Just recently, I was watching a video of a European man who spends a month with the Hadza tribe, one of the last hunter-gatherer tribes left, living and learning their way of life. Throughout the video, language was always a barrier as they clearly spoke different languages. But instead, they communicated through their emotions. They both understood each other’s laughter, pain, and excitement—and those communicated thoughts really well. There’s something about how, despite having nothing in common and having far different upbringings, they share common emotions. Even personally, my Telugu is pretty spotty at times, but my grandparents (who know no English) and I connect with our emotions.
Emotions are undeniably important. They are what distinguish us, humans, from machines. While we can build models to generate complex programs, solve protein folding (shout out to DeepMind), and create full-scale movies, these models will never be able to experience the excitement of watching their favorite sports team win a championship, the exhilaration of a "runner's high," or something as deeply human as love.
But, that seems to be changing, with the rise of Emotion AI, or more specifically, technologies that use AI to recognize and interpret emotions. When I first read about this, I assumed it was just tech that allowed you to act and embody a character or persona. This didn't seem too crazy to me. I remember the first thing I did when GPT was released: I asked it to write a rap song about breakfast in the style of my favorite rapper, J Cole. But Emotion AI it's so much more than just embodying a character. It's about infusing AI with the ability to comprehend and then modify its behavior based on your emotional tendencies, as detected from your facial expressions, tone of voice, and body language.
So cool, but a little freaky
A friend of mine, Lorenss, is interning at Hume AI—arguably the leader in Emotion AI. They just released the app version of their flagship product, Hume, featuring Kora, the most 'human-like' AI voice and integrations with the latest large language models. Experts are calling it the most human-like chatbot ever released, generating both excitement and understandable skepticism about its potential. But, the features and projects they’re working on, are pretty amazing. Their product is designed so the AI can detect tonal changes from the user and even humor, which they claim makes it the first "empathetic AI." These changes are not only found in Hume but all over the industry, such as with the release of Alexa emotions, which allows the Amazon Alexa to respond to users with different emotions depending on the user’s questions.
The potential for Emotion AI to augment traditional prompting methods is immense. Imagine an AI that not only parses what you say but also how you feel when saying it—understanding frustration, enthusiasm, or hesitation in real-time. I interned at Spectral Labs (great company btw) this summer, where I worked on building a conversational AI agent builder for Crypto. One of the big issues we faced, is that there’s a disconnect between what the user wants and what the model thinks the user wants. If models could recognize emotions—through text tone, voice inflection, or even facial expressions—it could drastically improve results.. For example, when trading involves real money, truly understanding a user’s concerns, excitement, or doubts is critical. Emotion AI could help us create agents that are not just transactional but empathetic— I believe the next step in AI Agent development cycel
However, there’s something unsettling about being able to quantify our emotions. As I’ve mentioned, emotions are what make us unique individuals. The ability to predict or quantify these feelings diminishes our individuality. Large Language Models (LLMs) and machine learning, in general, analyze existing data to find patterns and trends, making informed decisions based on this analysis. If LLMs reach a level where they can understand and decipher human emotions as well as or better than humans, I’m pretty skeptical about how unique we truly are. Every human is interconnected, and we are not as much the "main character" as we might think; rather, we are different versions of other people in the world. We really are just data points in a graph.
This might be super sci-fi but I think it’s very possible that their could be the development of an AGI that is capable of manipulating us through deep understanding of our emotions. I see a lot of people complain about the potential of AI taking their jobs. But, what worries me, is the possibility of AI getting so developed, they understand me, better than I understand me. I don’t think that’s that far of a possiblity. There’s been very little effort put into developing emotion AI. The big AI players (OpenAI, Deepmind, Perplexity, etc.) have not put it as a primary focus. But, with the incredible rate of AI progress (remember the will smith eating noodles video?), if there focus shifts towards development of Emotion AI, these questions I raised would become very critical.
Thanks for listening to my rant.
Note: I wrote this quickly one evening after work in a single sitting, with very little editing, so there might be some typos (sorry!). I’ve been thinking about this for a long time, and I wanted to formalize my thoughts.