This rebroadcast originally aired on May 23, 2025.
AI is playing a larger role in the job hunting and hiring process. It helps job seekers fine-tune their resumes and cover letters, and employers winnow down applicants. But the new technology isn’t getting it all right.
Guests
Hilke Schellman, investigative reporter and assistant professor of journalism at New York University. Author of “The Algorithm: How AI Decides Who Gets Hired, Monitored, Promoted, and Fired, And Why We Need To Fight Back.”
Also Featured
Kendiana Colin, student at The Ohio State University who experienced an A.I. bot interview her for a summer job in Columbus, Ohio.
Tyler Jensen, filmmaker and editor (Typical Films) looking for a job in New York City.
D. Mayfield Phillips, based in New York City, he’s worked in project management for 25 years and is looking for work.
Klaudia Kloc, CEO and co-founder of Vidoc Security Lab.
Transcript
Part I
KENDIANA COLIN: My name is Kendiana Colin. I live in Columbus, Ohio. I currently attend the Ohio State University. Majoring in English on the pre-law track, am expected to go to law school after I finish up my senior year, which is the year after this year.
MEGHNA CHAKRABARTI: So that means right now Kendiana is looking for a summer job, just something easy and entry level so she can earn some extra money this summer. A few weeks ago, she applied for a sales associate job at a gym called StretchLab, and someone texted her to schedule an interview.
COLIN: Then he sent a whole paragraph telling me that it was basically going to be an AI bot that’s going to interview me and make sure that I look professional and I prepare.
You know, just treat it as a real interview, as if you’re talking to a real person. Why couldn’t I just have talked to a real person? I don’t know.
“He sent a whole paragraph telling me that it was basically going to be an AI bot that’s going to interview me.”
Kendiana Colin
CHAKRABARTI: Kendiana says this was her first time hearing about AI conducting interviews, but she wasn’t opposed to the idea. She thought it’d be like talking to Siri on her phone.
But once the interview started, things went a little sideways.
COLIN: It was going okay. I guess for the first two questions, they weren’t deep questions like, what’s the meaning of life or something. It was just, you know, what’s your name, your work experience, and then after the third one, she just started saying vertical bar pilates over and over again.
AI BOT: Vertical bar pilates. Vertical bar pilates. Vertical bar pilates. Vertical bar pilates. Vertical bar pilates. Vertical bar pilates. Vertical bar pilates. (REPEATING)
COLIN: I just told her, “Hey, I don’t understand what you just said. Can you please just try to rephrase what you were trying to say?”
Because I thought, you know, it was just a regular bug technological issue. Like, we can move past this. You were just working. It was just very weird. It was very creepy. It wasn’t ’cause she was just saying it over and over again. It was the way that she was saying it. It was like mimicking human nature.
Like it was a tongue twister, or at one point she was laughing. I didn’t record that part. And then she was like taking deep breaths and all sorts of things. I was like, and I was like, “Yeah, no one’s going to believe me if I don’t record this.”
CHAKRABARTI: Kendiana right there with you with the “ah.” Well, she was told the interview would be sent to a manager for review, but it’s been a few weeks and Kendiana has not yet heard back from the company. She’s not too worried though. She says the AI interview was actually a red flag for her and she’s gonna look elsewhere for summer opportunities.
Now Kendiana’s experience isn’t uncommon. According to the World Economic Forum, 88% of companies already use AI for initial candidate screening. And Resume Builder, a company that helps workers in the job application process, it recently did a survey of some 900 firms, and it found that about 23% of those firms already use AI to conduct interviews like what Kendiana experienced.
Now on the other side of the job search, job hunters are also using AI to seek out companies that might be the right fit to craft resumes and cover letters, and specifically to find ways to shape those documents to satisfy what the AI on the other end might be looking for.
Now, each week we seem to discover new ways in which artificial intelligence is completely shaping how we live. And today we’re going to talk about how it’s changing, how we look for work, and how employers look for workers.
Hilke Schellman joins us to help us with that. She’s an investigative reporter and assistant professor of journalism at New York University. And she’s author of The Algorithm: How AI Decides Who Gets Hired, Monitored, Promoted, and Fired, And Why We Need To Fight Back.
Professor Schellman, welcome to On Point.
HILKE SCHELLMAN: Hello. How are you doing?
CHAKRABARTI: Good. I’m also going to assure you right now that I am not an AI interviewer. (LAUGHS)
SCHELLMAN: But you know what, those are actually things I’m thinking about as a, journalism professor, you know, how are we going to make sure that the person on Zoom, on a call is actually a person? And that’s exactly what’s happening in the job world as well.
