I was at the Chattanooga Convention Center for Project Voice, a major gathering for software developers, venture capitalists, and entrepreneurs in conversational AI. The conference, now in its eighth year, was run by Bradley Metrock, an uncommonly tall man with rousing frat-boy energy who is, per his professional bio, a “leading thought leader” in voice tech. “I’m a conservative guy!” he said to me on a Zoom call some weeks prior. “I was like, ‘What kind of magazine is this? Seems pretty out there.’”
So that what? So that no one could accuse the AI pioneers of ignoring existential threats to culture? To facilitate a brief moment of self-flagellation before everyone hit the bars? I wasn’t sure, but I sensed my presence had less to do with balance and more to do with sport. Bradley kept using the word “exciting.” A few years ago, he said, the contrarian speaker stormed onstage, visibly irate. As she railed against the wickedness of the Echo Dot Kids, Amazon’s voice assistant for children, a row of Amazon executives walked out. Major sponsors! That, said Bradley, was very exciting.
I wondered if I should be offended by my contrarian designation, which positioned AI as the de facto orthodoxy and framed any argument I could make as the inevitable expression of my antagonistic pathology. The more I thought about it, the more I became convinced I was being set up for failure. Recent discussion of conversational AI has tended to treat the technology as a monolithic force synonymous with ChatGPT, capable of both cultural upheaval and benign comedy. But conversational AI encloses a vast, teeming domain. The term refers to any technology you can talk to the way you would talk to a person, and also includes any software that uses large language models to modify, translate, interpret, or forge written or spoken words. The field is motley and prodigious, with countless companies speculating in their own little corners. There are companies that make telemarketing tools, navigation systems, speech-to-text software for medical offices, psychotherapy chatbots, and essay-writing aids; there are conversational banking apps, avatars that take food orders, and virtual assistants for every industry under the sun; there are companies cloning celebrity voices so that an American actor can, for example, film a commercial in Dutch. The field is so crowded and the hype is so loud that to offset a three-day circus with thirty minutes of counterpoint is to practically coerce the critic into abstractions. Still, I accepted the invitation for the same reason I took the job with the real estate start-up: it was a paid opportunity and seemed like something I could write about.
On the first afternoon of the conference I took a lap around the floor and tried to make sense of what I was seeing. Tech companies had arrived with their sundries: bowls of wrapped candies, ballpoint pens, PopSockets, and other bribes; brochures fanned on tables; iPads with demos at the ready. The graphics, curiously alike across the displays, were a combination of Y2K screen saver abstractions and the McGraw Hill visual tradition. Many companies had erected tall, vertical banners adorned with hot-air balloons, city skylines at dusk, dark-haired women on call-center headsets, and circular flowcharts with no discernible content. If conversational AI had a heraldic color, that color would be blue — a dusty Egyptian blue, chaste and masculine, more Windows 2000 than Giotto. It’s a tedious no-color, the color of abdicating choice, and on the exhibition floor it was ubiquitous in calm, flat abysses backgrounding white text.
***
The only booth that stood out was at the far end of the exhibition hall. A company had tented its little patch of real estate with an inflatable white cube that looked like a large, quivering marshmallow. Inside the cube was Keith, a soft-spoken man whose earnest features and round physique conveyed a gnome-like benevolence. Beside Keith was a large screen. On the screen was a woman. The woman had dark hair, dark eyes, and purple lips that endeavored a smile. Her shoulders rose and fell, as if to suggest the act of breathing, and though she looked toward me, her gaze was elsewhere.“This is Chatty,” Keith shouted over the roar of the blowers keeping his enclosure erect.
Keith worked for SapientX, a company that makes photorealistic conversational avatars powered by ChatGPT. SapientX had custom-built Chatty for Project Voice. Chatty could answer questions about the conference agenda and show you a map of the exhibition floor, except she couldn’t do it just then, said Keith, because they couldn’t seem to connect her to the wi-fi.
Keith was happy enough to walk me through the visuals. Chatty’s face was the collaborative effort of fifty different companies. A company in Toronto did the eyes. “There’s like eight guys and all they do is eyes all day,” he said.
Chatty’s face was a composite of several different races. Her voice was a composite of several different women. Her voice still needed some work, he admitted. “Right now she’s kinda mean.”
I picked up a brochure that featured a roster of “digital employees,” complete with their names, headshots, and “personality scores.” I wondered what industries might hire them.
“They’re mostly for kiosks,” Keith responded with a tone of defeat. “Like at a mall or a museum. Also military training. Stuff like that.”
Keith directed my attention to the exterior of the cube. A large banner depicted an older male, prosaically handsome, with a square jaw, a custardy dollop of silver hair, and pale, limpid eyes. This was Chief, said Keith. “He’s a navy guy. And he talks like a navy guy. We work in forty different languages. So if you’re training someone in Ukraine how to operate an American tool, we have that language built in.”
Keith went back inside to rustle me up a T-shirt. He told me that the company was also breaking into health care — nursing homes, to be precise. Keith explained the vision. Your mom is old, and you’re constantly reminding her to take her medicine. Why not leave that to an avatar? The avatar can converse with your mom, keep her company, fill up the idle hours of the day. Plus, you can incorporate a retina scanner to check her blood pressure and a motion sensor to make sure she isn’t lying dead on the floor.
“Say there’s an elderly woman with dementia,” he said. “Her avatar will look like she did when she was younger. So she has someone to identify with. Does that make sense?”
I imagined a future geriatric Keith, lying in a nursing home bed, conversing with his younger self. Would such an arrangement appeal to him?
“There’s not going to be a choice,” he said. “A lot of old people are going to be talking to avatars in ten years, and they won’t even know it. When I was touring facilities in San Francisco for people with dementia and stuff, those places are like insane asylums. But some patients still have some cognitive function, and that’s who the technology would be for. It’s definitely not going to apply to the guys that are comatose.”
We stood in silence for a moment, and he faced Chatty, who hovered before us, drifting in her strange, waking trance.
“I wish they could fix the internet,” said Keith. “I swear, she gets nasty. She like, looks at me bad.”
by Laura Preston, N+1 | Read more:
Image: Dana Lok, Typist. 2023. Miguel Abreu Gallery, New York.
