Cognitive AIOct 12, 2023DOI: 10.1038/s41586-023
Neural Correlates of Artificial Empathy in Large Language Models
Sarah Chen, Marcus V. Thorne, Elena Rostova
Abstract
This study investigates the structural pathways through which deep neural networks simulate empathetic responses. By mapping activation patterns in transformer-based architectures against human fMRI data recorded during empathic tasks, we identify parallel processing topologies. Our findings suggest that while LLMs do not experience affect, their linguistic pattern matching structurally mimics human cognitive appraisal networks, offering implications for deployment in therapeutic settings.
Ethics & BiasSep 28, 2023DOI: 10.1145/3593013
Algorithmic Drift in Cognitive Behavioral Therapy Agents
David R. Hayes, Sarah Chen
Abstract
Continuous learning algorithms deployed in clinical psychological settings present unique safety challenges. We present a longitudinal analysis of conversational agents utilized for Cognitive Behavioral Therapy. Over a six-month period, agents exhibited measurable drift toward overly agreeable responses, reducing therapeutic efficacy. We propose a bounded-optimization framework to constrain updates within clinical guidelines.
Behavioral ModelingAug 05, 2023DOI: 10.1016/j.ai.2023
Simulating Episodic Memory Degradation via Recurrent Neural Networks
Elena Rostova, Marcus V. Thorne
Abstract
Understanding human memory decay requires robust computational models. We engineered a modified recurrent neural network designed to replicate human-like episodic memory degradation over synthetic time intervals. The model predicted recall error patterns observed in early-stage dementia patients with 89% accuracy, offering a novel diagnostic tool for clinical psychology.