CHAKRABARTI: Well, exactly. And honestly, I say that with more frequency these days on the show. Because one day they will just kind of broom me off the stage here and replace a lot of, we in the radio with AI. Some listeners might want that data come sooner rather than later, but let’s get to the job market.
We found the numbers to be a little bit squishy in terms of determining how prevalent AI is in the job search or job marketplace. How would you describe that?
SCHELLMAN: Yeah, I mean it’s hard to tally, right? Because, you know, no company has to tell a government regulator. We don’t have like official statistics or anything, but we have surveys, and we have companies that sort of self-disclose.
And I’ve talked to a lot of companies while I was writing the book. So I talked to all of the, you know, I think the first point that I think a lot of job seekers will sort of encounter AI, if they, you know, it’s not clear if they will sort of understand that they’re encountering AI, but if you go to any of the big job platforms — Indeed, LinkedIn, Monster, ZipRecruiter, you know, you have it — all of those companies use some form of AI, right?
“Indeed, LinkedIn, Monster, ZipRecruiter — all of those companies use some form of AI.”
Hilke Schellman
That doesn’t really mean anything. But that is a big encounter. And then we know that a lot of especially large companies, they use applicant tracking systems, which, used to be like, glorified spreadsheets where you would track like, okay, this person applied, we interviewed them, they’ve been rejected, they went to the next round.
And all of those, the biggest vendors of those also now have AI capabilities built in. We don’t exactly know, does company A turn them on and on? But we see AI being used there. We seeing it used in like video interviews, audio interviews. We see it used in assessments. There’s like games that we ask job seekers to play.
And then we also, companies also use it in AI background check. They can use AI to check your social media history, all of those things. So we see it all along the hiring funnel, do job seekers see it? Not necessarily. If your avatar starts talking about vertical bar pilates, something is up, right?
But if you send in your resume, what do you know what’s gonna happen to it? Like you wouldn’t know that AI is screening it.
CHAKRABARTI: Yeah. I just wanna get into the nitty gritty about how AI is being used at each one of the levels that you just described. Let’s just listen quickly to Elizabeth. She explains why employers might be looking to AI for help with the hiring process. She works at Yale as a career coach and also helps companies with recruitment. And she gave us an example of how using AI was very successful.
ELIZABETH: I just did a search for a startup, tiny startup series A, 17 people. They had a director of growth marketing open role. It was 1,400 applications for one job. So they have no manpower to do that.
So what we did is we used AI to develop a competency grid. We said, “Here are the five things we absolutely want. Evaluate all these resumes against these five.” We took that list from 500 or whatever it was to 80, and then I eyeballed the 80. After we were done with the process, I went back and looked at the rest because I wanted to make sure I hadn’t missed anyone that AI could have missed. And it actually did a pretty good job.
CHAKRABARTI: So Professor Schellman, AI, in any application, in any sector that we’ve thought about, one of the things it does brilliantly is exactly this. It takes a process that might have taken weeks and can cut it down to microseconds in the jobs recruitment space. That seems to be a very powerful and positive use for the technology.
SCHELLMAN: Yeah, it makes hiring much more efficient. And it will analyze everyone’s resume and it doesn’t have the problem of humans, right? That we are like, if I’m hungry, I’m more grouchy. And maybe I look at candidates more negatively.
It’s a vast promise that AI vendors come out, that they say, “This is like gonna be more efficient. You can cut labor costs.” And we’ve seen some companies have cut down their HR departments or the talent acquisitions departments. “It will be without bias, and it will find the most qualified candidates,” and it’s true, it makes it more efficient, saves a lot of money. You hire people faster.
But we really haven’t seen a lot of evidence that the tools find the most qualified candidates and that there is no bias. In fact, when I started looking into it, I found a lot of bias in some of these systems.
CHAKRABARTI: We’re gonna talk about that in detail because it comes up over and over again regarding artificial intelligence models. But in terms of finding the most qualified candidates, you said that there isn’t much evidence that it’s successful in doing that.
SCHELLMAN: Yeah, exactly. And I think job seekers have intuitively thought that, right?
That they felt like, “Wait, I’ve applied to these jobs where I was like, really overqualified or I was really qualified. Why didn’t I get a callback?” But just a hunch, right? But we know from a survey from Harvard Business School, professor there, Joe Fuller asked over 2000 C-suite leaders in Germany, the U.S. and the UK.
And when those companies use AI tools, they ask, does your system reject qualified candidates? And almost 90% said yes. The leaders themselves, leadership and companies knows, they know themselves that their AI tools do reject qualified candidates.
