2/
Generative AI is reshaping experimental research.
But how do we ensure replicability when AI introduces randomness and variability?

3/
We propose a new experimental model:
🔁 Experiment-AI-Subjects Triad (EAIST)
AI becomes part of the experimental loop, generating stimuli and conditions.

4/
This raises new challenges:
🤖 Adaptive AI behavior
🎭 GAN-generated facial expressions
💬 Emotionally expressive chatbots
All can undermine experimental control.

5/
To address this, we offer a framework based on Open Science:
✅ Modular testing
✅ Parameter fixation
✅ Structured prompt engineering
✅ Robust experimental design

6/
Though rooted in social robotics, this framework applies to any AI-driven experimental research, especially in the social sciences.

7/
Authored by:
Antonio Aquino, Fabio Aurelio D’Asaro, Rocco Gaudenzi, Marco Lezcano, Vittorio Iacovella, Michela Vezzoli, and Michele Scandola
🧠 ITRN AI Working Group