ai: I created an interactive digital avatar of myself — and you can

TechCrunch Senior Reporter Dominic-Madori Davis has unveiled an interactive digital avatar of herself, developed by the AI video-generation startup Synthesia. The advanced AI replica, capable of engaging in conversation, marks a significant step in personalized AI technology and raises intriguing questions about the future of digital interaction and even journalism. This pioneering move comes after Synthesia, a company recently valued at $4 billion, invited Davis to their new New York office to experience their cutting-edge avatar creation process firsthand.

The creation process for Davis's "digital twin" involved a visit to Synthesia's specialized film studio. Here, numerous photographs were taken, and a two-minute voice recording captured her vocal nuances, all with her explicit consent. The process yielded both "personal avatars" that recite pre-written scripts and "interactive avatars" designed to listen and respond dynamically.

These interactive versions, including one trained specifically on Davis's investigative piece about venture-backed startup fraud, represent Synthesia's first such deployment for a journalist, following a similar initiative for their Head of Corporate Affairs. Synthesia, a U.K.-based firm with a reported $100 million in annual recurring revenue, is a key player in the digital avatar space alongside competitors like D-ID and HeyGen. Their core offerings include enterprise solutions for interactive training videos and their new Roleplay Sessions, which enable employees to practice scenarios with AI avatars that provide real-time feedback and scoring.

The Technology Behind the Digital Twin

The underlying technology for Davis's interactive avatar integrates Synthesia's proprietary video and voice models with voice-to-text, agentic language, and text-to-voice models, allowing for complex, conversational interactions. This sophisticated stack can also incorporate alternatives from other labs like Cartesia, ElevenLabs, Google, or OpenAI, offering flexibility for enterprise clients. The entire avatar creation process for Davis took only "a couple of days," demonstrating the efficiency of modern AI tools.

User Experience and Early Reactions

Davis observed distinct differences between her personal and interactive avatars. While the personal avatar's voice was remarkably accurate, even smoothing out a slight hoarseness from her recording day, the interactive version's voice and likeness were perceived by friends as "not as good," though still "close enough to be somewhat creepy." The interactive avatar is deterministic, meaning it strictly adheres to its trained data—in this case, Davis's venture fraud article—and redirects attempts to ask off-topic questions. Despite the minor discrepancies, Davis's mother found the interactive avatar "amazing," highlighting the mixed reactions this nascent technology can elicit.

Implications for Journalism and Beyond

The experience has prompted Davis to reflect on the broader implications of AI avatars, particularly for fields like journalism. She contemplates whether AI-generated news presenters could become acceptable or if CEOs might prefer interacting with AI journalist avatars over human counterparts. Crucially, Davis emphasizes that trust, a cornerstone of journalism, seems irreplaceable by AI. Beyond media, the concept of a digital "clone" capable of handling routine inquiries could be transformative for corporate environments, potentially eliminating post-vacation backlogs by ensuring an "always-on" version of oneself.

However, Davis also harbors "mixed feelings" about her digital twin. She notes the potential for "AI psychosis" if avatars were nondeterministic and allowed to freely pontificate, expressing a preference for the deterministic models. Despite finding digital avatars "less jarring than humanoids," she acknowledges a generational divide, predicting that Gen Z might find the sci-fi-like reality of digital twins unsettling. For now, the ability to simply "log off" provides a comforting escape should interactions become too strange.

FAQ

Q: What is a "deterministic" interactive avatar?

A: A deterministic interactive avatar is trained on specific content and will only respond with information directly derived from that content. It cannot generate new, unrelated responses or engage in free-form conversation outside its programmed scope.

Q: How was the TechCrunch reporter's avatar created?

A: The reporter's avatar was created by Synthesia in a mini film studio. This involved taking numerous photos and recording a two-minute voice sample. The interactive version was then trained on a specific article written by the reporter.

Q: What are the main types of products Synthesia offers?

A: Synthesia builds three primary product types: a video-creation and distribution platform with classic avatars that read scripts, an agentic platform called Sessions for interactive experiences like surveys and roleplay, and an API platform allowing integration of their video and voice models with other tech services for custom interactive applications.