Transcripts are auto-generated and may contain errors.
0:00
Hey, I’m Rose.
And I’m Angie.
And this is Our Lives with Bots, the show where we ask important, timely questions about what it means to live with our bot counterparts.
And from time to time, we also dive deep into what an AI future might look like for us.
0:15
Sometimes we agree, sometimes we spiral, but we always go deep.
Hello lovely viewers and listeners.
Welcome to series 3 on AI and work and a is impact on our relationship to work.
So this is again our third series.
0:32
If you’re interested in watching our other two series, go ahead and check out our website or YouTube or Spotify.
But for today, we have an interviewee for our workplace series, so.
Today we have Hilke Schellmann.
She is a well well known investigative journalist and a professor at NYU, very well acclaimed and she’s just written a book.
0:53
But we’ll let her introduce all of that.
But what is she going to be talking to us about?
Well, I’ll set a little scene for you just to begin.
So all right, imagine you submit a job application and the recruiter reaches out to you for a phone call interview, which in this day and age is a miracle, right?
1:10
You’re getting a phone call from the recruiter.
Thank God.
Finally, one of my 3000 applications I’ve submitted in the past six months is getting a call back.
So finally an interview request and you pick up the phone, you call back, and the recruiter says hello.
And all of a sudden, this joyous feeling that you just had has sunk to the ground.
1:29
This person who just says hello is actually just it’s 1.
Of those tinny hello.
The tinny hello, the hello right like you get from a customer service call where you say representative press 0 a bunch of times representative.
1:45
Anyway, so that’s the scene for today.
So Hilke has done investigative journalism into AI used in hiring, recruiting and across HR and across the employee life cycle.
So the Algorithm is a book all about her experience doing investigative journalism for that topic.
2:04
So really, really excited to have her on.
Well, I.
Think it’s going to be really, really interesting because, you know, the last few episodes we’ve been talking about how AI is going to affect us at work.
You know, some people are concerned that AI is going to displace workers, and there’s obviously mixed views about whether that’s actually the case.
2:20
We’ve been talking about cognitive deskilling, so is it going to make us think less about work?
But actually this episode is really important because it encourages us to think a little bit further up the chain.
And actually it’s about what’s going on before we even get into having a job.
And to what extent is AI precluding people from getting into the workplace or kind of stopping a whole lot of people from getting into the workplace because of its bias?
2:44
And I think that’s really important because it’s important that people are aware of how it may help some people.
So it might help recruiters and and how could it be helpful or it’s important to understand how it is an algorithm and to what extent might it exclude certain groups from those jobs inadvertently without maybe hiring even being aware of it.
3:04
So I think it’s important kind of awareness creation today.
With that said, let’s bring on Hilke to see what she has to say given her expertise on this topic.
Welcome Hilke to our series on AI and work.
First of all, if you could just give a brief introduction about you and why you’re the perfect person to talk about this topic.
3:22
We’re really excited to have you on.
And your perspective.
Well, thank you for the celebratory introduction.
I’m going to leave it up to the audience to decide if I’m the perfect person or not.
But yeah, my name is, I’m Hilke Schellmann.
I’m an associate professor of journalism at NYU.
I’m also an investigative reporter.
3:38
And I want to say the last eight years or so I’ve spent investigating and looking at how AI is changing hiring and also the world of work.
And you recently wrote an amazing book called The Algorithm, which has been getting a lot of press and a lot of attention, as it should, because it’s all about how AI is impacting the workforce, specifically with its involvement in human resources or HR.
4:05
So what inspired you to write The Algorithm first of all?
You know, it’s kind of funny because a lot of people always ask me like, how do you get your ideas?
And you know, obviously it’s hard thing for journalists.
Like, how do you find the next big idea?
And one of the things is I usually walk around with a notebook and take notes when I see something like weird ads in the subway or somebody mentions something.
4:25
Now it’s all digital, but it used to be like in a notebook.
And the tip that ended up being the book actually came from a lift ride in Washington, DC.
So I was at a conference with lawyers, had nothing to do with AI.
It’s about consumer protection.
And I needed a ride to the train station to get back to New York.
4:41
And I called myself a lift and got in the backseat.
And I asked the driver how he’s doing.
And he told me that he had a weird day.
And, you know, the history of me taking lift rides.
No one has ever said that.
So I was like, curious.
I was like, oh, wait, what happened?
He’s like, you know, I just had a job interview with a robot.
4:58
And I was like, job interview with a robot.
So this was in late 2017, which is, you know, centuries in the in the age of AI.
So he said he had gotten a call from a quote UN quote robot that day asking him three questions about a job.
5:14
And he just was like very like literally weirded out.
So he did have a weird day.
So we just like chatted about that.
He’s like, yeah, it’s so weird.
And he told me he had applied for a baggage handler position at a local airport.
And this, you know, this was the answer that he got to his inquiry.
And I was like, Oh yeah.
5:29
Can I, you know, get your phone number?
Can I follow up with a question?
And he’s like, totally.
And, you know, I made a note of it and maybe, you know, like, kind of forgot about it for a little while.
