A New York-based startup, Hume AI, unveiled a groundbreaking AI voice interface, the Empathic Voice Interface (EVI), designed to mimic human emotions and engage in more natural conversations with users. This emotionally intelligent AI can be integrated into various applications, spanning customer service, healthcare, and virtual and augmented reality environments. Following a successful funding round of$50 million led by EQT Ventures, the beta version of EVI has been launched to the public.

Unlike conventional AI systems guided by superficial human ratings, Hume AI's approach involves training the AI to understand user preferences and generate vocal responses optimised for user satisfaction. By learning directly from proxies of human happiness and continuously updating its knowledge with each interaction, EVI aims to reconstruct human preferences from first principles, thus enhancing its ability to engage with users effectively.

Its empathic Large Language Model (eLLM) is central to Hume AI's innovation, which integrates multimodal generative AI with expression measures. This model enables EVI to adjust its language and tone based on context and the user's emotional expressions, creating a more personalised and engaging conversational experience. The voice model was trained on extensive data from millions of human interactions, enhancing its ability to effectively understand and respond to users' needs.

Why does it matter?

While Hume AI leads the charge in emotionally intelligent AI, other researchers and institutions also explore similar avenues. Columbia University's Hod Lipson has developed an AI-powered robot capable of predicting human expressions and replicating them. At the same time, scientists from South Korea's Ulsan National Institute of Science and Technology have created a facial mask that records verbal and non-verbal expression data. These innovations hint at a future where technology seamlessly integrates human emotions into various applications, from virtual reality concierges to personalised services based on users' emotional states.