I am SO excited to share my latest #preprint with collaborators @RobbRutledge, @zebkDotCom, @MartinJChadwick, @summerfieldlab, and others from DeepMind: "Increasing happiness through conversations with artificial intelligence" https://t.co/RGmAAGeqBS A short 🧵 1/7

Want to dive deep? Submit a symposium! Have an idea to workshop or results to share? There are posters for that! Want to speak on empirical or theoretical work? Give a flash talk! Want to spark discussion? Chair an innovation forum! So many ways to engage! https://t.co/rdaCYBQVcr

Why do human–AI relationships need socioaffective alignment? As AI evolves from tools to companions, we must seek systems that enhance rather than exploit our nature as social & emotional beings. Published today in @Nature Humanities & Social Sciences! https://t.co/y92riRuvDF

A putative neural correlate of mood! One big (scandalous?) idea, simple analyses, and the STRONGEST brain/behavior correlation I've EVER seen (which is shocking, given that it's mood). Work with: You-Ping Yang, Catrina Hacker & Veit Stuphorn https://t.co/cI3b3uvhWP

Fixed it: --> last year, the U.S. sold more educational services to the rest of the world than it sold in natural gas and coal **combined** --> every $1 of research funded by NIH generates $2.56 in economic activity --> more than 1.1M foreign students come to the U.S. each year

Our goal for this project was to establish a baseline for understanding how human-AI conversations influence happiness. When millions of people are interacting with these AI chatbots daily, even a small effect on well-being can have significant societal ramifications. end 🧵

What about the dynamics within a conversation? We provide evidence for a bidirectional relationship between participants and the AI, such that participants adopt the AI chatbot's language more for negative topics: 6/7

Across all topics, the AI chatbot mirrored participants' sentiment and the emotional tone of both AI chatbot and participants predicted post-conversation happiness: 5/7

We quantified people's unconstrained responses using LLM-assisted sentiment analysis and developed a novel computational model using sentiment PEs to explain happiness ratings. In short, the AI chatbot caused users to express more positive sentiment than expected: 4/7

We found that, across the board, conversations with AI increased momentary happiness compared to journaling, with the largest effect sizes on discussing topics that typically would make people unhappy: 3/7

We started this research with a single question in mind: How do conversations with AI influence subjective well-being? Participants either engaged in conversations with an AI chatbot or wrote journal entries on the same randomly assigned emotional topics: 2/7

This was a joy to work on and is the first of many language-based mental health insights to come from the lab so stay tuned :)

5. Dissociable neural signals for reward and emotion prediction errors https://t.co/YNz1TMvG68

3. A probabilistic map of emotional experiences during competitive social interactions https://t.co/KuomsmZFAY 4. Emotion prediction errors guide socially adaptive behavior https://t.co/GdqsJzmLvx

1. A generalizable framework for assessing the role of emotion during choice https://t.co/f2PwJsvKqc 2. Emotional responses to prosocial messages increase willingness to self-isolate during the COVID-19 pandemic https://t.co/1DDlkYPZiw

If you'd like to read more about our research, you can find the published works from the dissertation below. A huge shout out to all of my collaborators who made these projects possible: Jae Young-Son, @marclluis, @froemero1, Matt Nassar, and of course @orielf

I am so honored and happy to have received the Best Dissertation Award 2023 from @SocforNeuroecon https://t.co/ZyracwEL7S I owe a great amount of gratitude toward my amazing PhD mentor @orielf for her unwavering guidance and support on these projects

Congratulations! I remember seeing bits and pieces of this over the years, it's incredible to see how @harrison_ritz has weaved them together to explain how the brain monitors and uses multiple control streams. Excellent paper!!

New paper out! Large Language Models are biased to overestimate profoundness. https://t.co/F74jn4BmQd

It's wonderful working with thoughtful collaborators @marclluis & @orielf! This project was a delight to work on. We took an unconventional approach toward decision-making and showed that mental representations about uncertainty predict gambling decisions. Details in the paper!