Businesses may soon be able to designate duties like customer support or training exercises to life-like virtual personas through a new tool from artificial intelligence startup Soul Machines.
The San Francisco-based company debuted a digital brand studio this week that lets clients customize their own digital person by choosing from a set of realistic CGI avatars and uploading conversational trees with natural language processing systems from Google or IBM.
Founded by Academy Award-winning visual effects engineer Mark Sagar and entrepreneur Greg Cross at the University of Auckland in New Zealand, Soul Machines is part of a small but growing scene of startups exploring how life-like human avatars can be put to use in business contexts, whether as virtual influencers, extensions of celebrity personalities or for interpersonal interaction practice.
“The objective here is not to replace people, it’s to really focus on things that are very, very difficult for organizations to deliver, like infinitely scalable customer interactions or infinitely scalable customer support at a completely different level of economics,” Cross said.
The company has raised $47.5 million to date from investors including Hong Kong-based Horizon Ventures and Salesforce’s venture capital arm. It has already worked with a select set of clients on customer support avatars, including a digital customer service rep for Air New Zealand named Sophie, a car salesperson named Sarah for Mercedes-Benz and a virtual financial advisor named Jamie for Australia and New Zealand Banking Group. It also built a digital version of Will.I.Am that was featured on Robert Downey Jr.’s YouTube series about AI.
Cross said it’s the quality of the animation that sets Soul Machines apart from competing avatar companies, though he stresses that the intent was never to deceive people into believing they are conversing with a real human.
“We absolutely believe that the highest quality CGI and the highest quality animation are an absolute prerequisite for this technology to be successful,” he said. “[But] we’re certainly never trying to fool people; we’re always very upfront—and our customers are upfront—that you’re talking to a digital person, not a real person.”
While the avatars can detect and respond to facial cues and preprogrammed queries across a variety of languages, responses are relegated to a set of prompts written by the brand, putting them nowhere near the level of free-wheeling, versatile intelligence one might expect from a science-fiction notion of AI. Dialogue generated from whole cloth is still very much in experimental research stages and too unpredictable—and not brand-safe—for commercial use, although some companies are beginning to experiment with it.
“It’s likely to be quite some time—and I’m talking even probably at least five years—before big brands and organizations are going to trust technology like this to generate conversational content on the fly,” Cross said.