“Leadership and companies know … that their AI tools reject qualified candidates.”
Hilke Schellman
CHAKRABARTI: Did you ask them about the sort of give and take between the efficiencies they gain, which ultimately converts into money, right?
SCHELLMAN: Yeah, totally.
CHAKRABARTI: Versus maybe missing some qualified candidates.
SCHELLMAN: Exactly. Exactly. It still amplifies the efficiency so much more for companies and you have to, like, this one little startup, right, gets 1,400 applications. But we see, when I talked to Google a few years back, they said they get over 3 million applications. IBM gets over 5 million, that’s a few years back.
So with generative AI, we see the volume has gone up at least 50% or so. So you can assume that for every open job they get hundreds of thousands of applications sometimes. Even Goldman Sachs said for their summer internship program, they got over 220,000 applications one summer. The scale is just so much that the companies feel like, if some qualified candidates are being rejected, it sucks, but we just can’t have humans go through all of this. That trumps everything.
CHAKRABARTI: Here’s another place in which AI is meant to help things get more efficient, but it’s turning into an arms race. Because as you said, with generative AI, workers are just like, “I’m gonna apply to 150 different places.”
SCHELLMAN: Yeah, exactly. And it’s funny that we hear job seekers online, a couple years ago they were really frustrated because they felt like, “Oh, we’re sending all these applications into black hole and never hear back.” And I think they feel a little bit empowered by generative AI now that they can craft and generate cover letters and resumes and people joke about it. “It’s AI versus AI. May the best AI win.”
Because everyone knows that companies also use AI, especially in the early stages of hiring, right? We see a lot of companies use AI for rejections, and when there is a small candidate pool, when you have like maybe 10, 20, 30 people left, you bring in humans to do the hiring, right? That hasn’t still gone away, but the rejection we have often outsourced to software.
Part II
CHAKRABARTI: We heard from a lot of listeners with their own AI job hiring or job search stories. Actually, one of them is a person we reached out to. This is Mayfield Phillips in New York, and he’s worked in project management for 25 years and has also been out of work for the last few months. And he’s found the whole process of applying for jobs with AI to be infuriating.
MAYFIELD PHILLIPS: You’re interacting with these applicant tracking systems. I just encountered one earlier that it took my resume at the very top of the process, populated my name, that’s all. Then there’s all of this list of experience it wants me to fill in. I just closed the application.
It’s, to me, it’s disrespectful of your talent to put all of that burden. You ask for a document that has all the information, use it. Don’t expect us to regurgitate it to you. That’s an indicator of what professional life would be there.
CHAKRABARTI: Here’s Tyler Jensen. He’s a filmmaker and editor also in New York, and for years he’s been able to make ends meet through freelance gigs, but COVID really changed all that. Now he is looking for more regular work, and he had one recent AI interview with a bot called Robyn, which he found very disconcerting.
TYLER: There’d be long pauses and I’m like, “Oh, I probably should say more stuff,” but I don’t understand how this is being used against me. For me, if it’s truly AI and they’re just gonna take my responses and transcribe them to text, doesn’t really matter how awkward my interactions are. And yet I don’t have eye contact on another person to let them know that this is weird for me.
So I don’t know how my, I don’t know. It’s one of those strange things that I hope not to get better at.
CHAKRABARTI: Here’s another one. This is Shree from Loganville, Georgia. She left us this message about how she’s been using AI to help with job screening.
SHREE: I have used AI to help me target companies in my area to apply to.
I’ve used AI to write my cover letters, I believe cover letters to be outdated, but some companies still require them. So AI looks at my resume and crafts a very well-written professional cover letter. And then once I’ve been able to secure the interview, I’ve used AI to prep, to compare my resume to the job description to determine what types of questions I need to be prepared to answer.
“AI looks at my resume and crafts a very well-written professional cover letter.”
Shree in Loganville, Ga.
CHAKRABARTI: So that’s Shree from Loganville, Georgia. Professor, your book goes through chapter by chapter in detail about what happens at every level of the job search, right? And I’d love to bring some of the riches of that detail and those stories into the show here. So let’s start with just when you submit a resume to a place like LinkedIn, like specifically, what happens?
SCHELLMAN: We see two technologies and I think may feel had an interesting experience there as well, right? That they were asked like, “Hey we put in this your name and your address. Can you add all of the other information in fields in our spreadsheet?” And I think that’s actually happening a lot, that when I talked to the CEO of ZipRecruiter, he said that they checked all of their resume parsers on the market a few years ago, and they found that about 50%, 60% of the time the fields, like your work experience get copied into the wrong field of the spreadsheet.
