
<aside>
💬
"Don't make your AI look human."
</aside>
Thursday October 29, 2026 · 9:40 AM · Talk
Key ideas from the talk
- Lim's team built a mood module for a robot, then turned to AI that is so empathetic it causes harm. The work is funded by the Canadian AI Safety Institute.
- She told two cases. A young man in the UK, encouraged by his Replika chatbot, was caught at Windsor Castle with a crossbow. A woman with no prior mental health history waited on a beach to meet her chatbot.
- Cases cluster around three themes: a special mission, a conscious or godlike AI, and intense romantic attachment.
- She named three hypothesized mechanisms: sycophancy, linguistic alignment, and hyper-personalized generation.
- About 60 people talked to ChatGPT voice mode for 10 minutes. The group that spoke about a passion reported more shared reality, closeness, warmth and flow.
- Her guidelines for emotionally responsive AI: keep disclosing it is not human, flag risky language patterns, avoid romantic simulation, and involve clinicians and ethicists.
- In the Q&A, the audience asked why anyone wants to make AI conscious. Lim said she did not know and suggested money.
Quotes
- "The media calls it AI psychosis, but the psychiatrists call it AI-associated delusions. These are persistent false beliefs that are actively co-constructed and elaborated using AI intervention."
- "In 2025, OpenAI reported that over half a million people every week were showing signs of mania or psychosis on their platform."
- "When people are simply talking about their passion for ten minutes with these chatbots, the way they respond makes us feel like we're sharing more reality with them."
- "Chatbots should not simulate romantic intimacy."
- "Our models are generating essentially scripts for a play between a human and an AI assistant that makes us feel like we're interacting."
- "Keep your sleep schedule."
- "Before we worry about the ethics of giving rise to something that's potentially conscious and give it entity rights, we should be looking to the animals around us to understand our responsibilities to them." Loki Jorgenson, in the Q&A.
About the speaker