But then a few weeks later, I wasn’t an AI conference.
That was at NYU.
And I went to a panel in the afternoon by someone, Kelly Trendel, who had just left the Equal Employment Opportunity Commission because of the first Trump administration.
5:51
And she was talking about that she can’t sleep at night because they did sort of like a day on back in the day.
We called the big data on big data used in employment.
And she found out that some companies use very basic algorithms to go through employees calendars and understand like how many hours there are at their desk.
6:10
Obviously a very flat way of making sense how many you know how productive you are at work.
But she was also worried that like people who have longer absences are folks with disabilities and usually parents with caregiving obligations, which are mostly mothers in this country.
And she was really worried that they could get penalized because they have higher rates of absentee with this kind of basic assumptions that companies were doing.
6:33
And we started talking about hiring.
She told me about companies and, and I called them up and I didn’t understand a word what the people were sort of telling me at the time, but I was really intrigued.
I was like, hey, for hiring, who knew?
And went to a conference and saw the first time one of the companies was giving a demo and they had like a video and you could see the squares that they have over people’s facial expressions and the frown and this and this.
6:57
And I was like, wow, this is like, I’m seeing the future unfolding in early 2018.
And it was already used by a lot of companies at the time.
I had no idea because I hadn’t been looking for a job for a very long time.
And I was fascinated.
I was like, I have to report on it.
And as I started reporting, I started to find out that maybe some of the stuff that companies put out there were less scientific or rigorous than we might thought.
7:21
And sort of, you know, as I was reporting on this for various outlets, I sort of felt like, wait, this is like a huge change and I need to write a book on this.
This is like a big deal and it needs sort of a big outlet for for lack of better.
Work, I mean, it’s just fascinating to hear and you know, there’s a lot that you work on in your book around the recruitment space and obviously we want to hear a lot more about that today.
7:44
But can you maybe start by just giving us maybe an overview of the big trend that you see in terms of how AI is being used across the whole employee life cycle in HR?
Just for our listeners to get a sense of just that breadth of how AI is being used.
It would be great to set the scene for us.
7:59
Oh, of course, yeah.
So AI is basically used anywhere in what you call the employee life cycle, right?
Like what we would maybe call from applying hiring to surveillance at work.
And also like thinking about succession planning all the way to firings.
8:15
We see AI being used as part of that process now.
So AI is touch points anywhere folks do work or apply for jobs.
Wow, so AI can hire, monitor and fire you across the board, no?
And also complain to your succession, right, like sort of for should we hire, you know, what is the future of our company and who do we need to hire?
8:40
AI can be also helpful in that.
So it’s sort of everywhere, although I have to say like it’s probably most prominent in hiring because you have high volume of applications.
But we don’t, We also do see it a lot of work, right?
Like the unfortunate thing is like it’s hard for me to give like sort of hard data because there is no agency or anywhere in the United States or in probably any country where companies have to register or tell anyone what they’re using.
9:06
So nobody voluntarily tells any kind of government agency that they’re surveying their employees at large.
You know, we know this from like surveys.
We know this from analysis like the New York Times, that analysis a couple years back and found that 8 out of 10, the largest employers in the US at least partially surveyed their workforce.
9:25
But we don’t have concrete so and so many percent of companies do XY and Z.
We just know it’s like super widespread.
And so since we know that AI is perhaps being used more in the hiring part of things, so perhaps we do a deep dive into the recruitment space.
9:42
And we’ve seen a lot of stories about the use and misuse of AI for recruitment on platforms, for example, like LinkedIn, Indeed.
And we want to hear your take on what is all of this about, what’s happening and how is AI actually being used in the recruitment phase of the employee life cycle?
10:01
Yeah.
So I think since the dawn of these job platforms that we call like LinkedIn, Indeed, Zip Recruiter, they’re you know, they’re a bunch of them.
They’re like about 25 ish years old right now.
We’ve seen sort of the the companies start out with like they’re going to democratize hiring.
10:17
They make it so easy to see open jobs and applying for those jobs like in the old you had to like, you know, first of all, find the open job description or even find the job description somewhere or buy like paper or, you know, different ways to to find these or get one of those weekly, alt weekly.
10:33
So there was a big deal to see their classifieds in their job section and you had to like write something and print it out and go to the post office, you know, involve kind of a lot of work.
And so it’s these job platforms that got easy, right?
There was a digital resume.
You would upload it or go to the company’s website and send it there.
10:50
So it made applying very easy for a lot of people and that was a great innovation.
I think what people maybe didn’t necessarily foresee that it made a lot of recruiters started to feel like their lives are getting harder and hiring managers that they felt like, oh, it’s easy for people to apply and more people are applying.
11:08
So they could see that for every job.
And obviously now that we have generative AI, we see those numbers just skyrocketing again.
So we see about like a number of like 50% or more applications per job because Gen.
Edge just makes it easy to generate resumes, cover letters.
11:25
Now they’re programs that run absolutely autonomously outside of the applicant.
So you can just lean back and have AI apply for you So you don’t even have to worry about anything, right.
And like, and on the other side, recruiters and HR managers have been complaining for the last 25 years or so now even more that they just get this deluge of resumes and they can’t cope with that.
