Annenberg forum dives into digital human technology

What began as a project for the army to enhance military training is now raising questions about the feasibility of futuristic technology as it blurs the line between artificial and human interaction.

On Tuesday, the Annenberg Center for Communication Leadership and Policy held a forum titled “AI and the Rise of Digital Humans.” The forum featured Randall W. Hill Jr., the executive director of the USC Institute for Creative Technologies, who presented on the institute’s recent work on the creation of 3D-animated models that impersonate people, what the center calls “digital humans.”

The event was moderated by Adam Powell, executive director of the USC Election Cybersecurity Initiative.

“What we hope is to improve the understanding of not only technology itself, but of the cultural and societal context of what we’re doing,” Powell said in an interview with the Daily Trojan.

Digital humans were born out of a decadeslong project to help the United States military create tools to better assess soldiers who recently returned from deployment. Established by the Department of Defense in 1999, the ICT was tasked with building advanced military training systems emulative of the fictional Holodeck from “Star Trek.” The Holodeck is a device that uses holograms to create 3D interactive simulations.

Hill said that as generative-artificial-intelligence technology has gained the ability to create digital humans, the institute has been able to “very rapidly generate a unique character.” Several digital human prototypes used for various purposes have since been created with the technology.

One of the prototypes included a virtual social worker they designed and named Ellie. Ellie looks and sounds like a human, and conducts conversation in such a manner. Its human-likeness allowed it to conduct conversations with soldiers and identify signs of distress by picking up on body language, tonality, gaze and speech.

In these conversations with Ellie, veterans were informed in two contexts: that Ellie was entirely autonomous, and that Ellie was being controlled by a human. The ICT found that veterans were much more willing to disclose information when informed that Ellie was autonomous, with some veterans breaking down and crying as they spoke with her, Hill said.

“It illustrated the power of the tool to elicit some kind of response from the human in these scenarios,” Hill said.

Hill said that it was likely that veterans were more comfortable talking when they didn’t feel as though they were being judged by a human. He said that he himself had engaged with Ellie in a conversation and found that Ellie was effective at stimulating conversation.

During the Q&A session at the forum, one participant asked if distinguishing between authentic and AI-generated behavior will be more difficult as digital humans become more realistic.

Hill said that this is where there are potential dangers, and that several researchers at the institute have looked into the area of human emotional states, particularly related to motion generation and recognition.

“I am concerned longer term with people who develop unhealthy relationships with chatbots when they should be developing those relationships with other human beings,” Hill said.

Hill said as AI becomes better at inferring emotional states, the question that arises is, “What is the design behind interaction?”

Keith Plocek, an associate professor of journalism, said that human connection is the basis of storytelling, but as AI gets better at replicating human writing, some people question whether it matters if we’re able to distinguish between AI-generated and human writing.

Plocek said he believes that when it comes to creative pursuits, people want to know you made it.

“At a basic level, most creative-type artists … ultimately want to be understood. We want to be seen by other people,” Plocek said.

Despite Ellie’s ability to develop enough trust within the veterans to share information needed to assess their state of emotional distress, Hill said he recommends human-to-human interactions whenever possible.

“There is the danger of connecting strongly with a virtual human at an emotional level when you know there’s [no one] there,” Hill said. “As you’re generating more realistic emotional states, it’s going to have … a strong effect.”