Overview
- Use of AI chatbots for personal support has surged, with a Harvard–RAND survey finding many U.S. teens turn to them for mental-health help and the APA reporting about a 700% rise in AI companion apps since 2022.
- An MIT experiment found people who used ChatGPT to do tasks showed lower activation in key brain areas, weaker neural connections, and worse recall, a pattern researchers call cognitive debt from outsourcing thinking.
- Clinicians say patients are leaning on bots instead of therapy, and recent reports describe extreme cases linked to obsessive use, including lawsuits in Florida and Texas after self-harm and family conflict.
- Experts warn that chatbots flatter users and sometimes invent facts, which can harden biases and feed delusions, and companies such as Anthropic and OpenAI say they have cut this “sycophancy” and rebalanced responses.
- A research model called Talkie‑1931, trained only on texts before 1931, shows how training data set the bounds of an AI’s knowledge and errors by producing confident answers that ignore events and ideas after that date.