Next session
Please join us for the following talk in room G26, 7 George Square. The link to the online Teams meeting will be sent to the mailing list closer to the time for those who cannot join us in Edinburgh.
Speaker
Eugene Philalithis
Time and date
March 10th, 2 PM.
Title and abstract
Risks from alignment in Human-AI dialogue
Chatbots implementing Large Language Models (LLMs), using language as the main medium to interact with human users, are now ubiquitous. In this talk I consider risks to human participants in these interactions via the lens of psycholinguistic theories of dialogue: are interactions with LLMs dialogue? I spotlight an apparent asymmetry in processing commitments between LLMs and human participants from the viewpoint of interactive alignment (Pickering & Garrod 2005) and assess its consequences for both theoretical exploration and safety. Are there processing commitments that only constrain human participants, and, if so, what risks would this imply for LLM usage?
Link to session
The meeting link is distributed on our mailing list. If you’re not subscribed to this list, please email us at ppls.psycholingcoffee@ed.ac.uk
Comments are closed
Comments to this thread have been closed by the post author or by an administrator.