Jack Dorsey's AI-focused company Block has unveiled a new conversational AI model named Dessa, designed to simulate the experience of chatting with an awkward but genuine new friend. The model, trained from scratch exclusively on human conversation data, aims to create more natural and relatable interactions compared to traditional chatbots.
Training on Authentic Dialogue
Dessa's development marks a significant departure from conventional AI training methods that often rely on large datasets of pre-existing text. Instead, Block's team focused on creating a model that learns directly from authentic human conversations. According to Ragavan Thurairatnam, lead on the Dessa project, the model's training data consists entirely of real dialogue exchanges, which helps it understand the nuances of human communication.
Awkwardness as a Feature
One of the most distinctive aspects of Dessa is its ability to produce responses that can feel slightly awkward or uncanny—intentionally so. In a demonstration shared by the team, Dessa engages in a conversation that reflects the uncomfortable yet honest nature of meeting someone new. The AI's responses sometimes miss the mark or come across as slightly off-putting, mimicking the real-world experience of social interaction where people often struggle to find the right words.
This approach challenges the typical goal of AI chatbots to be perfectly smooth and helpful, instead embracing the imperfections that make conversations feel human. While this may not be ideal for all applications, it could be valuable for creating more relatable and emotionally intelligent AI systems.
Implications for AI Development
Block's Dessa represents a shift in how AI models are being designed, focusing on authenticity over perfection. As the company continues to refine its approach, Dessa could influence how other organizations think about conversational AI. The model's unique training methodology may pave the way for more emotionally intelligent and socially aware AI systems in the future.