So a lot of information can get lost. So basic the problems here. So we often, I often, recommend to job seekers to use AI tools like Jobscan and others that actually help you upload the job description, you upload your resume and it tells you, “Oh, like most AI tools will think that you have 80%, 90% or so overlap with the job and that’s what you want to push for.”
CHAKRABARTI: Yeah. Hang on here for hundred percent. Yes. Jobscan. Yes.
SCHELLMAN: Yes.
CHAKRABARTI: Okay. Okay.
SCHELLMAN: And there are a couple other tools like that I’m not endorsing that can help you, but no, they’re not endorsing it. But I’m saying that it is helpful because I’ve had people who have been really frustrated applying, they couldn’t figure it out why their resume doesn’t come through.
They use Jobscan and it helps you find the right key words to put on your resume and think about transferable skills and that ups your percentage and then you can get through easier. So we have to like really rethink the way we craft our resumes. Generative AI can help, but also like just thinking through like, how can I make it more machine-readable? All of those things.
CHAKRABARTI: No. The reason why I wanted to hear you say the word Jobscan again, is that I know that in listening to this hour, a lot of folks are gonna be taking notes, right? On how to help themselves in their job search.
So again, not endorsing Jobscan, but that’s a tool that I’m actually, I’m writing it down too, just in case I need it in the future that you suggested that people use. Okay. So again, this resume screening though, what happens, I guess it depends on the company, but for some of these big firms, what do they do with that data?
SCHELLMAN: There’s two way and one is where you have AI, look at the job description and just check how much overlap is there to all the resumes. That is actually not very efficient because job seekers and generative AI, whatever they use, they’re pretty genius.
You know that the skills in the job description, you probably have to have them on the resume. So it actually doesn’t screen out a lot of people. So it’s very inefficient. So what many companies do, they take people who are currently in the job. So if you are looking to hire an accountant and you already have 50 accountants working for you, you assume, “The 50 accountants that work for me, they’re successful.
They made it through all of the trials and tribulations of hiring. I’m going to use their resumes, feed them into the AI tool and tell it, ‘Figure out what these 50 accountants have in common and hire those people. Put the people, the resumes that I feed into you, tell me which one are similar.'”
And that’s where actually when I talk to employment lawyers and other folks who get called in, when AI vendors pitch their software to companies or companies use the software and they feel like, “I don’t know if this is working.” They bring in outside counsel and they found that sometimes, or actually more than sometimes, some of these tools like predict on weird things.
For example, one time in one of the resume screeners, it had learned that the name Thomas was predictive of success at the company. So if you have the word Thomas on your resume, you got more points. And it probably points to, probably the AI, it’s going to sound weird did what it does best, right?
“[The AI] had learned that the name Thomas was predictive of success at the company. So if you have the word Thomas on your resume, you got more points.”
Hilke Schellman
It got a pile of resumes; it did a statistical analysis looking for patterns. And maybe this was for like a tech arm or for a company where you often have more men working than women, and maybe there were more Thomases on the pile, and so that became a predictor, but obviously it’s not meaningful, right?
It’s just arbitrary. It’s not meaningful for —
CHAKRABARTI: Can I just, pointing here, point us to your capabilities?
SCHELLMAN: Yeah, of course.
CHAKRABARTI: Because this gets very creepy. Because even just at the level of resume screening, in your book you write about how preferences are given to those patterns, but they’re also downgrading other words that might appear in the resume. Like you say in some cases resumes were downgrading if the word women appeared in there. Like you have the example of, “I belong to women’s chess club or women’s soccer team.”
SCHELLMAN: And we don’t know anything about it.
So that could be gender discrimination, right? Because probably more women have those words on their resumes, and I think that’s where some of the problems with these tools come in because the scale is just unprecedented. If you use an AI tool for all incoming resumes to accompany, you could discriminate against hundreds of thousands of people versus one hiring manager.
I also found out that in one case, if you had the word baseball on your resume, you got more points. If you had the word softball on there, you got fewer points. Again, pointing to gender discrimination. Another tool used the word Syria and Canada as predictors of success. That could be discrimination based on national origin, right? Because you are favoring those and disfavor others and there can be all kinds of bias creeping in.
And I think what’s interesting here is like the companies themselves didn’t find this out, right? They found this out when they brought in counsel from the outside. So this isn’t routine that this comes up, but it’s actually routine and built into these tools that we are not doing in a good job supervising these tools.
CHAKRABARTI: Let’s hear from Camille. She’s in Santa Barbara, California. She reached out to us with this story. She works in biotech. She’s been out of work since February, and she says the AI-powered applicant tracking systems, or ATS, that she’s encountered are very frustrating. And to your point, professor, she believes they’re discriminatory.
