🎧 He Built an AI Model That Can Decode Your Emotions

TL;DR: Today we’re releasing a new episode of our podcast How Do You Use ChatGPT? I go in depth with Alan Cowen, the cofounder and CEO of Hume, a research laboratory developing an empathetic AI model. We dive into the science of emotions and break down how Hume’s AI functions. Watch on X or YouTube, or listen on Spotify or Apple Podcasts. 

The future of AI technology isn’t just faster or more powerful—it’s empathetic. My guest for this episode, Alan Cowen, is leading the charge with the first-ever emotionally intelligent AI.

Alan is the cofounder and CEO of Hume, an AI research laboratory developing models trained to identify and measure expressions of emotion from voice inflections and facial expressions. The best part? Once it understands these emotions, the AI is designed to interact with users in a way that optimizes for human well-being and leaves them with a positive emotional experience.  

Previously, Alan—who has a Ph.D. in computational psychology—helped set up Google’s research into affective computing, a field focused on developing technologies that can understand and respond to human emotions. He operates at the intersection of AI and psychology, and I sat down with him to understand the inner workings of Hume’s models. Alan walks me through the shortcomings of traditional theories of emotional science and breaks down how Hume is addressing these challenges. While talking about the potential applications of the models, we also discuss the tricky ethical concerns that come with creating an AI that can interpret human emotions.

This is a must-watch for anyone interested in the science of emotion and the future of human-AI interactions. Here’s a taste:

• Decode what people really want. Alan believes that emotions are a way through which we can understand people’s hidden desires. “Your preference is whatever is going to make you happier or more awe-inspiring or amused…understanding people's emotional reactions is really key to learning how to satisfy people's preferences,” he says.
• Learn more from every customer service call. If emotions can reveal our hidden desires, the way we speak, coupled with the language we use, can reveal our emotions. “So [voice inflexions are] something that just accompanies every single word. In certain situations, it conveys twice as much information to consider the voice versus language alone,” Alan explains.
• Recognize our shared human experiences. People across cultures experience similar underlying emotions, even though they might use different words to describe it. “[L]et's say that in the U.S., it was more common to say ‘shock’ than ‘fear’ or ‘surprise,’ and in the U.K., people said ‘fear’ and ‘surprise’—we wouldn't say people in the U.S. and U.K. actually experience different emotions, it just turns out that they use different words as their basic vocabulary instinctively,” he says. 

I’m curious to know more about how Hume functions, but before we dive into that, we take a step back to understand how Alan thinks about the theories around emotions. We start by discussing the way science has traditionally divided up emotions into a discrete number of categories or dimensions.

• Paul Ekman's theory identifies six basic emotions—happiness, sadness, fear, disgust, anger, and surprise—that he proposes are universally experienced and recognized across all human cultures. According to Ekman, these emotions are hardwired into our biology and are expressed through universally consistent facial expressions.
• Lisa Feldman Barrett's theory suggests emotions are built from two dimensions: valence (how pleasant or unpleasant something is) and arousal (the level of energy or activation). Barrett argues that our brain interprets sensory inputs using these dimensions, and then constructs the specific emotions we feel based on the context and our past experiences.

Hume’s research is driven by the belief that both of these theories fail to map the full spectrum of emotional experiences because they are based on presumed categories or dimensions of emotion. For this reason, the Hume team has developed a new approach called the semantic space theory.

• Semantic space theory is best understood in the way it is different from the traditional theories discussed earlier. Instead of assuming the existence of certain types of emotions, it examines emotional responses and maps the dimensions of emotional experiences. Alan explains, “The general approach of emotion scientists has been, ‘Let's posit what emotions are and then study them in a confirmatory way,’ and semantic space theory is doing something different: ‘Let's posit the kinds of ways we can conceptualize emotions and then derive from the data how many dimensions there are, what the best way to talk about them is, how people refer to them across cultures and so forth.’” 

In order to study emotions through the lens of semantic space theory, Hume is leveraging technological advances in computing, data collection, and online data storage to measure what people are feeling. Alan shared two methods the company employs to do this:  

• Labeling emotional states, or asking people to label how they are feeling in different moments of a voice or video recording, and then finding common patterns or dimensions that explain these feelings across individual data points.
• Evaluating facial expressions, or studying videos of people across different cultures and contexts with the intent of measuring their facial expressions.