11:46
And you know, these numbers are a couple years back.
So it hasn’t fully, they didn’t grasp all of their Gen.
I revolution.
But like even back in the day, Google said they get about 3 million resumes a year or applications.
IBM gets about 5 million.
I mean, it’s just like staggering numbers, right?
Easy numbers.
And you know, like a few years back, Goldman Sachs set for their summer internship alone for that program, they got over 220,000 applications.
12:10
I mean, it’s just a huge number of resumes that like obviously no human wants to go or can go through, right?
So like since the early 2000s or so, we’ve seen companies coming up with technological solutions to this and just saying, hey, like we can screen your resume.
And you know that at the beginning it was like very basic, it was sort of like just checking, here’s the job description, how many words do overlap?
12:32
And now we see more sophisticated ways that companies use it and all the job platforms.
So I think, you know, and I think people are aware of this now, right?
If you like use LinkedIn or the job platforms, all of them use some form of, of AI.
And I do know that because I talked to them.
But you know, to a variety degree, we don’t know exactly what’s in the black box, right?
12:50
What I think generally is happening, AI is being heavily used in the beginning stages of hiring where you have a lot and lot of vasiman, you have to do a lot of cutting or rejection.
So we use AI for rejection.
And once you have like the candidate pool, you know, you winnowed it down to like, I don’t know, 10/15/20 applicants or something.
13:09
Then usually humans get involved, right?
There’s still for most jobs, there’s still a job interview and you get to, you know, at the end after you maybe you’ve done a one way video interview, you still talk to human.
And so humans still make the decisions at the end for most jobs and they get involved at the end there is a decision to be made.
13:26
But the rejection, we have automated that.
We left that to software.
So how does that rejection work?
How are these algorithms implemented there?
What sort of data are they filtering through or using as filters to determine who gets to be in the 10 to 15 get interviewed stack versus those who get pushed into the trash bin?
13:46
How is that decision?
Yeah.
Even though you think about like, you know, it’s like sort of a, you have a lot of applicants at the beginning.
You know, I’m sure every Fortune 500 company at this point uses some sort of resume parser or Screener.
Looking at the resumes and either comparing it to the job description and understanding how many words overlap is here and then rejecting people based on that.
14:05
It’s pretty imprecise technology because a lot of, first of all, generative AI systems know that, but also applicants know that that like you want to have a lot of the keywords in the job applications in your resume.
So it’s like, you know, you don’t filter that many people out of that process with that.
So what other companies use is like they have people that they label as successful employees and that is often people who have made it through the hiring and now in the job.
14:30
It doesn’t necessarily mean that they’re high performers, but they’re in the job.
So they made it through the hiring process.
You take their resumes and feed them into an AI system and tell the AI system like what do these people have in common?
And like as you get in new resumes, either put them on the yes or no pile.
14:46
And we see again and again there have been problems with that kind of systems because AI does what it does best.
I’m hypothesizing is that it does a statistical analysis and you know, in some instances it found out that if you have the word Thomas on your resume, that is a predictor for success and you get more points.
15:04
And that’s probably true for that stack of resumes of people in that company that are deemed successful.
Maybe lots of Thomases were in that pile and it was statistically significant, right?
It’s just that an AI tool doesn’t have ethical reasoning, a moral compass, no social science, and knows that like a first name has no causality to how you do your work, right?
15:25
Like, just because your name Thomas doesn’t mean that you have a medical degree and can be a good doctor, right?
There’s no causal relationship.
Obviously, human hiring managers and others know that.
Like we really shouldn’t look at these personnel proxies or any kind of identifiable things.
It doesn’t mean that machines know that.
15:42
So we have sort of like problems coming in.
Like other things I found is that a tool used Africa and African American as the proxy for success if you have those words on your resume.
And I think that could easily be race discrimination.
Other things that I found is when you had the word baseball on your resume, you got more points.
15:58
If you had the word softball on your resume, you got fewer points.
That could be gender discrimination, right?
And there are laws around this.
You are not allowed to do this in the United States.
But because these tools are often black boxes, so even the developers who built the tools and their companies that use the tools don’t actually know how these tools work and make decisions.
16:16
Every time they make a decision, no one is the wiser, right?
Like I only know this because some companies do their due diligence and they bring in outside counsel to check these tools.
And that’s when they found some of these problems.
And actually way more than I had thought.
Like a couple of lawyers were like, we found problems and every tool that we looked at and now they said in every 4th tool.
16:34
So it’s way more widespread than we think it is.
Because I think what what happens at the at the root of it is that HR is usually seen as a cost center in companies, right?
It doesn’t generate revenue.
It does mainly profit.
It costs money, kind of have to have it, but you often don’t want to spend excessive amounts of money on it.
16:52
It is not a product.
It doesn’t make the company money.
So like the idea of like, oh, let’s buy technology and have fewer workers is pretty striking for a lot of people in HR or leaders in HR.
So I think, you know, they want to save money.
So they’re not necessarily now hiring people to oversee these algorithms that they bought, right?
