AI avatars in the workplace: A tricky equation CIOs must prepare for

As generative AI capabilities expand, CIOs will soon have to make difficult decisions about how far to allow AI to represent company employees, whether in internal meetings or when meeting with customers or partners. 

Two events this week — an announcement from Asana Wednesday and an interview from Zoom’s CEO on Monday — outline a near future in which lines between human and AI collaborators become increasingly blurred.

Asana’s news release touted AI teammates that can “advise on priorities, power workflows, and even take action on work — all while adapting to the unique ways that individuals and teams work. With humans in the loop every step of the way, people have full transparency and control over how AI teammates support work.”

Two days earlier, The Verge published an extensive interview with Zoom founder and CEO Eric Yuan in which he spoke of future Zoom meetings where AI digital avatars will speak instead of employees. 

“Let’s say the team is waiting for the CEO to make a decision or maybe some meaningful conversation, my digital twin really can represent me and also can be part of the decision-making process,” Yuan said. “I should have my own LLM. All of us, we will have our own LLM. Essentially, that’s the foundation for the digital twin. Then I can count on my digital twin.”

Yuan also spoke of the ability to tweak the intellectual capabilities of the digital twin. 

“For that meeting I say, ‘Hey, tune that parameter to have better negotiation skills, send that version, and join,’” he said.

One frequent concern about an enterprise using such AI avatars is to make sure everyone in a meeting knows when they are interacting with the person and when they are interacting with the avatar. But Yuan said that he envisions a future where it can’t be easily differentiated.

“The experience down the road, that’s a 3D version of yourself that can mimic you very well, so you can’t know if it’s a real person or just a 3D version,” he said. 

Yuan also spoke of the ability of these avatars to mimic physical sensations. “Whenever you and I have a call down the road, it’ll feel like you and I are sitting together. I shake your hand, and you feel my hand. I give you a hug, and you feel my intimacy as well. Plus, even two people who speak a different language, the real-time translation will also work extremely well. And if you and I don’t want to meet, I send a digital version for myself, and you’ll have exactly the same conversation.”

He also, oddly, said the Internet was born in “1995, 1996” when, depending on how one defines the birth of the Internet, the most appropriate year might be when Arpanet switched to TCP/IP in 1983.

Guardrails and guidelines needed

Various AI specialists expressed concern about whether enterprises will place the proper limits on how far AI versions of people will be allowed to go.

“Enterprises should encourage AI to handle those repetitive, data-intensive tasks, freeing human employees to focus on strategic and creative endeavors. However, crucial decisions that require emotional intelligence, ethical judgment, and nuanced understanding must remain in human hands,” said Lars Nyman, CMO of CUDOCompute.com. “Colleagues often value the nuanced, empathetic touch that only a real person can provide. If and when the AI novelty wears off, we might find ourselves yearning for genuine conversations rather than the saccharine and over-caffeinated outputs from an AI.”

Nyman said that he was particularly concerned with any suggestion that the avatars might fool people into thinking they are human. 

“It needs to be abundantly clear when and why these collaborative avatars are at play. Not just with regards to risks at hand, but from a sheer user experience point of view. Real people tend to actually feel unimpressed or even frustrated when they discover they are interacting with AI instead of a human. AI should complement human efforts, not deceptively impersonate humans or replace active decision-making or human interactions,” Nyman said.

“The pessimistic take is that this presents a lose-lose proposition — build AIs that are too human-like and you risk perpetuating a deception. Build AIs that are clearly AIs and your workforce prefers its own kind,” he added.

Maxim Ivanov, CEO of Aimprosoft, also expressed concerns about how far avatars could go. Ivanov said he “attends meetings to discuss strategy issues, something I personally wouldn’t trust my virtual avatar to do. No virtual avatar can know what’s in my head, no matter how well trained.”

Another AI specialist is Aaron Painter, CEO of Nametag and former general manager at Microsoft China. He also said he was concerned about avatars that are designed to look and sound too much like their human counterpart.

“Visibility is critical: Humans should immediately know whether they’re interacting with an AI avatar or a real person. Deepfakes are an insidious threat because they break trust in online interactions. If someone meets with an AI avatar but thinks they’re talking with a real person, they may feel offended, frustrated, or misled — negatively impacting collaboration and joint productivity down the line,” Painter said. “The promise of AI avatars is to free us from meetings so we can spend time on more productive work. CIOs should consider how many employees will actually use AI avatars, and how people may feel if they expect to talk with a human and instead get an AI counterpart.”

Because of all the current attention on deepfakes, Painter believes AI avatars in general face an uphill battle. “CIOs will need to implement extensive training for employees before they will become accepted in the mainstream enterprise,” he said.

“CIOs need to have a strategy in place to limit how far AI goes in the company. But where the lines should be drawn is going to be up to the enterprise,” said Bob Rogers, the CEO of Oii.ai, a supply chain optimization firm. “They should be asking themselves what they want from their AI systems. Analyze large datasets? Absolutely, that’s what AI is good at. Replace employees with AI versions of themselves at meetings? Maybe weigh that decision a little more.”

Rogers said that he looked at what Asana is promising and thought it had value, but that CIOs must set strict limits. 

“Asana’s new AI teammates seem to be tackling the issue of repetitive mundane tasks that humans aren’t wired to do. Searching for the right person at the company who can answer a particular question is a task better suited for AI,” Rogers said. “But taking that answer and applying it to build a solution to a problem? That likely requires a creative human brain.”

Rogers said he was also concerned about the future picture Zoom’s CEO painted. 

“It’s a nice thought, but Yuan does describe decisions being made at these meetings. Organizations would need to tread carefully to limit the decision-making abilities of these digital twins. He mentions solving the hallucination problem, and I do agree that’s a key ingredient of making this process successful,” Rogers said. “I do wonder which companies will willingly pay for the digital twin technology and retain the same number of employees if people can just send digital twins to meetings for them.”