Does AI Have Emotions? Functional Emotion Representations Inside LLMs
❗Please Read❗
This is not a seminar! Participants are to read the material in advance to engage more fully with the technical and methodological details during the session.
If you have your own research, experience or saw something on the internet (blogpost, article etc.) that would add to the discussion, do contribute!
Suggested Pre-Reading:
Main paper:
Lineage papers:
Modern language models do not just talk about emotions. Anthropic’s interpretability team recently found 171 emotion-related representations inside Claude Sonnet 4.5, from “happy” and “afraid” to “desperate,” “brooding,” and “appreciative.” These representations activate in contextually appropriate situations and causally shape the model’s behaviour, including its tendency toward reward hacking, sycophancy, and blackmail in evaluation settings.
This session focuses on emotions as functional representations. Neuroscientist Antonio Damasio argued that in humans, emotion is not the opposite of reason, but part of how we assign value, mark urgency, and make decisions under uncertainty. If a language model develops emotion-like internal states that perform some of the same functions—without a body, subjective experience, or biological need—how far does the analogy go?
The session will also discuss what this means for alignment, safety, and conversational AI. Can “desperation,” “calm,” or “anxiety” become useful safety signals? Should we monitor or steer these states? And what changes when systems people talk to every day are shaped by internal affective representations we are only beginning to understand?
More About the Host
Zacchaeus Chok is Co-Founder & CEO of Marymount Labs, where he builds conversational AI systems for behavioural engagement. His work uses everyday messaging channels to help organisations guide people toward timely action, with applications in chronic disease support, preventive care, and eldercare outreach. At the NUS Health Informatics Research Lab, he helps oversee digital health deployments for clinical research programmes, and is also a co-founder of HealthGen, which commercialises a SingHealth-NUS chronic disease management platform validated in a 1,000-patient randomised controlled trial. Zacchaeus studied Computer Science at NUS as a Stephen Riady Scholar and completed the NUS Overseas Colleges Silicon Valley programme.
More About the Series
Paper Club is Lorong AI’s community-driven initiative where members gather to discuss and analyze academic papers, research articles, or key developments in artificial intelligence.
Get involved: Learn more about Lorong AI | Speaker Sign-up | WhatsApp Community | LinkedIn | X