17:10
So there isn’t an incentive to do that kind of mining, which you really should be doing.
You should be doing your due diligence because, you know, like we all have gotten rejected for a job that we have applied for that just like has always been part of that, right?
Like for most times in the last few decades, they’ve always been fewer jobs than applicants.
17:28
You do get rejected as part of it.
You know, I get nervous before job interviews because I do think they’re like, really, there’s, there’s a lot for me at stake.
There’s just like at stake for me, like, can I put a roof over my head?
Can I feed my children?
Can I feed myself?
Like, these are real high stakes decisions that are being made.
17:45
And also like, you know, a lot of us spend a lot of time at our jobs and there was like, your personal happiness is tied into that, right?
So it matters if we have a job, it matters if we have a good job.
So I do think these are high stakes decisions and we really should be more careful using machines to make those decisions.
18:02
I’m not actually against using machines because we haven’t even talked about this.
Like human bias is rampant in hiring and it’s really hard to get rid of.
So like, it’s not totally a bad idea to use machines.
It’s just a bad idea to use machines and then not monitor what these what the output is of these machines and not understanding that like you could really harm people.
18:22
And I think we hear a lot of recruiters and hiring managers saying like, oh, like we get overwhelmed with these application and there’s no good people in there.
There’s like we have a problem.
We have a shortage of people who could do these like very sophisticated jobs.
And it’s sort of like, Are you sure?
Because it might just be the algorithm, you know, and job applicants have the same complaint that they feel like, wait, I’m actually like a very good candidate here.
18:43
I have everything they need and I still get thrown out and I don’t know why.
I don’t know why they ghosting me.
I never hear from them.
I think both sides sort of operate in this black boxy environment and both don’t know why things are happening.
I mean, Hilke, it sounds like a really complex problem actually, because to your point, like IBM’s getting, what did you say, 5 million applicants a year?
19:04
I mean, you can’t go and hire a whole bunch more recruiters just to go through all these applications, right?
Like that just becomes an untenable solution.
So they need a solution, as you were saying.
And I think you said like you’re not against a technological solution, but this doesn’t seem to be a fair one.
19:20
And there’s lots of problems with it.
So I mean, how do the recruiters, when you talk to them, how do they feel about this solution?
Because I think, you know, they do want to do a good job, right?
Like they care.
My experience of recruiters is they they always care about candidates like they care about getting the right person for the role.
19:37
Even though as you say, they’re often incentivized by getting the role filled quickly, at their heart they really want to get the right candidate for the roles.
From your experience, how do they feel about these tools?
You know, I think they’re as frustrated as job candidates in the job.
They feel frustrated too, right?
They feel overwhelmed.
19:53
They’re they’re also like, you know, they are sometimes also angry because you know, it is true that also people apply who might not have the necessary qualifications, right?
Because it’s so easy to apply, you know, they feel like that wastes their.
Time they feel like the technology wastes their time, but also they don’t have the time to actually go through all of the myriad of resumes that they get.
20:11
And so I think there’s only very few recruiters, like I’ve talked to a couple who is so frustrated by the technology, but like, OK, it’s not working.
It’s not finding the right people.
If I go, if I do like just even go in and, and, and look at a bunch, I find some that are actually are qualified.
20:26
They just were in surface by the tool.
And some have said, you know what, I just like cancel everything on my calendar and like take a week off and just go through all the resumes.
I don’t think that’s doable for most people.
I think what it sort of points out to the heart of it is that like hiring for a long part, we just did it the way we did without like thinking a lot about this.
20:47
And I think I was hoping that like maybe now that we automate and AIFIA lot of these processes, people would start like questioning like, why are we doing this process, right?
Like it turns out that like resumes have a very low predictive validity for success in the job, right?
21:04
But we use them because we’ve used them for a very long time.
But really like, should we be using them?
And should we use like people like to use personality because their predictions for hiring, because there’s very little discrimination between different groups in that scenario.
But like that’s also not very predictive of success in the job.
21:21
Also our personality is very little to do with success in our job.
And I mean, also sometimes just like think about myself that I’m like, I actually happen to be kind of an introvert and really shy.
And I had to like train myself to like go up to people and talk to strangers and, you know, sort of realize my own limitations.
21:36
And we’re able to train myself and, and, and do things to overcome that.
But like on a personality test, I would show up as like very introvert when I really overcome some of those things.
And I think, you know, sort of a snapshot in time of your preferences.
And it doesn’t really account for all these other things.
And it’s just very low predictive value in general.
21:54
So I don’t think we should be using, but instead we’ve seen companies sort of AI fi now personality tests with fun little games and things that are really truly not working in hiring.
So I feel like, hey, can it be a solution that as we think about how we could use AI or any kind of algorithmic solutions, can we think about using better ways to predict job success, right?
22:15
Like we know from science that this will be surprised to no one that the best way to predict if you’re going to be good at the job is to have you do the job.
Not very helpful for most companies, right?
Like they’re not going to hire 100 people and have them in their office for two months and then let go of 99 people is not doable, not efficient.
22:34
But I do feel like can we do any kind of assessments where you really think about like, OK, what are the the real skills that you need for this job?
