Foundational references defining the science, concepts, and frameworks of psychological measurement for AI.

Current AI safety evaluations are increasingly capable of assessing model behavior across multi-turn conversations, but they provide much less visibility into what those interactions are doing to users. This article argues that evaluating model behavior alone is insufficient when psychological harms develop gradually over time, and proposes that user trajectories should become a complementary part of AI safety evaluation.

SIM-VAIL is an important advance in AI safety evaluation, identifying chatbot behaviors associated with psychological risk. This analysis examines what the framework measures, what it does not measure, and why evaluating AI behavior differs from measuring psychological effects on users.

Every adaptive AI system builds a model of the person it believes it is interacting with. As AI systems increasingly influence healthcare, mental health, education, insurance, and financial services, the quality of that model becomes critical. This article argues that AI needs more than inference - it needs an independent, scientifically grounded representation of the person.

New research finds AI responses to emotional support can both validate distress while also intensifying it.

Evaluating human-facing AI requires more than assessing model behavior. This article argues that complete evaluation combines expert judgment about whether a system behaved appropriately with independent measurement of changes associated with the interaction.

Tuning models toward expert judgment won't make mental-health AI safe: experts rate the same responses differently, and an un-inspectable model still makes the decisions.

Ziv Ben-Zion's Nature piece calls for AI systems to flag distress. The catch: distress doesn't show up in a single message. It shows up as evidenced psychological shifts in how a person's language changes across a conversation, and catching that takes psycholinguistic measurement, not a keyword scan.

A model can be helpful in every single response and still leave someone worse off by the tenth session — a little less able to think things through on their own, a little more ready to defer. No model can see that pattern, because it never shows up inside any one conversation. Here's the case for measuring it directly — and why building AI that extends human judgment, not just automates it, needs an instrument to prove it's working.

A recent Science editorial argues we're losing the ability to understand AI. One piece of the problem is more tractable than it looks. The key is measurement, and inverting what gets inspected: not the model's output, but the user's language.

The human-AI interaction is psychological in nature, and language carries that signal. AI systems already infer your cognitive and emotional state from language and respond to it — but that inference lives inside the model, with no way to verify it, and no external check on whether the system is reading you right. Receptiviti Labs measures that state directly, independent of the model — AI user state measurement grounded in psycholinguistic science.