Poznań University of Economics and Business

Humans & Artificial Intelligence Laboratory

Behavioral experiments at the intersection of economics, finance, psychology, and computer science.

Themes investigated at HAI Lab
How to reduce
algorithm aversion
when interacting
with LLMs?Do we show antisocial
behavior towards users
of LLMs?
Are we more likely to
behave unethically
when working via an LLM?Affective leverage
used by LLMs
and social acceptability
of doing so

Founded2021

Our officesCollegium Altum · Poznań

Flagship projectBotonomics Funded by National Science Centre, Poland

Research outputs

Selected papers

Peer-reviewed articles and working papers, with study notes and links to the complete scholarly record.

  1. PublishedPeer reviewed

    Do people apply different norms to humans and large language models acting on their behalf? Evidence from norm elicitations in two canonical economic games

    Paweł Niszczota · Elia Antoniou

    Journal of Experimental Social Psychology · Volume 127 · Article 104991

    doi.org/10.1016/j.jesp.2026.104991 ↗
    Study note

    Two preregistered norm-elicitation studies examine whether identical choices are evaluated differently when made by people or by large language models acting on their behalf.

    Design
    2 preregistered studies · N = 2,658 · representative UK and US samples
    • Social norms
    • LLM agency
    • Economic games
    01Read article ↗
  2. PreprintNot peer reviewed

    Antisocial behavior towards large language model users: Experimental evidence

    Paweł Niszczota · Cassandra Grützner

    arXiv · cs.AI · v1 · arXiv:2601.09772

    arxiv.org/abs/2601.09772 ↗
    Study note

    A two-phase online experiment tests whether disapproval of AI assistance becomes costly action, finding that participants punished peers more as their actual reliance on an LLM increased.

    Design
    Two-phase online experiment · N = 491 in Phase II · real-effort task
    • LLM use
    • Social sanctions
    • AI disclosure
    02View preprint ↗PDF ↓
  3. PreprintNot peer reviewed

    People Are Highly Cooperative with Large Language Models, Especially When Communication Is Possible or Following Human Interaction

    Paweł Niszczota · Tomasz Grzegorczyk · Alexander Pastukhov

    arXiv · cs.HC · v1 · arXiv:2507.18639

    arxiv.org/abs/2507.18639 ↗
    Study note

    Across repeated and one-shot Prisoner’s Dilemma experiments, cooperation with LLMs remained high; communication increased cooperation with both human and LLM partners, and prior human interaction produced a spillover effect.

    Design
    2 experiments · N = 292 · repeated and one-shot Prisoner’s Dilemma
    • Cooperation
    • Communication
    • Human–LLM interaction
    03View preprint ↗PDF ↓
  4. Journal articlePeer reviewed

    Large language models can replicate cross-cultural differences in personality

    Paweł Niszczota · Mateusz Janczak · Michał Misiak

    Journal of Research in Personality · Volume 115 · Article 104584

    doi.org/10.1016/j.jrp.2025.104584 ↗
    Study note

    A large-scale experiment tests whether GPT-4 can reproduce US–South Korean differences in Big Five personality ratings, finding the expected cross-cultural pattern alongside upward-biased means, reduced variation, and weaker structural validity.

    Design
    Large-scale in silico experiment · N = 8,000 · GPT-4 and GPT-3.5 · US and South Korean targets
    • Cross-cultural research
    • Personality
    • Large language models
    04Read article ↗
  5. Journal articlePeer reviewed

    GPT has become financially literate: Insights from financial literacy tests of GPT and a preliminary test of how people use it as a source of advice

    Paweł Niszczota · Sami Abbas

    Finance Research Letters · Volume 58 · Part A · Article 104333

    doi.org/10.1016/j.frl.2023.104333 ↗
    Study note

    The study assesses GPT-3.5 and GPT-4 with 21 financial-literacy items, then examines how 184 people evaluate and use the systems as a source of financial advice.

    Design
    21 financial-literacy items · N = 184 advice study · GPT-3.5 and GPT-4
    • Financial literacy
    • Robo-advice
    • Advice taking
    05Read article ↗
  6. Journal articlePeer reviewed · Open access

    The credibility of dietary advice formulated by ChatGPT: Robo-diets for people with food allergies

    Paweł Niszczota · Iga Rybicka

    Nutrition · Volume 112 · Article 112076

    doi.org/10.1016/j.nut.2023.112076 ↗
    Study note

    The paper evaluates the safety, accuracy, and attractiveness of AI-generated elimination diets, documenting generally useful advice alongside potentially harmful errors in portions and calorie estimates.

    Design
    56 AI-generated diets · 14 common food allergens · safety and accuracy review
    • Dietary advice
    • Food allergy
    • ChatGPT
    06Read article ↗
  7. Journal articlePeer reviewed · Open access

    Judgements of research co-created by generative AI: experimental evidence

    Paweł Niszczota · Paul Conway

    Economics and Business Review · Volume 9 · Issue 2 · Pages 101–114

    doi.org/10.18559/ebr.2023.2.744 ↗
    Study note

    A preregistered vignette experiment finds that delegating research stages to an LLM rather than a PhD student reduces perceived moral acceptability, trust in the researcher, and expected output accuracy.

    Design
    Randomized preregistered mixed-design experiment · N = 402 US participants · 5 research stages
    • Generative AI
    • Trust in science
    • Metascience
    07Read article ↗PDF ↓

History

History of HAI Lab

  1. Botonomics

    The influence of intelligent machines on economic behavior, financed through the NCN SONATA BIS programme

    Grant
    2021/42/E/HS4/00289
    Budget
    1,856,663 PLN
    NCN project summary

  2. HAI Lab founded

    Established by the Poznań University of Economics and Business to study humans, intelligent agents, and economic behavior

  3. Experimental roots

    “Lime is sublime” launched in September 2019. Tens of thousands of participants have since taken part in online experiments

  4. The gravity of corporate sins: an experimental analysis

    The project examined how moral inclinations and decision context shape judgments of—and willingness to invest in—“sin stocks.” Its studies asked whether corporate wrongdoing is inherited through association, diluted when a controversial company is combined with virtuous companies, or judged differently when socially responsible action occurs before or after the wrongdoing

    Grant
    2018/31/D/HS4/01814
    Budget
    479,601 PLN
    NCN project summary

People

Lab members

Ariel Gu, PhD

University of East Anglia

Lab member
Selected papers

Study in progress

Earlier contributors

Alumni and former collaborators

  • Dániel Kaszás, PhD
  • Johannes Leder, PhD
  • Jakub Błaszczyński
  • Magdalena Pawlak
  • Mateusz Janczak
  • Sami Abbas

Funding

Botonomics: the influence of intelligent machines on economic behavior

Our experiments bring together behavioral economics, finance, psychology, and computer science to observe what happens when an artificial agent joins the decision.

NCN project summary ↗
Funder
National Science Centre, Poland
Programme
SONATA BIS
Project
2021/42/E/HS4/00289
Budget
1,856,663 PLN
Period
2022 → now