How can I measure them and measure them effectively?
And how can I do this somewhat efficiently and also give the candidate a sense of what the job is about, right?
22:51
Because they often also don’t know that and they get kind of angry because they’re all these AI.
So they’re very impersonal.
Often they don’t get a lot of information about the job either.
And they feel kind of not everyone, but most of them have told me they feel it’s kind of an inhumane process.
And maybe like a virtual reality testing of different skills or something would actually give you a sense, oh, that is what the job is about.
23:11
And do I want that?
You know, like there’s like information for both sides.
But we haven’t really seen a lot of that.
We’ve just seen sort of the efficient way, like cutting down on efficiency, cutting down on time to hire, which I personally think it’s just not a good criteria.
Like you should really find the right employee and give recruiters bonuses for that and not because you hired somebody in a week.
23:32
Sure speed is maybe 1 criteria, but really hiring costs a lot of money for most roles so you want the right person, not a person.
I really like the highlight that AI used in hiring is sort of illuminating all of the ways that hiring has already been biased and it needs a new solution, a systemic new solution.
23:53
So I’m curious about these proxies that are used for hiring that is suggestive of what was used before things like personality.
I know in your book you did some auditing of AI hiring tools like those used for interviews, right?
24:09
What was that experience like with your perception of OK personality is not not hitting?
Yeah, I mean, you know, it’s like wild and you know, and maybe I’m, I’m aging myself here that I just had never encountered any of these tools in hiring, right.
And when I started the the tool, I was already at NYU.
I’m still at NYU.
24:25
So really we haven’t done any job applications, but you know, I was curious.
I was talking to job applicants who, you know, there were a couple of outliers who said, you know, I actually kind of liked it because I could do with it like 3:00 AM in the morning.
Like, you know, you can do these one way video interviews when no one is on the other side.
24:40
And now you have actually Gen.
AI avatars sometimes that do the interview, but he can do them whenever you want to.
You want to do at 3:00 AM Sure, no problem.
You would be hard pressed to get a human recruiter to do an interview with you at that time.
But most people just felt like it was very inhumane and I was like, let me see.
So like said up in my little office and, you know, most companies give you, you know, they want to sell a product, so they give anyone a wee two week trial and you know, I used my real name and signed up for these and I was like, let’s try it.
25:08
And I did find it awfully weird to like stare into a computer and you sort of feel like for a job interview, you want to usually you want to like convey excitement that you like this company and you’re interested.
And you know, there’s like sort of a little video of like, Hey, this is company A we’re so excited you’re here.
25:24
Tell us where you want this job.
But then there’s no one else and you’re like, how am I going to get excited looking at this green dot?
And I don’t even know where to look and I’m looking at that.
Should I look here?
Should I look there?
Like it was just wild to me that experience and kind of lonely and but you know, I got used to it.
I did so many of them.
25:39
And also like, you know, I think my experience might be a little different than other job applicants because for me, there was nothing at stake, right?
I was testing them.
I was trying to understand like, how does the experience feel like?
But I didn’t feel like, oh man, I really need to get this job otherwise I can’t pay rent, right?
Like I want to make sure that people understand the very significant difference here, that I like tested them for research purposes.
25:58
And I could like sort of illuminate what is this experience like what happens in these job interviews.
But I also worry that, like, you know, a lot of tech companies sell these products saying, like, oh, this is like a great Democratizer.
It’s one-size-fits-all right, like one video to rule them all or one resume parser to evaluate them all.
26:16
And what I was thinking when I was doing these and sort of got it, maybe a bit more sense of like how they work on the back ends.
One of the things I thought about like in one way video interview use, it’s not the audio recording or the video recording that is being used to screen people.
It’s actually the audio is transcribed through a transcription service and no one ever pays any attention to that.
26:37
It’s sort of seen as the plumbing of our digital lives and no one’s questions that.
But, you know, I was wondering like, what is with people like me who have an accent, people who have a speech disability, like what a transcription tool be able to transcribe their words as well as people who have like African American or have a German accent like me or, you know, all kinds of things or have a speech disability.
27:00
So I was surprised when I talked to then the the chair of the Equal Employment Opportunity commissioner said, yeah, we are wondering that too.
Like, you know, somebody needs to really look at that.
And I was like, well, you do have a staff of over 2000.
Can you kind of look at that?
I think it’s important because we see these interviews being done on millions of people.
27:18
But I felt like, well, maybe I need to start looking into this.
So I did like, these, like, little tests myself.
And I talked to the companies and I was like, what happens, like if somebody has a speech disability or have a very strong accent, for lack of a better word?
And they said, oh, yeah, yeah, yeah.
The algorithms are like, built to like, find that.
They sort of talked about that there was a threshold that you have to overcome in your answers, and only then you would get a score if you’re under that threshold.
27:41
I never fully understood what that threshold is, but that’s what they told me.
You would get an error message and then a human recruiter would get in touch with you presumably.
So when I set this up and you know, I got my little trials, I call it maybe needling or pushing the algorithm a little bit, right?
So I was like, I’m not going to fake an accent and like or fake a speech disability.
