Receptiviti Labs is the AI-focused arm of Receptiviti, developing scientifically grounded measurement infrastructure for human-state-aware AI systems.
Our work focuses on making cognitive and psychological interaction dynamics measurable, observable, and usable within AI evaluation, monitoring, and conversational systems.
Current AI systems adapt to users implicitly through language, but the underlying interaction-state signals influencing model behavior remain largely opaque, inconsistent, and unavailable to the systems meant to evaluate or govern them.
We are interested in approaches that make interaction state explicit, measurable, longitudinally trackable, and inspectable.
200+ psychological dimensions · 34,000+ peer-reviewed citations · 30+ years of psycholinguistic research
Our current work explores how explicit interaction-state measurement can support:
Salecha, Ireland et al., 2024 — Large language models display human-like social desirability biases in personality surveys. PNAS Nexus. LLMs shift responses when they infer they're being evaluated, with effects up to 1.20 human SD across GPT-4, Claude 3, Llama 3, and PaLM-2. Co-authored by Molly Ireland (Receptiviti).
Entwistle, Hoemann, Nightingale & Boyd, 2025 — Psychosocial dynamics of suicidality and nonsuicidal self-injury: a digital linguistic perspective. npj Mental Health Research. Large-scale naturalistic study of language dynamics surrounding suicidality and self-injury across 992 individuals, 66,786 posts. Co-authored by Ryan Boyd (UT Dallas).
Boyd & Markowitz, 2026 — Artificial intelligence and the psychology of human connection. Perspectives on Psychological Science. Introduces the MIRA model — a framework for when and how AI functions as a relational entity in human ecosystems, with language as the primary modality through which that relationship operates.
Vu, Boyd, Eichstaedt et al., 2026 — PsychAdapter: adapting LLMs to reflect traits, personality, and mental health. npj Artificial Intelligence. A lightweight architectural modification generating text that reliably reflects Big Five personality traits (87.3% accuracy) and mental health variables (96.7% accuracy).
Chi, Ganesan, Boyd, Ungar & Guntuku, 2026 (preprint, under review) — When support escalates distress: regulation and escalation in LLM responses to venting and advice-seeking. Across 178,800 Reddit posts, LLM responses to venting simultaneously regulate and escalate distress — escalation invisible to standard safety evaluations.
Full list and ongoing work: Research
Our measurement approaches draw on validated psycholinguistic and behavioral research methods designed for structured observation of interaction dynamics within AI systems.
We are particularly interested in evaluation and observability approaches that treat interaction state as measurable infrastructure rather than latent implicit inference inside model behavior.