LLMs as social and cognitive actors
How LLMs represent knowledge and reproduce or amplify human biases, studied with free-association norms and forma mentis networks; how LLM agents agree, polarize or drift in debates, and when they persuade humans.
← Research
Research line
GenAI & Social Twins
LLM cognition and biases, opinion dynamics among LLM agents, and LLM-powered digital twins of social media platforms such as YSocial.
Large Language Models are becoming participants in the social systems we study: they write, recommend, moderate and persuade, and they increasingly interact with each other.
This research line looks at generative AI through the lens of network and computational social science. We study LLMs as social and cognitive actors, and we use them to build social twins, simulated platforms where controlled experiments on online behaviour become possible.
Two complementary directions
Social twins
LLM-powered digital twins of online platforms, where synthetic users with personas, interests and memory post, comment and follow, enabling reproducible and ethically safe experiments on moderation and recommendation.
Our social twin: YSocial
The open-source platform where we put these ideas into practice.
YSocial Social media digital twin YSocial is an open-source framework to design, simulate and analyse social media platforms populated by LLM-driven agents. It lets researchers test recommender systems, content moderation strategies and opinion dynamics in polluted information environments, without touching real users.
- Open source
- LLM-driven agents
- Recommender systems & moderation
- Local or remote LLMs
Simulation running#urbanplanning · illustrative feed
M
Marta @marta_rome · 2m
New bike lanes downtown finally approved. Curious how the shops on the corner will react #urbanplanning
12 replies♥ 487 shares
L
Luca @luca_b · 1m
Less parking means fewer customers. Nobody asked the people who work here.
4 replies♥ 212 shares
S
Sara @sara_k · now
Two good points here. A pilot on one street might settle it #evidence
1 reply♥ 91 share
D
Dario @dario_m · now
Pilot for three months, publish the footfall data, then vote. Deal?
0 replies♥ 60 shares
Objectives
- 1
Characterise the cognitive and social biases of LLMs and compare them with human behaviour.
- 2
Model opinion dynamics in populations of LLM agents and in hybrid human–AI networks.
- 3
Build open, reproducible social twins to evaluate the societal impact of platform design choices.
Related publications
All publications-
PhDGPT: Measuring bias in academic mental health across quantised and uncensored LLMs
Information Processing & Management, 64(2)@article{deduro2027phdgpt, title = {PhDGPT: Measuring bias in academic mental health across quantised and uncensored LLMs}, author = {Duro, Edoardo S. De and Taietta, Enrique and D'Alfonso, S. and Bentenuto, A. and Rossetti, Giulio and Stella, Massimo}, journal = {Information Processing \& Management}, year = {2027}, volume = {64}, number = {2}, pages = {105173}, } -
The role of System 1 and System 2 semantic memory structure in human and LLM biases
arXiv (Cornell University)@article{abramski2026role, title = {The role of System 1 and System 2 semantic memory structure in human and LLM biases}, author = {Abramski, Katherine and Rossetti, Giulio and Stella, Massimo}, journal = {arXiv (Cornell University)}, year = {2026}, doi = {10.48550/arxiv.2604.12816}, url = {https://doi.org/10.48550/arxiv.2604.12816}, publisher = {Cornell University}, } -
Network Effects and Agreement Drift in LLM Debates
arXiv (Cornell University)@article{cau2026network, title = {Network Effects and Agreement Drift in LLM Debates}, author = {Cau, Erica and Failla, Andrea and Rossetti, Giulio}, journal = {arXiv (Cornell University)}, year = {2026}, doi = {10.48550/arxiv.2604.11312}, url = {https://doi.org/10.48550/arxiv.2604.11312}, publisher = {Cornell University}, } -
LLMs can persuade only psychologically susceptible humans on societal issues, via trust in AI and emotional appeals, amid logical fallacies
arXiv (Cornell University)@article{carrillo2026llms, title = {LLMs can persuade only psychologically susceptible humans on societal issues, via trust in AI and emotional appeals, amid logical fallacies}, author = {Carrillo, Alexis and Citraro, Salvatore and Ardebili, Ali Aghazhadeh and Taietta, Enrique and Rossetti, Giulio and Ferrara, Emilio and Veltri, Giuseppe and Stella, Massimo}, journal = {arXiv (Cornell University)}, year = {2026}, doi = {10.48550/arxiv.2604.16935}, url = {https://doi.org/10.48550/arxiv.2604.16935}, publisher = {Cornell University}, } -
LLMs and Cognition
@incollection{rossetti2026llms, title = {LLMs and Cognition}, author = {Rossetti, Giulio and Abramski, Katherine and Cau, Erica and Pansanella, Valentina and Stella, Massimo}, year = {2026}, pages = {1--39}, doi = {10.1007/978-3-031-61050-9_22-1}, url = {https://doi.org/10.1007/978-3-031-61050-9_22-1}, } -
Forma mentis networks predict creativity ratings of short texts via interpretable artificial intelligence in human and AI-simulated raters
Journal of Computational Social Science, 9(1)@article{haim2026forma, title = {Forma mentis networks predict creativity ratings of short texts via interpretable artificial intelligence in human and AI-simulated raters}, author = {Haim, Edith and Fischer, Natalie and Citraro, Salvatore and Rossetti, Giulio and Stella, Massimo}, journal = {Journal of Computational Social Science}, year = {2026}, volume = {9}, number = {1}, doi = {10.1007/s42001-025-00446-z}, url = {https://doi.org/10.1007/s42001-025-00446-z}, publisher = {Springer Nature}, } -