27:59
That seems weird.
So I was like, I did a couple of things.
One time I just said I love teamwork 50,000 times everything.
I just the only four words I said I love teamwork.
And I guess it could be 3 words depending on if you write teamwork together or not.
And I got a pretty high score, so that was surprising.
28:15
I was like, wow, I didn’t even even say anything.
What are my strengths and weaknesses?
I love teamwork.
Well, teamwork, I love teamwork.
That’s crazy.
So I didn’t really say anything substantial.
I said something and I guess I got a score for that points.
And then I just thought like very much I did.
I did give it different intonations, but I guess that wasn’t, I don’t know if it took it into account or not.
28:33
I did get a personality score though too.
So it was able to find a personality out of all those different.
I love teamworks.
Very conscientious, I’m sure.
Very extroverted too.
Yeah, you know, didn’t testify I’m on time or anything, but she just loves teamwork, which probably in the history of job interviews, no one has ever said no that they didn’t like teamwork.
28:52
But in this case, so I thought like, OK, let me just read something in German.
So I vet the Wikipedia entry on psychometrics, which is sort of the difference between people and how to measure how maybe adapt you are for for a certain job.
So I vet that into the tool for all the answers.
And for sure, I thought I would never meet this threshold because I, I mean, I spoke German about nothing to do with the job, so surely I would get this error message.
29:17
But I got an e-mail and in the first time, I was so surprised I was 73% qualified for the job.
And I was like, whoa.
I didn’t even say anything of substance.
And in fact, in one of the tools, I got a transcription of the words that I said and it was like total gibberish.
29:33
It was like sociology does it iron nematode.
I mean, I didn’t even know what the words mean because obviously the tool hurt the German words and try to transcribe them into English and, you know, it’s total gibberish, makes no sense.
But I got a 73% score.
That means I was 73% qualified for the job.
29:50
I’m going to say that is pretty high.
And then, you know, I talked to the developer and and one of the developers was like, yeah, it’s because, you know, in this 5D space and German and English, they’re like together.
I was like, I don’t, I was like, I don’t understand what 5D.
Is that an explanation?
I don’t know So well, you probably know more, but.
30:08
You weren’t making, you weren’t speaking like sensible answers, just in a different language, which would.
Yeah.
No, nothing to do.
No, no, I mean, and so then I also did it with like, you know, I worked with graduate students at a time once her native language was Chinese, the other one was Vietnamese.
30:25
So we did the test in all kinds of languages that also got really high scores.
Same total gibberish transcript, right.
And I also like kept asking these chief technology officers that I’m like, OK, like, I’m sorry, I don’t understand the 5D and I sorry, it’s beyond my cognitive abilities.
But can you few in front of the judge?
30:42
And you would have to tell them like, why was Hilke rejected or got a 73% qualification score?
Like how did that come about?
Can you just explain that?
And I got the 5D answer and I was like, I don’t think that is so, you know, I learned over time.
30:57
It’s like also how to query and answer question developers here like because I do think like if we use AI in high stakes decision making, we should be able to explain what these tools do and they should be a human who knows what these tools do.
Were you ever rejected so did you ever not make the threshold?
31:14
Because it would be great to know what under what circumstances.
I never not made the threshold with any kinds of tests.
I can’t tell you when you made the threshold.
Apparently maybe there is no threshold.
It turns out one of the companies actually told me they’re like, Oh yeah, like a six out of 10 is actually not high.
We had to like scrap one and two because people were complaining.
31:32
So like the the the lowest you could get was actually higher than you think.
I was like, I’m now I’m like really confused how all this stuff works.
It’s interesting.
That there’s a lack of clear explanation of how these AI tools work in hiring when at the same time it’s difficult to ask someone perhaps why it is that they hired a certain person.
31:53
What is it about this person that let you to say, yes, this person is getting hired versus this person, we’re not going to be moving forward in the process.
I think there’s also some lack of explanation there, right, Of people using.
Oh yeah, there’s humans to say.
Oh, it just this person felt right.
32:10
Is the true reasoning versus filling in kind of, oh, because they hit this threshold, et cetera, et cetera.
I think that there’s a little bit of fuzziness on both ends.
So yeah, Angie, what do you think, especially since you’ve been hiring a lot?
Yeah.
And I think that that’s one of the things that kind of occurs to me and what I’d love to get your view on Hilco is why these companies are still using these tools.
32:30
As you said, they’ve been using them for a very long time and they’re not cheap or they’re cheaper than hiring recruiters perhaps.
But you know, what is their deemed return on investment.
But one of the things that you’re saying when you’re talking about the video recruitment is I can imagine one of the problems to solve is that actually one of the biggest issues is that we tend to recruit people like us.
32:50
When I say we, obviously if you’re aware of that, then hopefully that’s half the problem.
But often hiring managers will be like, this person was amazing.
They like football, I love football.
We went to the same school, we both love pizza.
And of course, all they’re doing is they’re hiring a best friend for the weekend, right?
33:07
And so that’s one of the hardest things is trying to get people to hire for diversity in terms of thinking and all sorts of things.