Alan and I discuss some of the insights that the Hume team has discovered as they dive deep into the science of emotions:

• Identifying similarities across cultures. While measuring emotion through facial expressions, the team found that there is consistency among the emotional experiences of culturally diverse people. “[W]hen you take out the labels, you realize that there's more cultural consistency, which implies that expressions have similar meanings and the language actually imposes more cultural differences on how those meanings are interpreted and what's salient to people depending on what kinds of experiences they've had in life,” Alan remarks.
• Accounting for individual differences. The Hume team unearthed similarities among cultures, but their model is also trained to account for individual idiosyncrasies by analyzing subtle behavioral cues. “[I]n order to be able to make good predictions, [the model] appreciate[s] individual differences and resting facial expressions and resting voices and how people modulate their voices over time and so forth,” Alan says.

Alan tells me that his mission for Hume is to develop an empathetic AI that “truly understand[s] what humans want,” “measure the impact” that it would have on people’s emotions, and train it to “optimize for the positive emotions that people can have in life.” Given this context, we discuss some of the ways in which Hume can be used:

• AI that people love talking to. Since Hume is optimized to give the people it interacts with positive emotional experiences, Alan thinks potential use cases for the model include customer service, a character in a video game, and some version of a therapist app. “[P]eople talk to it in a kind of a way they might talk to a therapist or a friend and really get something out of it because it's already optimized for people to be satisfied coming out of the conversation, and naturally it does these things that people enjoy,” he explains.
• Hume’s potential as an interface. Hume has many applications because it can be integrated into different products to optimize user experiences. “[I]t's an interface you can build into anything—products, apps, robots, wearables, refrigerators, whatever you want…the core of what [the] AI is doing is trying to figure out, ‘What bits can I flip to make you happier?’” Alan says.
• Not your average AI assistant. While discussing Hume’s uses as an interface, Alan explains that it’s distinct from an AI assistant because as opposed to carrying out tasks for users, Hume serves as a seamless intermediary between users and the underlying technology. “We're trying to take the tools that developers are building for an AI to operate and be the interface that the person talks to that deploys those tools,” he explains. 

As we explore Hume’s applications, we also discuss some of Alan’s concerns about the ways in which the AI could be misused: 

• Align the use of AI with ethical concerns. To prevent misuse, Alan intends to evaluate the use cases the company pursues against the guidelines for AI ethics created by Hume Initiative, the nonprofit associated with the company. “I think there's ultimately an alignment between the long-term interests of the business, which are obviously to make money, but also the long-term interest of humanity, which is we won't permit AI to destroy our society,” he says.
• AI that indexes for human well-being, not fragmented attention spans. Having said that, Alan believes that AI characters designed to maximize engagement is one of the avenues in which Hume could be misused, and wants to select for well-being instead. “[AI characters] should be optimized for somebody's health and well-being and not for somebody's engagement because if it's optimized for engagement, it can manipulate you to be sympathetic to it in ways that are inappropriate,” he explains.

You can check out the episode on X, Spotify, Apple Podcasts, or YouTube. Links and timestamps are below:

• Watch on X
• Watch on YouTube
• Listen on Spotify (make sure to follow to help us rank!)
• Listen on Apple Podcasts

Timestamps:

1. Dan tells Hume’s empathetic AI model a secret: 00:00:00
2. Introduction: 00:01:13
3. What traditional psychology tells us about emotions: 00:10:17
4. Alan’s radical approach to studying human emotion: 00:13:46 
5. Methods that Hume’s AI model uses to understand emotion: 00:16:46 
6. How the model accounts for individual differences: 00:21:08
7. Dan’s pet theory on why it’s been hard to make progress in psychology: 00:27:19
8. The ways in which Alan thinks Hume can be used: 00:38:12
9. How Alan is thinking about the API v. consumer product question: 00:41:22
10. Ethical concerns around developing AI that can interpret human emotion: 00:44:42

What do you use ChatGPT for? Have you found any interesting or surprising use cases? We want to hear from you—and we might even interview you. Reply here to talk to me!

Miss an episode? Catch up on my recent conversations with LinkedIn cofounder Reid Hoffman, a16z Podcast host Steph Smith, economist Tyler Cowen, writer and entrepreneur David Perell, Notion engineer Linus Lee, and others, and learn how they use ChatGPT.

If you’re enjoying my work, here are a few things I recommend:

• Subscribe to Every
• Follow me on X
• Subscribe to Every’s YouTube channel
• Check out our new course, Maximize Your Mind With ChatGPT

The transcript of this episode is for paying subscribers.

Dan Shipper is the cofounder and CEO of Every, where he writes the Chain of Thought column and hosts the podcast How Do You Use ChatGPT? You can follow him on X at @danshipper and on LinkedIn, and Every on X at @every and on LinkedIn.