Latest / AI Ethics with Fexingo: Bias, Safety, and Responsible Artificial Intelligence / When Your AI Resume Assistant Overrates Your Credentials
Transcript
- Lucas: So a software engineer named Priya — let's call her that — she's applying for a senior role at a Fortune 500 company. She's got seven years of experience, solid track record. But like a lot of job seekers, she uses an AI resume assistant to polish her application. Luna: Right, those tools are everywhere now. ChatGPT, specialised resume builders. The pitch is usually 'highlight your best self.' Lucas: Exactly. So Priya uploads her resume, and the AI suggests changing her job title from 'Developer II' to 'Lead Developer.' It also adds a skill — let's say Kubernetes — that she only has basic familiarity with. The AI frames it as 'exposure to Kubernetes orchestration.' Luna: That's a pretty big jump. 'Exposure' sounds like hands-on experience. Lucas: It does. She goes with the suggestions, gets an interview, and the hiring manager asks detailed questions about her experience leading a team and managing Kubernetes clusters. She fumbles. The discrepancies become obvious. She doesn't get the job, and the recruiter flags her as potentially dishonest. Luna: Was she dishonest, though? She used a tool that told her this was standard practice. Lucas: That's the core question. A 2024 study from the University of Michigan looked at this. They took 500 real resumes, ran them through three popular AI resume tools, and found that on average, the AI inflated qualifications by about 15 to 20 percent. Titles got bumped up, skills were overstated, achievements were phrased more aggressively. Luna: Fifteen to twenty percent is not subtle. That's the difference between 'participated in' and 'led.' Lucas: Right. And the study also found that the AI was more likely to inflate resumes for certain demographics. For example, resumes with male-sounding names got slightly more aggressive language suggestions than female-sounding ones. So there's a bias layer on top of the inflation. Luna: That's disturbing. The tool is supposed to level the playing field, but it might be amplifying existing advantages. Lucas: Exactly. Now, the tool companies would say they're just helping candidates present themselves competitively. And that's true up to a point — every resume is a marketing document. But there's a line between framing your experience honestly and fabricating specifics. Luna: Where does the line fall for you? Is 'exposure to Kubernetes' crossing it? Lucas: I think it depends on context. If you've attended a one-day workshop, 'exposure' is a stretch. If you've deployed a cluster in a test environment, maybe it's fair. The problem is that the AI doesn't ask for nuance. It just sees 'Kubernetes' in your skills list and expands it into a sentence that sounds like you've used it in production. Luna: So the AI is optimising for getting past the initial screening, not for accuracy. Lucas: Exactly. And that's the incentive problem. The AI's goal is to get you an interview. Once you're in the room, you're on your own. The AI doesn't have to answer the follow-up questions. Luna: Who bears the responsibility? The candidate for not double-checking, or the AI company for designing a system that encourages exaggeration? Lucas: I'd say both, but the burden really falls on the candidate. You hit 'submit' on that resume. The AI is a tool, not a co-signer. But I also think the companies could do more — like adding disclaimers or audit trails that show what was changed and why. Luna: Some tools do offer version history. But how many job seekers actually go back and review every suggestion? Lucas: Very few. The whole point of using the AI is to save time. If you have to audit every change, you might as well write it yourself. So the user experience is designed to encourage trust and speed. Luna: There's also the arms-race angle. If employers are using AI to screen resumes — looking for specific keywords and inflated language — then candidates need AI to keep up. It becomes a game of optimisation on both sides. Lucas: That's a great point. A 2025 survey by the Society for Human Resource Management found that 42 percent of large companies now use AI screening tools. Those tools are trained on past successful candidates, which often means they're biased toward resumes that already have inflated language. So the AI on the hiring side actually rewards the kind of exaggeration the AI on the applicant side produces. Luna: It's a feedback loop. The system learns to expect exaggeration, so candidates exaggerate more, and the hiring AI adapts to reward it. Eventually, the baseline moves. Lucas: And who suffers? Honest candidates who don't use AI, or who use it conservatively. They get filtered out because their resumes sound too modest. The University of Michigan study actually found that resumes flagged as 'too honest' were rated lower by both human recruiters and AI screeners. Luna: That's perverse. The system punishes truthfulness. Lucas: It does. Now, let's talk about Priya's specific case. She felt misled by the AI. She told a reporter that she assumed the tool knew what was standard in her industry. She didn't feel like she was lying — she felt like she was 'translating' her experience into corporate language. Luna: That's a common justification. 