This summer, I was quite taken aback when Alexandru Voica, Synthesia’s head of corporate affairs, introduced me to their latest PR team member. It turned out to be an interactive virtual avatar of himself, designed to respond to typical media inquiries about Synthesia’s operations and functionalities. Just prior to this, I had participated in a panel discussion where PR professionals debated the acceptability of AI-generated text in pitches. However, Alexandru’s avatar represented an entirely new level, striking me as the ultimate application of AI in public relations.
I received an invitation to Synthesia’s new New York office in September. This UK-born company is a prominent digital avatar startup, competing with firms such as D-ID, HeyGen, and Colossyan. Earlier this year, Synthesia achieved a $4 billion valuation, and last year, it reported exceeding $100 million in Annual Recurring Revenue (ARR).
Synthesia provides businesses with the tools to create interactive training videos using AI avatars. They recently introduced Roleplay Sessions, a feature enabling employees to rehearse scenarios, such as sales presentations, with an AI avatar that offers interactive feedback and scores their performance.
At the new office opening, when offered my own AI avatar, I immediately accepted without hesitation. The prospect of having a digital twin was appealing, especially since I was pleased with my attire and hairstyle that day.
Prior to creating my digital twin, my view on avatars was neutral, though I sensed their growing presence in online daily life was unavoidable. I was aware of individuals on Instagram using avatars resembling themselves to produce social media content. My current lack of hesitation in introducing my own digital twin stems from this evolving interest. Notably, Synthesia had not previously created a digital avatar for a journalist (or indeed, for anyone other than Voica). This particular avatar is programmed to discuss my article detailing why venture-backed startups are more prone to fraud compared to their non-VC-backed counterparts, and it will exclusively respond to inquiries related to that specific piece.
To get started, simply click “start in a new window” below.
Feel free to ask questions such as:
- What prompted the creation of this article?
- What is the focus of the research paper?
- What were the key findings of the researchers?
The creation process involved stepping into a compact film studio within Synthesia’s office, where multiple photos were taken, and a two-minute voice recording was captured. After providing my consent, my digital likeness, ‘digital Dom,’ came into existence. Synthesia developed a personal avatar for me (which simply recites provided scripts), available with and without glasses. Additionally, they fashioned two interactive avatars (capable of listening and responding), also with and without glasses.
Following the selection of an article to train it on, a team subsequently constructed my interactive avatar. This avatar operates through an integration of voice-to-text, video, language, and text-to-voice models. While Synthesia’s proprietary video and voice models form the core of my avatar’s technology, clients are also given the flexibility to opt for alternatives from providers such as Cartesia, ElevenLabs, Google, or OpenAI. Furthermore, enterprises have the choice to host their avatars on their preferred cloud platforms or engage Synthesia for hosting services.
The process involves a voice-to-text model converting spoken words into written text, followed by an agentic language model interpreting this text and generating appropriate responses. A text-to-voice model then transforms these responses into audio, and finally, a video model (developed by Synthesia) provides the visual animation for the avatar as it speaks.
Synthesia offers three primary product categories: a video creation and distribution platform featuring conventional avatars that vocalize user-inputted scripts; an agentic platform known as Sessions, which facilitates interaction with avatars for purposes like surveys or role-playing; and an API platform, enabling users to integrate Synthesia’s video and voice models with external technological services to construct interactive avatars or other innovative products.
The Synthesia team completed my avatars within a few days. Initially, I experimented with the personal avatars, entering a general script to assess the sound of my AI voice. I prompted it to discuss the arrival of fall in New York, a season I particularly enjoy. The voice rendition was quite precise, and I was relieved it didn’t reproduce the hoarseness present during my original audio recording.
When I presented it to friends outside the tech industry, their reactions ranged from intrigued to unnerved.
Next, I demonstrated my interactive avatar. Being deterministic, it is confined to responding solely with information it was trained on, specifically my article on venture fraud. Despite my attempts to ask personal questions, such as my previous employer before TechCrunch or my New York neighborhood, it consistently redirected the conversation back to the venture fraud story.
My friends noted that the interactive avatar’s voice wasn’t highly similar to mine, and its visual representation wasn’t as precise as the personal avatar, though it was sufficiently close to evoke a slight sense of unease. In contrast, my mother lauded it as “amazing,” a significant compliment from her. Both she and my father repeatedly posed questions that only they would know about me, yet the model consistently reverted to discussing the venture fraud story.
After interacting with the model, she lightheartedly remarked, “I don’t recall having two children.”
This encounter prompted me to ponder the prospective evolution of journalism. Would audiences accept news broadcasts delivered by avatars? An investor I consulted promptly dismissed the idea. Undeniably, there’s significant contemporary resistance to the proliferation of uncontrolled AI content on social media and other news platforms. However, others I questioned held more nuanced views. Is it conceivable that avatars could enhance—or even supplant—journalists? Would corporate leaders prefer engaging with an AI avatar of a journalist over a human counterpart?
My enjoyment of journalism stems from connecting with individuals, crafting narratives, and exploring novel subjects. Yet, the cornerstone of this profession is trust, a quality I believe cannot be delegated to artificial intelligence.
Beyond the realm of journalism, I envision the concept of self-cloning as potentially attractive. The burden of post-holiday or vacation catch-up work could be alleviated, as a digital counterpart would be perpetually available to address inquiries.
The future of avatar implementation in corporate America remains to be seen. Possessing my own, however, leaves me with conflicting emotions. Once the initial allure faded, I scrutinized my avatar after it ceased speaking, finding myself waiting—though I’m uncertain for what. Perhaps for it to blink, or utter an unscripted remark, or smile, or simply convey an awareness that it understands.
Given their deterministic nature, my digital twins are incapable of such spontaneous actions. However, I can readily imagine how a non-deterministic avatar, driven by a freely conversational chatbot, could lead someone towards a form of AI-induced psychological distress.
I shared with an investor my belief that, regardless of the trajectory of digital twins, my generation, Gen Z, will probably struggle to fully embrace them. They possess a distinctly sci-fi feel, embodying elements previously confined to cinematic fiction. Nevertheless, I find digital avatars less unsettling than humanoids. With an avatar, should things become peculiar, I retain the option to simply disconnect.
In the meantime, behold my personal avatar, ready to deliver this week’s highlights from our website!