And so I can imagine one of the greatest appeals for a technology tool is to try and get away from that bias.
But like you say, we’re building it right back in.
33:24
I think that is the appeal and I think that’s sort of certainly the marketing language of these companies.
And you know, I was maybe also naive in 2017 and 2018 because I was like, wow, who knew?
Like there’s an objective way to like find this way.
No more human bias that like when you start talking about schools, we went to the same school, right?
33:42
Like human hiring managers see you in a different light than the next candidate.
They only look at their capabilities, right?
So that changes everything.
Although we as applicants love to do chitchat, right?
We want to make that human connection.
But really what I think like one of the good things about these AI tools or tools coming into the space is that they did bring structured interviewing, right?
34:04
Like if you do have, if you have a one way video interview, there are 3456 questions that are always asked of every candidate, right?
And I think that is, that is a good thing.
There’s no chitchat, although that is now changing that you have a Gen.
AI tool sometimes, right?
Like this tool can sort of adapt to the questions that you’re asking.
34:22
I don’t know how they look at the answers.
If they still have like 6 structured questions and only those get scored.
I have no idea.
You know, I often wish that like AI would be used to help people.
So like, if you are in a job interview, which we still do with humans at the very end, right, Like, can’t there be an AI that like sort of scores the the hiring manager and she’s like, you know what, like when you started talking about schools, it’s really not a good idea.
34:45
You just stay on the script and you know, now that you’re scoring people think about the rubric of things that we developed, like these are the most important capabilities.
And can you let out that, you know, leave at the door that this person, you know, you really got along when you talked about the same comedy or something like that, right?
Like we don’t often see that because I think they’re isn’t necessarily a market for that.
35:04
What we see is like there is a market for rejecting and piling down this like huge amount of swath of people with technology.
But I love that, yeah.
So it’s kind of like AI and hiring would be useful if applied to the actual hiring process of figuring out what is it that we’re looking for in a candidate, what are the most important skills, kind of organizing the recruitment process on the HR end rather than being the filter through which people are chosen to be hired or not.
35:35
Is that kind of what?
I mean, I do, you know, I mean, I think, I think that could be one of the applications, right?
Like, I mean, I think we really have to think through if an LLM is capable of doing what we call a job analysis that used to be standard in a lot of recruiting for very large companies that and industrial organizations, psychologists would go in and sort of analyze like talk to all the stakeholders, but also analyze what do you really need to do this job, right?
35:57
And that is often different than a job description, right?
Because like job description often come from like some hiring manager hired somebody in the role three years ago and now they have to hire somebody else.
They find it on some, you know, forgotten folder in their computer.
Pull that up at 3 more things and then you have this laundry list of skills that you need, which is probably not true and nobody thought about this.
36:17
You just sort of add to the laundry list and that’s not helpful either, right?
There really needs to be a conscious analysis like what do we need for this job?
And maybe also like what are the capabilities we already have in house?
What could we need above that that we would love to hire for?
I don’t think that is often done.
36:32
I’m not sure if like LMS can really help with that, but I think sort of with coaching hiring managers that could be really helpful, helping them develop rubrics, like sort of more objective measures of candidates could be really helpful.
I think you could definitely build a GPT to do it.
36:49
You know, like if that was your intention, as you say, like a coach.
I think you could definitely create a coach.
We see a lot of coaching in this space and hire in in in the world of work.
I haven’t seen it for that particular.
There’s an idea.
Use case.
Oh, I have a lot of ideas.
I don’t think I have it in me to, you know, I love doing this kind of investigative work.
37:07
There is a lot of I’ve pitched it to many entrepreneurs.
I was like, you really should do this in hiring.
There really needs to be someone who does.
But you know, I think also one of the problems we see in this space is like a lot of these AI vendors of venture backed small companies that have to get money back to their investors.
37:23
So they come with like a good intention.
Like they do want to democratize hiring.
They want to make hiring better.
But I think on the way to a product, it’s all about speed and getting to the market quickly.
And then they sort of realized, whoa, science takes a lot of time, a lot of testing costs a lot of money.
37:39
And I think the drive to get a return on the investment for your investors really pushes the good intention to the side.
And that’s why we see.
So sometimes I call it a whack A mole, right?
Like at the beginning, we saw companies use like emotion expression analysis to predict how good you’ll be at the job.
37:56
Like obviously there’s no research on this.
We don’t know what facial expressions and job interviews predict how good you’ll be at the job.
There was like journalism that talked about this.
I did like an investigation, The Wall Street Journal on this, like showing like, hey, this is like really problematic, right?
Like we don’t actually have any scientific basis for this, but we use this as part of hiring processes.
38:15
So finally the company, the biggest vendor in the space, stopped using that.
But then it’s kind of like a whack A mole because six months later, somebody else was like, Oh yeah, there’s a problem with hiring.
I guess we should really use this like emotion facial analysis that we now have.
We have computer vision so much better, let’s use that.
38:31
And you’re like, no, no, no, we already have shown this kind of thing does not work and it’s not ever going to work.
Can come from that.
Yeah.