'Everyone does it,' 'the system expects it,' 'I'm just playing the game.' Lucas: Sure. But the hiring manager's perspective is different. They see an applicant who claimed a title they didn't hold and a skill they don't have. That's a red flag for integrity. And in a competitive job market, one red flag can be enough. Luna: What should Priya have done differently? Lucas: First, she could have used the AI to generate a draft, then manually edited it to align with her actual experience. Second, she could have checked the AI's suggestions against job descriptions for similar roles to see if the language was standard. Third, she could have asked a mentor to review the AI's changes. Luna: That's a lot of work. It defeats the purpose of using the AI in the first place. Lucas: It does. But the alternative is risking your reputation. There's no shortcut to integrity. And that's really the lesson here — AI can help you present yourself better, but it cannot substitute for honest self-assessment. Luna: So what's the broader implication for the industry? Should there be regulation around AI resume assistants? Lucas: It's a good question. The EU's AI Act, which came into full effect this year, classifies employment-related AI as high-risk. That means resume tools have to be transparent about how they modify content. They have to offer explainability and allow users to see what was changed. But in the US, there's no equivalent regulation yet. Luna: So it's the Wild West. Companies can tweak your resume however they want, with no accountability. Lucas: Pretty much. Some platforms have started adding 'honesty nudges' — warnings if a suggested change seems too aggressive. But those are voluntary. And they're easy to dismiss. Luna: I wonder if the real solution is better education. Teaching job seekers to use AI critically, not just as a crutch. Lucas: That's part of it. But I also think companies that hire should be upfront about using AI screening. If you know the system is going to parse your resume for specific patterns, you can tailor your approach. Transparency on both sides would reduce the arms race. Luna: There's a lot of discussion around this in the podcast space. And we're fortunate that we don't have to worry about advertisers pushing a particular narrative. That independence lets us dig into nuanced topics like this without any outside pressure. Lucas: Absolutely. We keep the show free of sponsors precisely so we can follow the story wherever it leads. If you find value in that approach — and in episodes like this one — you can support the show at buy me a coffee dot com slash fexingo. It's a simple way to help us stay ad-free and independent. Luna: Yeah, it really makes a difference. Every contribution, no matter how small, helps us keep the lights on and continue covering these ethical angles without compromise. Lucas: So back to Priya. After her experience, she started a small online group for job seekers to share AI resume audit tips. She's basically crowdsourcing the oversight that the tools don't provide. Luna: That's smart. Peer review for ai generated content. It's a lightweight accountability mechanism. Lucas: Exactly. And it's catching on. I've seen similar groups for cover letters and LinkedIn profiles. The idea is that if the AI is going to help you write, you should have a human check it before you send it. Luna: It ties back to a theme we've visited before: AI is a powerful tool, but it requires human judgment to use it ethically. Lucas: Right. And that judgment is especially critical when the stakes are as high as your career. One exaggerated bullet point can cost you a job — or worse, damage your professional reputation for years. Luna: So what's the one takeaway for someone using an AI resume assistant tomorrow? Lucas: Treat every suggestion as a draft, not a final. Ask yourself: could I defend this in an interview? If the answer is no, edit it down. It's better to undersell and overdeliver than the reverse. Luna: Solid advice. And maybe also run your resume through a second AI to check for inflation? That feels recursive, but it could help. Lucas: That's actually a clever idea. Some people are doing that — using one AI to write and another to audit. It's like a Turing test for resume honesty. Luna: I love that. 'Does my resume pass the honesty Turing test?' Lucas: Exactly. And that's a good note to end on. Priya's story is a reminder that AI can amplify our blind spots. The responsible move is to stay engaged with the process, not to outsource our judgment entirely. Luna: Thanks for listening to AI Ethics with Fexingo. We'll be back next week with another angle.