ÌìÃÀ´«Ã½

2. Governing Human-AI Teams in the Workplace

2026 IDeaS Challenge Problem

AIOverview-Examples

Monday October 12, 2026

Tuesday October 13, 2026

Wednesday October 14, 2026

2026 Conference Papers:

Lynnette Ng, Kathleen M. Carley, “Large-Language Agent-based Network Dynamic (LAND) model for Social Simulation”

Abstract: Currently, there are three main paradigms that are used for modeling societies: agent-based models, large-language models and network dynamic models. This paper argues that these three traditions should converge rather than compete. We describe the Large-language Agent-based Network Dynamic (LAND) model, a framework in which the network-dynamic modeling supplies the interaction topology, agent-based modeling supplies the behavioral dynamics and large-language models supply the language and content generation. We also describe how the LAND model can be used as a pre-deployment policy testbed, and some open challenges and research agendas.

Samantha C. Phillips, Kathleen M. Carley, “Context collapse and identity complexity in online self-presentation: a cross-platform comparison”

Abstract: Social media users must curate their self-presentation in the face of varying degrees of context collapse and audience invisibility across platforms. These conditions may constrain identity complexity in self-presentations, defined here as the number of identities expressed and the conceptual and cultural diversity among them. To assess this possibility, we compare profiles from X and Bluesky users who posted about Eurovision during the same period in May 2026. X has been established for far longer, has a much larger user base, and defaults to an algorithmically curated feed that exposes users to content beyond those they follow, creating greater uncertainty about who may encounter a profile and the norms they bring than on Bluesky. We find that across all three dimensions, Bluesky users expressed greater identity complexity than X users. Specifically, they presented more identities, more conceptually diverse identities, and more culturally atypical combinations of identities, although differences were small overall. These findings are consistent with the possibility that broader and less predictable audiences constrain complex self-presentation and suggest that platform context can shape the forms of identity expression that users make visible online.

Luke Osteritter and Kathleen M. Carley,” Exploring the use of local Large Language Models to Generate Guidance and Stylized Facts for Agent-Based Models”

Abstract: Where can modelers find value in using large language models (LLMs) in the creation of agent-based models (ABMs)? We seek to begin answering this ques-tion by evaluating the output of several small open-weight LLMs on a set of four elicitation prompts regarding the creation of a model of insider threat. These prompts have two purposes: guidance for the creation of ABMs, and generation of “stylized facts” – empirical regularities that are generally regarded as true even if they don’t hold for every specific case – for the purpose of model validation. For each case we employed two prompt approaches: zero-shot, wherein a di-rective and schema for results are provided and nothing else and few-shot, wherein examples from other domains are provided. The models are capable of adhering to structure, with 93% returning the requested JSON schema. The few-shot approaches tended to worsen, not improve, the output as it led the models to return unrelated information. Finally, we find that asking these models to provide empirical sources results in fictitious citations 62% of the time, and only 4% of the sources proved to be both accurate and appropriate. We conclude that using these relatively small local models can be useful for structured ideation but are less useful for grounding work in empirical evidence.

Logistics Information

Hotel Accommodations

A courtesy room block has been reserved at the Hilton Garden Inn Pittsburgh University Place for in-person attendees. To receive the group rate, please book your room using the link below. If you experience any issues, contact Courtney.Burns@hilton.com for assistance.

If you prefer to stay elsewhere, ÌìÃÀ´«Ã½ is conveniently located near a variety of hotels offering amenities such as complimentary Wi-Fi and shuttle service to and from campus. Visitors can make a reservation with a local hotel for their visit.  Many area hotels also offer discounted rates for ÌìÃÀ´«Ã½ Mellon visitors, so be sure to ask about any ÌìÃÀ´«Ã½ special rates when making your reservation.

Campus & Parking

For details on visiting ÌìÃÀ´«Ã½, including parking options and campus maps, please refer to the ÌìÃÀ´«Ã½ Visitor Information page:
/visit/visitor-information

 Food and Refreshments

Lunch will be provided each day of the conference, along with coffee, water, and soft drinks.

Following Monday's sessions, we will host a welcome reception featuring a poster session presented by graduate students from the CASOS and IDeaS Centers.

Please be sure to indicate any food allergies or dietary restrictions during registration so we can accommodate your needs.

If you have any questions or need additional information, please contact us at centerforideas@andrew.cmu.edu.

_________________________________________________________________________________________

Media Partners