A Boise-area customer-service rep is training AI. Will it replace him — but use his voice?
A year ago, Bryan Conner started a customer service job. He works the night shift, taking remote calls for electric companies nationwide from his apartment in Nampa. But the recent introduction of artificial intelligence into his job has him questioning how long he’ll have one.
Several months ago, the company he works for implemented a generative AI engine to aid in answering calls.
At first, it wasn’t very smart, Conner said. It would listen in on calls, summarize the conversation afterward, and generate a script for Conner to use. Conner would validate its result — a step intended to help a machine learn enough to function in the real world.
But now, after months of learning from him and his colleagues, the system could almost be deployed on its own, Conner told the Idaho Statesman.
“I know that ultimately it is going to replace us,” he said.
Customer service agency implements AI engine
Conner, 45, works for World Connection, a “progressive, bilingual contact center” and business-process outsourcing company with a headquarters and a call center in Boise, where Adam Bentley is co-CEO, according to his LinkedIn profile. It has another headquarters in Guatemala and operations in Ecuador and El Salvador, according to the company’s website.
World Connection did not respond to a call, an email and a submission the Statesman made on its website’s contact form.
Conner said the AI engine, which was introduced with little explanation, is making his job harder. To help get the system up and running, Conner spends extra time revising the engine’s transcripts, but he’s only allotted 30 seconds to do so after each call. He said there are often small errors, where the AI either incorrectly transcribes a word or phrase or misspells a customer’s name.
“It’s challenging,” said Connor, whose shifts are from 10 p.m. to 6:30 a.m. Friday through Tuesday. “And if we don’t do it, we get in trouble.”
The company also cracked down on the length of calls, he said. To train the AI on the ideal call time, Conner is supposed to complete each call in four to five minutes. Conner said he was written up when he had a call that lasted three minutes and 56 seconds.
He said the company implemented a series of monthly contests and incentives push employees to get the AI up to speed even faster.
“I feel like we’re being manipulated into training it just so they can save money on labor,” Conner said.
He’s concerned that he’ll be laid off and forced to retrain in another industry. He’s also worried that his voice or likeness could be used without his consent.
“It constantly comments on all of my quality assessments with ‘Great tone,’” Conner said. He added that the scripts he is provided are often cheesy: “’How can I save your day?’ Or — ‘how can I make you smile today?’”
“Half of the time people call it’s the middle of the night, and they’re standing in a foot of sewage or water because their pipes are broken,” he said.
‘The way of the future’
Conner is no stranger to generative AI. For a decade ending in 2017, he worked on standardized testing at McGraw-Hill, one of the “big three” publishers of educational content. In the hand-scoring department, he said, he helped develop an auto-scoring AI engine for long-answer, handwritten responses to essay questions.
His team would score papers by hand and then run a range of examples and their scores through the engine to train it. After years of running other samples and tests, the system got to a point where it was “really effective,” Conner said.
Then a competitor bought the testing division he worked in, and he and his coworkers were laid off, Conner said.
“I kept reading stories about AI replacing jobs, and I breezed over those articles thinking it was in the future or that it wasn’t going to affect me,” he said. “And I don’t want to rail against the technology. I know this is the way of the future. But is there some sort of social responsibility with the implementation of these things?”
Researcher says white-collar jobs face greater risk
Morgan Frank, an assistant professor in the School of Computing and Information at the University of Pittsburgh, told the Statesman by phone that the sophistication of current AI models was “unimaginable” just 10 years ago.
His research is focused on trying to figure out which labor markets are most exposed to AI or machine learning, and how those labor markets will adapt to AI-related disruptions.
At first, the technology was seen as having the most potential to eliminate jobs or decrease wages for manual, routine occupations, such as robotics or manufacturing. Most white-collar jobs, such as academic research or coding, and jobs that have a social aspect, like customer service, were not thought to be at risk of automation, he said. But that might be changing.
Frank said that what’s “really interesting” about large language models like OpenAI’s popular chatbot, ChatGPT, is that they seem to ignore blue-collar jobs.
“Instead, the types of work ChatGPT can perform are squarely in the cognitive, nonroutine category,” Frank said. “Those are things that it can do convincingly that previous technologies could not do.”
Swapping out workers for AI
Frank said a recent study that looked at what types of work are exposed to automation from large language models wasn’t clear on whether exposure is good or bad for those workers.
The study, published by OpenAI in 2023, found that about 80% of the U.S. workforce could have at least 10% of their tasks affected by the introduction of GPTs, or generative pre-trained transformers, while nearly 20% could have at least 50% of their tasks impacted. The influence spans all wage levels, with higher-income jobs facing greater exposure, the study said.
In academic research, Frank said, AI can be used as a tool to smooth over first drafts or to help with computer code.
“For me, it’s a boost,” he said. “For customer service workers, at least to the extent that customers don’t realize they’re interacting with AI, that it might be a pretty good substitution. And we’re starting to see companies play around with it. But we’re still kind of in an experimental first wave.”
A playbook might be emerging.
In 2023, an India-based e-commerce platform, Dukaan, fired 27 customer service agents and replaced them with a chatbot, the Washington Post reported. But first, the company had used ChatGPT to improve its in-house customer service chatbot, training it primarily on the company’s help center content over the course of several months, according to the Post.
Companies could charge more to talk to a human
On the other hand, Frank said some customer service companies could use AI to take on more clients or offer varying tiers of service, where maybe the lowest cost of service involves speaking with a chatbot and, for a higher cost, customers could speak with a real person.
“You can imagine that the business model shifts in a way that elevates human customer service representatives to a different role in a higher tier of service,” he said. “There’s some research out there that suggests very clearly there’s a premium or a benefit to when customers call and at least believe they’re talking to a human.”
He said many people may have to retrain and transition to other industries because of AI. For people midway through their career, shifting to something else may prove more difficult.
“In extreme cases, really big shifts to skill demand can actually alter career pathways and change workers’ prospects for the future,” Frank said.
