Recent work on Empathic AI has focused on whether people like receiving empathy from AI. But the top use of generative AI in the past year has been for therapy and companionship, so we already know people are going to AI for emotional support. The more important question is: what happens to everyone else? When we learn that others are turning to AI for empathy, do we still feel compelled to show up for them ourselves? 

My research suggests we don't. I find evidence for "empathy offloading:" when people see AI as capable of providing emotional support, they become less willing to do the hard emotional labor of empathizing with others. Giving empathy is morally good, but it is uncomfortable and costly work. When AI can do it instead, people are happy to let it. This has real consequences for human relationships. Kindness revolves around exchanging empathy, and if empathy seekers go to AI and empathy givers stop showing up, the kindness that holds relationships together starts to disappear. My work reveals the dark side of Empathic AI: empathy offloading can quietly break the human connections it was meant to support.

Victimhood is one of the most powerful forces in human social life. When a person is recognized as a victim, others offer empathy, provide resources, and demand accountability. When a claim is doubted or dismissed, the same person may face indifference, suspicion, or hostility. But nearly everything we know about how people respond to victimhood comes from controlled experiments where participants read brief vignettes and rate their reactions on Likert scales. These studies don't capture how victimhood operates as people actually encounter it in the real world: multimodal, dynamic, and socially contested. A real victimhood claim has a face, a voice, a production style, and an audience that responds in real time with empathy, skepticism, counter-claims, or offers of help. 

Using the official TikTok Research API, I am conducting the first large-scale empirical investigation of how victimhood functions in naturalistic online settings. I develop a measurement architecture that captures both sides of the victimhood communication process: what narrators express and how audiences respond. On the expression side, I analyze facial expressions, vocal prosody, and structural presentation choices using machine learning methods. On the perception side, I analyze comments using NLP and machine learning methods and engagement metrics. By crossing the two, I can identify which features of a narrator's multimodal presentation predict whether audiences believe or dismiss their claim. Over five billion people now use social media, meaning billions of people are deciding who deserves support or backlash, which movements succeed or stall, and how communities unite and divide. Understanding how victimhood actually works in this environment matters because these judgments shape real outcomes across law, politics, and everyday life.

As a society, we value efficiency. AI has been a great help in delivering it, and has even beat us in many areas. But in our pursuit of efficiency, we have crossed an invisible moral line. From conducting job interviews to generating fashion models to writing bar exam questions, more and more companies are outsourcing tasks to AI -- tasks that carry social, intellectual, or ethical significance to human employees and consumers. This is not just about adopting a new tool. It is about replacing human involvement in moments that shape people's careers, opportunities, and identities. And when companies make this choice, it signals something about their character: that the people affected by their decisions are not worth the human time, care, or attention. My research examines how consumers perceive companies that outsource to AI, and finds that reactions go beyond doubting the algorithm's competence. People are making moral judgments about what the company chose not to do. When dozens of executives, engineers, and managers agree to let machines take over human judgment, it amplifies the moral signal. It communicates detachment, indifference, and self-interest rather than integrity or care. Outsourcing aversion is not algorithm aversion. It is a judgment about company character.