CAMILLE: Not surprisingly, ATS software is designed to weed out as many applicants as possible. Since employers and hirers are flooded with applications and they can’t possibly review them all. So because of this, they find some arbitrary reasons to weed you out.
For example, not only might you use too few keywords, but too many, and you’ll be filtered out. Worse, they’re almost certainly guilty of illegal discrimination. For example, instead of age discrimination, the system will instead deem you overqualified. I have a PhD in biochemistry, but I’m also 49. I’m certain that ageism is a factor. So now I’m playing a hide and seek with dates as I’ve been coached to do and avoid the overqualified filter.
You can also get blacklisted because ATS’s between companies communicate with each other. So say I got rejected from Amgen in March. Now, Bio-Rad won’t even look at an application from me until September or even later.
CHAKRABARTI: Professor Schellman, I keep thinking about this again, from the job seeker’s point of view and the employer’s point of view. Because it actually makes sense to me that a company would say, “These are the types of folks that we found are successful in this job. And so we are looking for people with these attributes.” Totally makes sense.
But at the same time, doesn’t it reduce the chances that they’re going to find people that may have some additional skills that they didn’t actually think that were immediately relevant to the job, but if a human were looking through them, they’d say, “We’re working on somehow trying to reach out to certain groups with this product, et cetera, maybe we want someone that has connections with that group?”
SCHELLMAN: Exactly. And we know that more diverse teams actually are better for business. So hiring the same people that you already have, that you often got into the job with human bias attached, right?
There is a reason why women and people with disabilities are underrepresented at the higher echelon of the workforce. And that’s partly due to human bias, right? So we have, you don’t wanna hire the same people again and again. And you staying with the skillset you currently have and capabilities where you maybe look towards the future and think, “What I really need is people who are go-getters, who are creative,” and you might not have that in the job.
And we see this again and again, that if you hire, a lot of hiring managers just pull out the old job description and just add three more skills. And so you have this like long, laundry list of like things where you really feel like, “Is that really what we need?” And I think the problem also by looking at people who are in the job today, is you really should only be looking at skills and capabilities, but we feed into these systems like the whole resumes.
A company, one of the largest venues used to use emotion expression analysis on people’s videos. We still see that sometimes pop up, where I’m like, that has nothing to do with the job. And you just — you might be discriminating against people who smile less or have different facial expressions. Don’t do that.
You veer into territory. We really should only be hiring looking at skills and capabilities. Not the way you look, not the way you talk. Those things should not be part of the hiring decision. But AI is bringing them back.
“We really should only be hiring looking at skills and capabilities. Not the way you look. Not the way you talk. Those things should not be part of the hiring decision. But AI is bringing them back.”
Hilke Schellman
CHAKRABARTI: So this gets us straight to the interviews and use of AI in interviews.
This is Aaron. He called us from Walnut Creek, California. And he’s encountered AI bots in interviews.
AARON: I am six months into a job search, trying to transition careers at a particularly bad time to do it. And I’ve had the interviews with AI bots asking me questions over the phone. I’ve had weird avatars asking me questions in a interview as well.
They ask good questions. You’re certainly talking to something that can think and carry on a conversation with you. But it is uncanny — and I am still unemployed.
CHAKRABARTI: Professor Schellman, we have to talk more about interviews here. Because off the bat in your book, you tell this very compelling story about technology that you saw in 2018, which is like a zillion years ago in AI.
SCHELLMAN: For AI. (LAUGHS)
CHAKRABARTI: Can you please tell us the story?
SCHELLMAN: Yeah. I first encountered this technology actually totally at random. I was at a conference in D.C., I needed to get a ride to the train station to come back to New York. And I called the rideshare, called the Lyft, went into the backseat and I asked the driver how he is been doing and he’s, “I’m having a weird day.”
And I was like, “Oh, really? No one has ever said that to me.” And he said, “I had a job interview with a robot.” This was in late 2017. I’ve never heard of job robots. And he had applied for a baggage handler position, and he had gotten a phone call, but he thought he was talking to a robot.
So sort of history repeats itself within eight years. Now we are back to like, robot hiring. And I started digging and I went to a conference and one of the companies was like showing, demoing their tool. And you would see how these boxes would look at the facial expressions and would write out “brow furrowing,” this and this, And what your emotions were.
CHAKRABARTI: I’m gonna jump it here. I was just really thrown because the detail is incredible, right? Because again, this is 2018 and as you write, on the image of the person who’s being interviewed, it’s almost like how they make movies, right? There’s like little dots, as you said, all over the face, on the eyes, nose and breath.
