IEEE TEMS/LMAG: "LLMs When Left Alone” (DigHum-Series)

#digital #humanism #dighum #regulation #AI #prediction #LLM #ethics #future #EU_AI_Act

Dear colleague,

We are pleased to invite you to this talk in our Lecture Series:

Tuesday, March 10, 2026 at 5:00 p.m. (17:00) Central European Time (UTC+1)

Topic: “LLMs When Left Alone”
(scroll down for abstract and CV)

Speaker: Stefan Szeider (TU Wien, AT)
Moderator: Julia Neidhardt (TU Wien, AT)

To participate in the talks via Zoom go to: https://tuwien.zoom.us/j/96389928143?pwd=UU5YRkNuRmdoWHV4MFBwMWRCcUErdz09
(Password: 0dzqxqiy)

The talk will be live streamed and recorded on our YouTube channel: https://www.youtube.com/digitalhumanism

For further announcements and information about the speakers in the Lecture Series, see https://dighum.org/#latest-news. Please note that you can access the slides and recordings of our past events via that link.

In case you missed the last lecture by Sayeed Choudhury you can watch the recording of “The Imperative of Openness in AI”.


Announcement of our next event:
21.04.26 DigHum Lecture Darja Djordjevic on AI companionship
reminder our new public calendar: https://calendar.google.com/calendar/ical/dighumtuwien%40gmail.com/public/basic.ics

We are looking forward to seeing you!

ABSTRACT “LLMs When Left Alone”:

Today, autonomous AI agents are so widely deployed that millions of people interact with them daily. These include coding assistants, chatbots, and systems that run for hours without human oversight. A natural question to ask is whether something is going on inside these systems beyond sophisticated pattern matching. The question is easy to dismiss but hard to settle. The standard dismissals (anthropomorphism, reductionism, training bias) each carry force, but none is conclusive. In this talk, we follow an empirical approach. Instead of speculating, we placed frontier language models in minimal autonomous loops with no task and no user. We observe that the behavioral patterns that emerged are surprisingly stable and model-specific. The patterns proved stable enough to identify the source model with high accuracy. We further examined whether AI self-reports of internal states can be trusted. We find that a “placebo tool” that does nothing reliably shifts what models say about themselves. These findings raise questions that go beyond computer science and concern society at large. These questions will only grow more pressing as AI agents become more autonomous.

Short Bio of Stefan Szeider:

Stefan Szeider is an Austrian computer scientist and full professor at the Faculty of Informatics at the Vienna University of Technology (TU Wien), where he chairs the Algorithms and Complexity Group. He is also a visiting scientist at the Simons Institute for the Theory of Computing at UC Berkeley. Specializing in algorithms, logic, computational complexity, and theoretical computer science, Szeider designs efficient algorithms for challenging problems in Artificial Intelligence and Logic. With a doctorate from the University of Vienna, he has held positions at universities in the UK and Canada. Szeider co-founded the Vienna Center for Logic and Algorithms, was the first Austrian computer scientist to receive an ERC Starting Grant, and has published over 250 articles in scientific journals and conference proceedings.

Short Bio of Julia Neidhardt:

Julia Neidhardt is an Assistant Professor at the Research Unit Data Science at TU Wien Informatics with a background in mathematics and computer science. Her work focuses on user modeling and recommender systems in e-commerce, tourism and news, online opinion-forming and behavior, and Digital Humanism. She has published in leading venues, including Nature Human Behaviour, and is regularly invited to speak internationally. She is a senior program committee member of ACM RecSys, associate editor of the Journal of Information Technology & Tourism, and a distinguished reviewer for ACM TORS. She leads the Christian Doppler Laboratory for Recommender Systems at TU Wien and, since 2023, serves as UNESCO Co-Chair on Digital Humanism.