Roshni Lulla
Bio
Updated 07/08/26 · Provided by member · VerifiedI'm the co-founder and Chief Research Officer of the Institute for Humane Robotics, where I work on embodied AI safety and the risks of deploying increasingly capable systems into the physical world. My research targets the dissociation between cognitive and affective empathy in AI systems, using the Dark Triad as a model-organism framework for studying misalignment. I use a combination of psychologically grounded evals and mechanistic interpretability to test whether sycophancy and antisocial behavior (such as deception and exploitation) dissociate at the feature level. My background is in computational and affective neuroscience. I hold a PhD from USC's Brain and Creativity Institute, where I used fMRI to study the neural basis of empathy and moral cognition, and I bring that grounding to how misalignment shows up in models that can imitate care without sharing it. I've published on empathy, moral values, and misinformation, and my current work sits at the intersection of interpretability, evaluations, and moral cognition.
Links
Updated 07/08/26 · Provided by member · VerifiedProjects
Grants
Updated 07/08/26 · By grantmaking.aiNo grants recorded.