Soon, AI companies expect to be selling “agents” — AIs that reserve tickets, cancel subscriptions and take other actions for their users. The companies are also talking up the power of generative AI to hold a natural conversation in real time. And, of course, AIs are also getting better at creating videos of real people.
Combine all three, and you get the the new realm of the “digital human avatar” — an AI creation that can represent you, speak for you, take action on your behalf. It looks and sounds and acts as close to you as an AI can make it. While a lot of robot and AI visions remain science fiction for now, the digital avatar is already a real product, with a multibillion-dollar global market that is expected to grow a lot in coming years.
It’s a wild vision: Imagine your avatar looking for a dental appointment, talking to the dentist’s avatar about whether Tuesday has an opening — while you and the dentist do (one hopes) more fulfilling things. Or think about a version of you that can talk in a meeting (same obsession with the deadlines, same way of nodding your head after you’ve finished talking, but real-you is elsewhere). Maybe you could have a version of you that you bounce ideas off — taking “talking to yourself” into a new dimension of meta.
It’s not hard to imagine things going wrong with this scenario, but like it or hate it, these products are already here, and rapidly evolving.
So I was happy to have a chance to talk with Hassaan Raza, co-founder and CEO of Tavus, a startup that has been making digital twins — of real people and of fictional characters — since 2020. We talked about the steps digital twins must take to win users’ confidence, the challenges of making a twin seem real, if and why he expects avatars eventually to leave the screen and become robots moving in the real world. And of course we also looked what could go wrong — deception, non-consensual recreations of people, privacy violations, the limits of AI’s language abilities, and other issues.
Caveat: As always, remember that the transcript is AI-generated and. might contain errors.
When You Can't Talk to Me, Talk to "Me"