosophy). If a Chatbot Can Lie, Can It Also Assert Truthfully?