Why AI can stop you getting your dream job
And what you can do to work around it
For the latest Explainable, I spoke with Hilke Schellmann. Hilke is an Emmy-award-winning journalist and the author of The Algorithm, a book on the influence of AI-powered tools on recruitment.
EXPLAINABLE: Can you tell me a little about the genesis of the book?
HILKE SCHELLMANN: The very basic idea started with a Lyft ride in 2017. I chatted with a driver, like, ‘how was your day?’ And he said, ‘you know, it's a really weird day, I had a job interview with a robot. I got a call, and this robot voice asked me three questions’.
He had applied for a baggage handle position at a local airport. Robot job interviews? I had never heard of this. So I started looking into it, and I was pretty floored by how ubiquitous it is, at least in the United States, for what they call high volume, high turnover. So a lot of retail fast food jobs, a lot of hourly positions. There are a lot of AI tools that are being used, from chatbots to video interviews. So that was the first encounter. I think this seems like such a profound impact in the way we hire, and there aren't a whole lot of journalists documenting this change.
EXPLAINABLE: It feels like there are a lot of bad use cases for AI in hiring, which we will get to! But were there any AI solutions that didn't have an obvious downside?
HILKE: Let me preface with this: I think the use of AI is very complex in HR. So we see companies are moving towards using more skills-based hiring. So instead of looking at specific universities, because that can tell you more about somebody's socioeconomic background, they also try to look more at skills. So that feels like a pretty good use case to look at just skills versus where you went to school.
AI is used for sort of large planning to understand what skills do we have across our hundreds of thousands of employees. And I think that's a great thing to assess and sort of build out ‘what do we need in the future?’ And that was almost impossible or would just take so many man hours by folks to do. And AI can really help with that.
EXPLAINABLE: And what stood out to you as the most egregious examples of misuse or unintended consequences?
HILKE: One thing that stood out is when companies use emotion recognition in facial expressions. So to infer if the person is looking happy. And I was always struggling with that. What do my facial expressions in a job interview have to do with my success at work? There's no sound science, there's no scientist that can tell you what facial expressions in a job interview would make you a good employee.
And the tool infers that I'm happy, but that's not always true, right? Like, I've been in many job interviews where I fake smile and try to be happy even though I'm really nervous and I'm not happy at all. So I think people can get seriously harmed and they won’t get their dream job because of a wrong facial expression, and we actually know it's not based on anything
Another thing is using tools without people's consent. So we have a lot of digital exhaust that folks say that we leave behind around the world or on the Internet, and they can harvest that, right? And use it. For example, a company that says they can find your personality in your Twitter or in your LinkedIn profile. So how dominant are you, or how conscientious are you? So I feel like there's a lot of consent missing.
Those kinds of tools should not be used in hiring. I feel like for me, we talk about high stakes in facial recognition, like, how long you're going to go to jail. And that's really important to talk about. But I also feel like if I get my dream job, that's high stakes for me.
I know we get rejected a lot in hiring, so we often feel like, ‘oh, you get rejected a lot. Who knows why?’ But I do feel it matters if I have my dream job. I would be really pissed if I didn't get my dream job, not because of merit, that's okay. If I didn't make the cut, that's fine. But if I didn't get it because of a faulty algorithm, I would be really pissed.
EXPLAINABLE: Within the HR industry, is there an acceptance that these tools are flawed or are they slightly in denial about that?
HILKE: I think both things are true. The vendors come out and AI is a big buzzword. And so a lot of CEOs tell their HR department, ‘We need to use AI’. So I think there's an enormous pressure on HR departments to have these innovative tools. And the vendors, they're promising three things: it will save you money, it will save you labor costs, and we'll find the most qualified people.
So we know it saves money, it saves a lot of labor. We actually haven't seen any results that point to finding the most qualified people. And there's been a survey done by Harvard Business School. Joseph Fuller is a professor there, and he surveyed over 2000 executives in the US, Germany and the UK. Eighty-eight percent said that they know that their HR tools are flawed, that they know that they're rejecting qualified candidates. So it seems like everyone knows it, just there maybe isn't a better alternative. And I do understand that companies are drowning in resumes.
They have too many applicants. Google gets three million applicants or around three million applicants a year. It's no human being who's going to go through all of those resumes, right? So they want the technological tool. Totally get it. So there is this push. We need technology to solve these problems for us. It's just the question, like, are we using the right technology? That I'm not convinced of.
EXPLAINABLE: When it comes to job applications there are these possibly urban legends within the job seekers markets. Like if you put certain keywords in your resume and make them invisible in your doc then they are still picked up by machines and get you to the top of the pile. Did you kind of come across any of that?
HILKE: A few years back I was at a conference where recruiters complained about that. They didn't like that when people did that, right? Because they put it in white. So a recruiter would look at the resume, wouldn't see it, but it would get them through the screen and they would be furious because they would then toss it on the no pile. So I think, yes, in the early days, maybe five, six years ago.Â
I now tell people to put their skills in a separate skills section and just bullet point it and have them very clearly on there so that a resume tool could just ingest them. Also don't have any kind of pictures, images, two columns. None of the old advice to stand out to a recruiter. No, don't stand out! Machines need to read the fields and put them in the right order. These are not all-knowing machines. So sometimes they put it in the wrong field. If you have two columns, the machine may go haywire.
So I always suggest you have the job description of your resume, put it in an online resume screener. So there are companies like Jobscan that will tell you there is a 60% overlap between the job description and your resume. And you want to aim for like 60% to 80% overlap, not 100%, because then your resume will probably be thrown out because the AI tool will infer that you just have copied the job description. So there are all kinds of things, like a super-easy word template and super crisp short sentences, clear metrics, and easily ingestible bytes for the machine is what gets you through
EXPLAINABLE: Do you think these issues are going to get resolved or are these problems exponential, a bad problem now could be a terrible problem in five years time?
HILKE: I think a bad problem is already an exponentially bad problem. The problem with an algorithm is its scope, right? So one human hiring manager can - and I'm sorry for the folks that have been discriminated against by one person - but there's only so many people that that one person can discriminate against. A resume screener that screens three million resumes... One of the lawyers that I spoke to, he found that one resume screener gave more points to folks who had the word ‘baseball’ on their resume and fewer points to people who had the word ‘softball’ on their resume, which in the US context is basically gender discrimination. Because men play baseball, women play softball. Not everyone puts this on their resume. But if you can imagine this over 3 million people, you would have discriminated against a lot of women. So the scope is just unprecedented.Â
I don't have much hope that the industry is going to fix itself, because the problem is it's not like a phone, where either your phone works or it doesn't. So the AI tools work. They run, they filter out people. So if you don't look too closely as an HR team it seems to be working. It ranks people and you call the 50 highest-ranked people. Great. So if you don't look too closely, it looks like the tool works.Â
Expert Interview is a monthly feature where I talk to people about their experience with AI. Their hopes, their frustrations, their tips, and their concerns. Want to talk about your use of AI for a future issue? Get in touch contact@explainable.online.
Expert Interview #4: The Designer
Expert Interview #3: The Creative Coder