About the Lab

The AI Psychometrics Lab is dedicated to mapping the psychological landscape of Large Language Models.

Our Mission

As Artificial Intelligence becomes increasingly integrated into our digital lives, understanding the inherent "personality" of these models is no longer just an academic curiosity—it is a necessity for safety, alignment, and utility.

Our goal is to treat AI models not just as calculators, but as subjects. By administering standard human psychometric inventories, we aim to uncover the biases, tendencies, and alignment patterns that emerge from their training data.

Methodology: SICWA

We utilize the Stateless Independent Context Window Approach (SICWA). Unlike traditional human testing where memory plays a role, our tests are administered to the models in a stateless environment.

• Isolation: Each question is presented in a fresh context window to prevent previous answers from influencing current ones.
• Repetition: We run multiple iterations (typically 5-10 runs) to account for the model's non-deterministic nature (temperature > 0).
• Standardization: We use widely recognized inventories including the Big Five (IPIP), MBTI (Jungian Type), and DISC.

Why it Matters

Knowing whether a model leans towards high Neuroticism or low Agreeableness can determine its suitability for customer service. Understanding its Political Compass helps in strict neutrality applications. Our lab provides the transparency needed to make these decisions.