YSocial: An artificial intelligence powered social media virtual twin
Big Data & Society, 13(3)@article{rossetti2024social, title = {YSocial: An artificial intelligence powered social media virtual twin}, author = {Rossetti, Giulio and Stella, Massimo and Cazabet, Rémy and Abramski, Katherine and Cau, Erica and Citraro, Salvatore and Failla, Andrea and Improta, Riccardo and Morini, Virginia and Pansanella, Valentina}, journal = {Big Data \& Society}, year = {2026}, volume = {13}, number = {3}, } -
Social Simulations: from Agent-Based Modeling to Digital Twins
arXiv preprint arXiv:2607.13693@article{cau2026socialsim, title = {Social Simulations: from Agent-Based Modeling to Digital Twins}, author = {Cau, Erica and Failla, Andrea and Pansanella, Valentina and Rossetti, Giulio}, journal = {arXiv preprint arXiv:2607.13693}, year = {2026}, } -
The “LLM World of Words” English free association norms generated by large language models
Scientific Data, 12(1)@article{abramski2025world, title = {The “LLM World of Words” English free association norms generated by large language models}, author = {Abramski, Katherine and Improta, Riccardo and Rossetti, Giulio and Stella, Massimo}, journal = {Scientific Data}, year = {2025}, volume = {12}, number = {1}, pages = {803--803}, doi = {10.1038/s41597-025-05156-9}, url = {https://doi.org/10.1038/s41597-025-05156-9}, publisher = {Nature Portfolio}, } -
Selective agreement, not sycophancy: investigating opinion dynamics in LLM interactions
EPJ Data Science, 14(1)@article{cau2025selective, title = {Selective agreement, not sycophancy: investigating opinion dynamics in LLM interactions}, author = {Cau, Erica and Pansanella, Valentina and Pedreschi, Dino and Rossetti, Giulio}, journal = {EPJ Data Science}, year = {2025}, volume = {14}, number = {1}, doi = {10.1140/epjds/s13688-025-00579-1}, url = {https://doi.org/10.1140/epjds/s13688-025-00579-1}, publisher = {Springer Nature}, } -
Language-Driven Opinion Dynamics in Agent-Based Simulations with LLMs
arXiv (Cornell University)@article{cau2025language, title = {Language-Driven Opinion Dynamics in Agent-Based Simulations with LLMs}, author = {Cau, Erica and Pansanella, Valentina and Pedreschi, Dino and Rossetti, Giulio}, journal = {arXiv (Cornell University)}, year = {2025}, doi = {10.48550/arxiv.2502.19098}, url = {https://doi.org/10.48550/arxiv.2502.19098}, publisher = {Cornell University}, } -
Bots of a Feather: Mixing Biases in LLMs’ Opinion Dynamics
In Studies in computational intelligence@incollection{cau2025bots, title = {Bots of a Feather: Mixing Biases in LLMs’ Opinion Dynamics}, author = {Cau, Erica and Failla, Andrea and Rossetti, Giulio}, booktitle = {Studies in computational intelligence}, year = {2025}, pages = {166--176}, doi = {10.1007/978-3-031-82439-5_14}, url = {https://doi.org/10.1007/978-3-031-82439-5_14}, publisher = {Springer Nature}, } -
A word association network methodology for evaluating implicit biases in LLMs compared to humans
arXiv preprint arXiv:2510.24488@article{abramski2025word, title = {A word association network methodology for evaluating implicit biases in LLMs compared to humans}, author = {Abramski, Katherine and Rossetti, Giulio and Stella, Massimo}, journal = {arXiv preprint arXiv:2510.24488}, year = {2025}, } -
LLM generated word association norms
In Hybrid Human Artificial Intelligence@inproceedings{abramski2024llm, title = {LLM generated word association norms}, booktitle = {Hybrid Human Artificial Intelligence}, author = {Abramski, Katherine and Lavorati, Clara and Rossetti, Giulio and Stella, Massimo}, year = {2024}, } -
From Perils to Possibilities: Understanding how Human (and AI) Biases affect Online Fora
arXiv (Cornell University)@article{morini2024perils, title = {From Perils to Possibilities: Understanding how Human (and AI) Biases affect Online Fora}, author = {Morini, Virginia and Pansanella, Valentina and Abramski, Katherine and Cau, Erica and Failla, Andrea and Citraro, Salvatore and Rossetti, Giulio}, journal = {arXiv (Cornell University)}, year = {2024}, doi = {10.48550/arxiv.2403.14298}, url = {https://doi.org/10.48550/arxiv.2403.14298}, publisher = {Cornell University}, } -
Cognitive Network Science Unveils Affective Bias in GPT Models Mirroring Math Anxiety in High-School Students
In Frontiers in artificial intelligence and applications@incollection{abramski2024cognitive, title = {Cognitive Network Science Unveils Affective Bias in GPT Models Mirroring Math Anxiety in High-School Students}, author = {Abramski, Katherine and Citraro, Salvatore and Lombardi, Luigi and Rossetti, Giulio and Stella, Massimo}, booktitle = {Frontiers in artificial intelligence and applications}, year = {2024}, doi = {10.3233/faia240221}, url = {https://doi.org/10.3233/faia240221}, } -
Cognitive network science reveals bias in GPT-3, ChatGPT, and GPT-4 mirroring math anxiety in high-school students
Big Data and Cognitive Computing@article{abramski2023cognitive, title = {Cognitive network science reveals bias in GPT-3, ChatGPT, and GPT-4 mirroring math anxiety in high-school students}, url = {https://www.mdpi.com/2504-2289/7/3/124}, doi = {10.3390/bdcc7030124}, journal = {Big Data and Cognitive Computing}, author = {Abramski, Katherine and Citraro, Salvatore and Lombardi, Luigi and Rossetti, Giulio and Stella, Massimo}, year = {2023}, }