Job seekers and hiring managers are using AI, and nobody’s happy - The Boston Globe

Ultimately, he got a job at a local bike shop after walking in with his resume in hand. “An interview is a conversation, a back and forth,” he said. “It’s an opportunity for me as a job seeker to get to know the people I could be working with.”

With the labor market getting weaker, the hiring process has become the latest battle ground for AI infiltrating important parts of the economy. In what employers and applicants describe as an escalating “arms race,” AI apps are increasingly being used by both sides — with mixed results.

Employers and hiring firms are using AI to help write job postings, winnow resumes, find talent in their workforces, and even conduct screening interviews. More than four out of five employers, including virtually all large companies, already use AI in the hiring process, according to the US Equal Employment Opportunity Commission.

Job candidates, meanwhile, are using AI to help write resumes and cover letters customized to a particular posting, practice their pitches in mock interviews, and even assist them during interviews. A Gartner survey this year found 39 percent of candidates said they used AI, and 52 percent believed employers were using AI. (Most of the job seekers use AI for resumes and cover letters, though 29 percent said they used AI to answer assessment questions.)

Both sides have complaints.

“Candidates feel like they’re sending their resumes into a black hole,” said Alexa Marciano, managing director at design and marketing recruiting firm Syndicatebleu. “But on the flip side, you have candidates who are just mass applying to roles. ... And when you’re getting 10,000 applicants to a role, it’s impossible to go through each of them. So it’s just becoming a circular problem.”

At Bentley University in Waltham, Alyssa Hammond, director of undergraduate career development, said AI apps have been useful for student job seekers to find job postings that match their interests and to get started with first drafts of cover letters and applications. Apps can scour many thousands of postings across corporate and job-listing sites to look for openings that fit a student’s skills.

But Hammond has also seen the apps exaggerate or make up information. Asking ChatGPT to write a cover letter for a student who ran an Instagram account for a high school sports team, the bot wrote of “a proven track record of creating engaging social media content.”

“It can be really dangerous, because you don’t want to come off like you’ve been out there in the world working for years, and you’re only a sophomore,” Hammond said.

William Battelle hadn’t created a resume in about 30 years when he was laid off from his job in events and conferences at Worcester Polytechnic Institute this year. The 60-year-old, who lives in Leicester, realized he needed to get a bachelor’s degree. But to apply to such programs and get some credit for his work experience, he needed to create a specialized document called a prior learning assessment. Microsoft’s Copilot AI app helped him start writing the document with the proper structure, he said.

“It gave me a foundation I could bring to a career counselor,” he said. “That was really helpful.”

For employers, the problem of being flooded with job applications didn’t start with AI. The trend goes back over a decade when online listing sites like Careerbuilder added features to let users quickly submit their resumes, according to Harvard Business School professor Christopher Stanton, who studies the use of AI in the workplace.

Job seekers responded by adding more keywords to their applications. The latest challenge for hirers, Stanton said, is job seekers using apps such as ChatGPT to burnish their resumes. With more applicants submitting AI-polished resumes that describe their skills using the correct terminology for a particular job posting, employers’ AI-powered screening software has a tougher time.

Even humans can have a harder time picking out the best candidates now, according to a study by researchers at Columbia University and Yeshiva University in New York. In an ingeniously designed experiment, inexperienced applicants using ChatGPT were able to write cover letters that fooled experts more often than applicants who didn’t use AI.

Even though AI only improved pitches a small amount, that still puts pressure on all applicants to start using it, Yeshiva professor and co-author Pablo Hernández-Lagos said. “If everybody’s doing it and I refuse to do it, I’m going to be the one losing out,” he said.

Employers will need to conduct more interviews to separate the truly qualified from those who were just getting a boost from AI, Harvard’s Stanton said.

But interviewing thousands of candidates is impractical. That’s created demand for companies like Hirevue, a Utah-based company that made the automated software that bike mechanic Peregrine encountered.

The company trained its app on thousands of recordings of real interviews, chief data scientist Lindsey Zuloaga said.

“Assessments have been around a really long time, and hiring is a very old problem, but we’re using new technologies rather than just a pencil and paper,” she said.

In a typical interview from Hirevue and similar apps, an applicant is asked four to five questions prerecorded by an executive. Questions are often general, such as “tell me about yourself” or “tell me something about you that’s not on your resume.” The interview often includes a specific question or two, such as why an applicant wants to work at the hiring company. The applicant may have anywhere from 30 seconds to a few minutes to record an answer. Usually, the app offers a chance to rerecord an answer.

Hirevue has an AI feature that can summarize an applicant’s answers and their fit for a particular opening. Employers typically still have a human review the applicant recordings.

REI said human employees reviewed all of the recordings and it did not use any of Hirevue’s AI software, such as to analyze candidates’ answers. The company has since switched from Hirevue to ICMS, which has more features including a similar automated interview app, REI said.

Boston-based Bullhorn, which makes software for recruiters and staffing agencies, rolled out a comprehensive AI app in May. Dubbed Amplify, the app can do everything from sourcing and matching candidates for jobs to conducting screening interviews and scoring the responses.

Bullhorn chief executive Art Papas said his company’s app only records an applicant’s voice, not video. Questions can range from whether a candidate is authorized to work in the United States, to what software a candidate has used to manage projects.

The use of the technology will continue to evolve rapidly, he predicted. “The candidates are definitely going to use AI, too,” Papas said. “’Talk to my AI bot. It can answer most questions for me.’ So that’s coming.”

The use of AI in hiring raises the specter of automated bias, because some apps were trained on material that reflected human biases such as gender, race, or age discrimination. In a study by the University of Washington last year, AI models used to rank resumes for jobs appeared heavily biased against female and Black-associated names. And a class action lawsuit filed in February 2023 in federal court in California alleged that AI software from Workday was biased based on age, race, and disability. The company has denied the claims.

Hirevue and Bullhorn said their apps are audited by outside firms to ensure they are not biased. And ultimately the apps don’t make hiring decisions, the companies said.

Eric Chester, who sometimes speaks with a stutter, completed two interviews with AI apps, but it was during a live interview with people over Microsoft Teams last month that he had a problem.

The 36-year-old, who lives on Long Island, said he was speaking quickly to avoid getting stuck verbally as he discussed a business analyst opening at an investment firm.

Suddenly the interviewer confronted him, asking if he was using AI “to cheat on the interview,” he recalled.

He denied the accusation but felt it had “completely destroyed my focus.... I never heard back, of course.”

Aaron Pressman can be reached at aaron.pressman@globe.com. Follow him @ampressman.