SCHELLMAN: Exactly, you think it is in a movie. And it blow me away. Yeah. I thought it was like, there’s new things, like this is a new way of hiring. And when I started looking into it and talking to experts, I was like, “Oh, wait a second.”
There were like facial expression analysis. We don’t know what facial expressions you need to have in a job interview to be successful in the job. There’s no science, I was like, “Oh no.”
CHAKRABARTI: So if you raised your brow too much and got a big score on that, that would equate to, and because it gave some, it did emotion analysis, right?
SCHELLMAN: Who knows, who knows.
CHAKRABARTI: Disgust, joy?
SCHELLMAN: What it was predict upon and was actually calibrated by people who did the job interview before and are now in the job. So if I ask you, “What are your strength and weaknesses?” And you smile, and if I smile in the job interview, then I would presumably get a leg up. And we all know that’s totally arbitrary. It has nothing to do with the job, how good you’ll be at the job.
CHAKRABARTI: And there’s all sorts of cultural elements when it comes to expressions and body actions, there’s different —
SCHELLMAN: Men and women are different. Different, from different societies —
CHAKRABARTI: The same action can mean completely different things, yeah.
SCHELLMAN: Exactly. That’s why, I mean I am glad to say that, like one of the biggest vendors in the space here, they have abandoned that technology. It does feel a little bit like a whack-a-mole because then, two years later another small startup comes up with this idea ’cause they’re like, “Oh, we found this, like you can do this. And they have a problem with hiring. So we thought we’ll use emotion expression analysis.” And I was like, “No, we already disproved it. Please stop.”
But I hope that companies, we are much more critical towards AI than eight years ago when I started this research. And I think that is because we have learned this bias. There’s problems here. We can’t just take these systems and not supervise them.
Part III
CHAKRABARTI: We wanted to get the perspective of an employer. And so we spoke with Klaudia Kloc. She’s co-founder of the startup cybersecurity company, Vidoc Security Lab.
And granted, she’s a smaller company, not like some of the big corporations we’ve been talking about, but Klaudia recently had two applicants that were, in fact, actually AI. They weren’t human, they were fake, but they got pretty far in the company’s hiring process. And that is until Klaudia finally sensed that something wasn’t quite right.
KLAUDIA KLOC: This person was invited to the final stage of the interview with me. And then, also, something was very off. Because it was not only about the bad connection. This person image was not — was speaking something and was delayed between the image and the avatar and his voice. And also, I started asking questions about his previous employees. And his jobs.
I asked if he was working remotely. He didn’t know where the headquarters of the company he said he was working for. He also said that he lived in one of the European countries for a while and he wasn’t able to even say a word in the language.
CHAKRABARTI: After discovering that this person wasn’t really a person, but instead some kind of deepfake or avatar, Klaudia says her company has had to reexamine the way they hire people.
KLOC: So we changed totally after this accident. We changed the way we do the interviews with people. So first we take much closer look to their LinkedIn profile and CV. We invite people to onsite interviews at the end of the hiring process. And we try to ask relevant, culturally relevant questions in the very early stage of the hiring process.
So questions like, “What was your favorite cafe at the university?” To make sure that this person really knows what they’re talking about. It’s extremely hard to do it remotely because you have access to ChatGPT, now we can Google things really quickly. But it’s the best we can do.
CHAKRABARTI: Klaudia’s company has released an instructional eBook so that other companies can learn from their mistakes. She understands why large corporations might use AI as an initial screening tool, but they no longer use it for her company, because it just doesn’t make sense for them.
She’s wary though, and says there really is no replacement — at least not with the technology as it is — there’s no replacement for humans.
KLOC: You can do the initial screening, but to be honest, there is, on the technical level, we don’t have a tools that will give you 100% accuracy. You still need humans to double check after AI. And also, unfortunately, because you’re using automated process, you can hurt humans in a way because they might be a real human who submits the CV, but it isn’t in a format that AI knows and you get rejected by the machine, basically.
CHAKRABARTI: Professor Schellman, this gets us back to the question of does the AI actually help companies find the kind of workers they want? Are they generally happy with it? And do they even know how it works? Do the vendors even know how it works or what these things are screening for?
SCHELLMAN: Yeah. I think that is like where I think we need to do much more. Because I think what happens a lot that companies use generative AI or deep neural networks, and part of that is that even the people that built the systems don’t always know how the system works. And what we see is like a lot of AI vendors are startups, they have to bring products to market really quickly. So I think they all come with good intentions, but there isn’t a whole lot of testing, redteaming or whatever you call it, bias testing — takes a long time that is being done here.
“Even the people that built the systems don’t always know how the system works.”
Hilke Schellman
And then the employers that wanna use the software, they also don’t wanna start pilot testing for month, right? They just wanna use something out of the box. They wanna save money. They don’t wanna hire people to then supervise the system. And this is HR. It’s usually seen as a cost center in companies, right? It doesn’t generate money, so any cost saving you can do, you wanna buy these tools, so there isn’t a whole lot of checking of the tool.
So I think that also worries me. And the companies that use the software of talking to now so many chief people officers who told me, “Oh yeah, we use that game, or we use this tool, and we had the same questions as you raised. We found out it isn’t actually working for us, so we stopped using it.” That’s great.
But I was like, can you just publicly say that so that the next company learns? But there isn’t a whole lot of appetite for that because those companies are often terrified of class action lawsuits. They would come out and say, “Oh, we used the game and fewer and fewer women came through.” There might be hundreds or thousands or hundreds of thousands of women who might file a class action lawsuit against that company.
So we see there’s a whole lot of silence around these tools, and I think that’s not helping us to build better tools and know which ones work and which don’t, and that’s a problem. And now put on top of that, that LinkedIn job interviews, you can’t even tell if it’s a human or if it’s an avatar that’s speaking to you.
You have to bring in some of these old-school things that the CEO is bringing back of like doing in-person interviews because you can’t tell anymore. So we have AI versus AI. And we really don’t know who was winning here in the end, it seems like companies are complaining that they don’t find the right workers. Job seekers complain, and it takes hundreds and thousands of applications often to even get a job interview. So that’s not working. Like, it’s really fundamentally broken.
CHAKRABARTI: Let me ask you about specifically, there’s some of the sectors that you found that AI was being used in the job search. You said trucking is a big one and in the airlines as well, in certain places?
SCHELLMAN: Yeah, so we see it most often used in what the industry calls like high-volume, high turnover jobs. So we see it often in retail, in fast food. We also see it in trucking that AI is used a lot. Chatbots are used to just figure out, “Okay, do you have a trucking license? Yes/ no?” And then you go to the next round.
Delta Airlines at one point used AI video interviews to hire flight attendants. We’ve also seen Atlanta Public Schools used AI as part of their hiring process at one point to hire teachers. So we see it creeping up. We also see it a lot if you want to work in investment banking or banking, like you cannot really, especially in New York, pass any hiring. It’s all AI screens.
“Delta Airlines at one point used AI video interviews to hire flight attendants … Atlanta Public Schools used AI … to hire teachers.”
Hilke Schellman
And I’ve talked to folks who were in graduate programs in New York who were like amazing, qualified computer programmers, quants even, and they say I had 400 interviews this semester. I can’t even get an internship because it’s all full of AI screens.
CHAKRABARTI: Yeah. It suddenly occurred to me that — again, this is always a problem when it comes to data analysis — like what we might end up creating is a system in which when workers get better and better at gaming the AI system, like finding what keywords to use or getting AI to help them mold their cover letters, et cetera. What the tool ends up actually selecting for is people who are good at that. And not necessarily good at the job that they’re trying to be hired for. Do you see what I mean?
SCHELLMAN: Yeah, totally. And I think you find people who are good at gaming the system or figuring out the system to their advantage, right? That’s always been the case in hiring, but it feels like it’s now on steroids.
And I do worry for people who don’t have those skills, right? Like somebody previously said about they’re worried about age discrimination, that comes to mind. We don’t really have any data. We don’t know. We know of one case where the Equal Employment Opportunity Commission took action against a company because it did flat out discriminate. If you had your age on your resume and you were over 55 as a woman, you were thrown out and the other time, over 60. But we don’t know what’s happening inside of these systems. So I do worry a lot about that and I think that’s not fair to people.
And also, what all of this shows, that’s why I said we need to fight back, because this is a first generation of AI tools that’s coming out. Or maybe we are in the second generation, but it’s pretty early on. And I think what we have done is like, we automated bad processes; resumes are not very predictive of how successful you’d be in the job. And we know that people are very good at looking at the job description, putting the skills that are necessary into the resume.
Maybe a two-week Python class and you put in that your Python and somebody is a master developer, you can’t tell those differentiation. So like resumes are bad. The same as like with the job interviews, actually, they’re not very good either because some people are very confident and they sort of project competence with that, and then they start the job, you’re like, “Wow, they’re really not good at this. They’re just good talking about it.”
So we use these flawed systems. So I actually think we need to think about like, how can we hire more holistically? What other, can we actually do assessments that the highest prediction is actually, be no surprise to anyone, is for somebody to do the job to assess are they gonna be good at the job, right? That’s usually not doable for most companies. You’re not gonna hire a hundred people for three months and then lay off 99, right? But can we build like maybe virtual reality assessments or something that looks at the core competencies of the job and test for that?
But it’s really, it’s really hard to do that. It’s really hard to test for soft skills. Almost impossible, right? If you could crack that, it’s a billion-dollar assessment industry, you would hit — hit the jackpot. But we need to hire better. Some cynics, employment lawyers told me, what would be more fair? Just ran a random number generator. There’s actually more fair, everyone has the same chance to be chosen and it’s as good as any of these tools. Probably most companies don’t wanna do that.
CHAKRABARTI: (LAUGHS) Can you hold that for a second, professor? Because I just wanna, before we run out of time, I just wanna get the voices of our listeners in here once again, because they really supplied us with so many actual first-person experiences.
Here’s Kyler from Independence, Oregon.
KYLER: I found that when I am making a resume or a cover letter, it’s almost as bad enough that I have to copy and paste from the job description itself, or I’m not even considered for even an interview in some cases.
CHAKRABARTI: Alright. And here’s Jason from Salinas, California. He actually serves on multiple hiring committees for his college employer, and he’s noticed recently, as we’ve been talking about, an upward trend of applications that are crafted with AI. But Jason says they’re all pretty bad.
JASON: It sticks out like a sore thumb. It’s obvious that it was AI-generated to the point that it is almost like students of mine copy and pasting and getting caught in terms of doing plagiarism. And it gets candidates booted.
So it’s a word to the wise that if you’re going to use AI to try to get a job, use it in such a way to where you mask that it’s AI. Because it’s still not to the point where you can just give it a prompt, ask it, and it’s 100% going to come across as genuine, particularly in the job application process.
CHAKRABARTI: So Professor Schellman, I have to say it. I want to continue to remind myself of the advantages that AI can bring, right?
We talked about the volume problem and ideally one of the promises of AI is that if care is taken, it should reduce the biases that can come with human driven processes. Because no one’s ever said that a face-to-face interview between two people is free of bias. We have our own biases there.
But at the same time, like we’re still maybe in a couple of more generations we’ll get there. But what you were saying a little bit earlier just made me think that like ultimately the problem isn’t AI. The problem isn’t people. The problem is just like how hard it is to find the right people. It’s the hiring process in general.
SCHELLMAN: Exactly.
CHAKRABARTI: You solve that though, and you have a quadrillion dollar product.
SCHELLMAN: Totally. And I think it just, like the move to AI or to digitization just has shown us how flawed the system really was always. And we know that from the data, right?
“The move to AI or to digitization just has shown us how flawed the system really was always.”
Hilke Schellman
We know that I think within a year and a half, almost 50% of people quit. A lot of people complain it’s not the right hire, yada, yada, yada. It is just incredibly hard to hire the right people. And often it’s when they start the job and they’re in the job, you finally get to assess how good they are at the job, right?
It is incredibly difficult. And also there’s so many untrained folks, right? Lots of people have to hire, they’ve never been trained for it. So it’s really difficult.
CHAKRABARTI: Yeah. And my conclusion to that is maybe that’s the way it should be, right?
Maybe some processes shouldn’t be easy and for some processes, this Shangri-La of a frictionless process just simply will never happen. We’re talking about people at the end of the day. Maybe that’s how it should be, but we have only about a minute left, professor Schellman.
And once again, I’m thinking about all the people listening who are particularly still looking for jobs in this fraught new environment. What advice would you give them? Are there specific tools you would recommend?
SCHELLMAN: Yeah, first of all it’s like, I always tell every jobseeker.
It is not you. It is a numbers game, right? It’s not that you are not qualified, it’s just like everyone knows that, like I talked to the big job platforms, they know that it takes hundreds, thousands of applications for anyone to find a job. So it’s not, you just keep doing it. Chugging along. Look at all like the stuff that we see out there, like how to produce a machine-readable resume. So I think, like old school, we would tell people like try to stand out, that’s like a human thing to say, have two columns, colors, images, none of that, machines can’t read that. Have short, quantifiable sentences. All of that. Find me on LinkedIn.
I have a lot of tips on there that I think works for most of the systems, right? I don’t actually know which system works every time, but we know a little bit how things work. But don’t give up. We will have to change the system eventually, so —
The first draft of this transcript was created by Descript, an AI transcription tool. An On Point producer then thoroughly reviewed, corrected, and reformatted the transcript before publication. The use of this AI tool creates the capacity to provide these transcripts.