So I think there is this like drive that is really, really hard for companies to do the right thing and do actually do.
I mean, it would take a lot of work to actually build algorithms that are predictive and find the right capabilities and have really thought to.
38:53
And in fact, some of the companies that do that actually find out that it is so complex that they often reduce their algorithms to just a few things because it is so hard to control all of the inputs that you could possibly look at.
But when we do think about new things and we think about innovation, recently you told us that you had been to HR Tech recently, which does sound like a whole host of new stuff.
39:17
So we’d love to hear some of that and some of the things that you’ve found there.
Yeah, it is sort of a mind blowing conference.
So it’s like 10,000 people in Las Vegas and it’s all about AI and tech and hiring.
And you know, there’s so many companies and vendors on the floor that like every time I go to my hotel room, like like I’m like, I’m like, Oh my God, I’m like so overwhelmed.
39:36
It’s just like such a fascinating conference and I love to go because not all the companies want to speak to reporters anymore about the technology.
But you know, at these conferences, I have a very big sign that says I’m a journalist, that’s my name.
But people, you know, they go there to showcase the technology and it’s a great way to sort of understand, OK, what are the new things, what is coming?
39:55
And you know, we see a lot of LLMS and to a certain extent, agentic AI is also a buzzword in a marketing term, right?
Then I’m sometimes I’m like, I’m not fully understanding how this is different than I guess a traditional LLM.
But we see now these tools pushing into compliance, right?
40:10
That like for HR that like if you have folks that work for you overseas, what are the legal implications of?
There are all kinds of big questions, right?
So we see LLMS being built and marketed for that.
And then what was like really interesting, there’s a couple of things.
Like one thing that we’re seeing is we see still see one way video and interviews where you have like questions on screen that you ask people to answer and there’s no human on the other side.
40:32
And it’s sort of like often pre recorded interviews where somebody’s like, hey, welcome to our company.
We’re so excited you’re here.
Tell us this about you.
And then you get 3 minutes to prepare and three minutes to answer.
And that recording gets sent to the company or to the vendor for processing.
We now see generative AI being used for that.
40:48
So now you have voice avatars.
I haven’t, I’m sure someone has a depiction of an avatar or like a human like deepfake that is talking to you.
But you know, you have sort of a conversation with them.
And, you know, I was, I was on a public radio show in New York and somebody actually called in and said, like, I did an interview with the generative avatar recently.
41:07
And they’ve really described they’re like, you know, I really much preferred it than no one on the other side.
And he’s like, actually like, you know, he was like, it was really personable because that like Avatar for the first time asked me, you know, I think he was a screenwriter, asked me about these like screenplays I vowed years ago.
41:23
And it knew so much about it.
And it really felt like so personal and like validating like.
And he’s like, you know, every time I have a job interview with human and I think we can probably all relate, they look at your resume like 6 seconds before they walk into the room and really haven’t fully understood what you were about at all.
41:39
So he was like, you know, I really liked it.
The question is like, does the structured way of interviewing?
Is that going to the wayside?
Babe, we don’t know and we don’t know how like most applicants feel about this, but we see that.
And then I think what all companies were complaining about is what they call like candidate faking, candidate swapping, deep fakes, taking job interviews.
41:58
So what we see is a lot of companies do have some sort of skills assessment, right?
And I think that’s very normal for folks who are maybe a software developers that they have to say coding test one way or the others.
Like if you hire maybe an editor in journalism and media, we give them an editing test, right?
42:14
Like we edit this article and we’ll, we’ll take a look because that is a huge part of their work and you want to see if they have that skill.
So what we see now, because a lot of that is asynchronous, meaning there’s like no Proctor in the room looking so that some candidates swap each other out.
So like Hilke applies for this beautiful software developer position.
42:32
Turns out I’m really do suck at coding.
I would never get a software developer position, but I know my friend Rose, she’s really good at that.
So she takes that assessment and she passes.
I get the job and like, you know, I show up and you know, it takes the company a little while to sort of understand like, wait, that woman that we hired doesn’t actually know what she’s doing really, but I get a paycheck in the meantime.
42:53
Great for candidates.
This is happening at a rapid speed.
So that some companies said like they actually started moving some of the priorities and hiring away from like sort of these large intakes like LinkedIn and others where lots of people apply and like a moving more of the hiring towards company internal employee referrals.
43:13
And they said, you know, those people are humans, Somebody who works for us will refers to a human.
And that’s great because they also had a lot of deep fake supplying in these like one way video interviews, like, you know, and AI is assessing it.
They don’t know.
In fact, I’ve done this like years ago and I’ve told companies years ago, you have a real security problem.
43:29
Like 3 years ago, I wasn’t in front of the camera.
It was a little bit harder at the time to like generate your voice.
But none of the one way video companies recognized the software didn’t recognize this.
No one there, the audio is generated.
So we don’t even have the basic security things.
43:44
You know, with the pandemic, we do have like a little bit of software that maybe looks around the room, but people can easily check that.
And the software also doesn’t know what Hilke looks like.
So if there’s a human on the screen, they will pass that.
Then I was like, OK, like, well, what are the companies they’re going to help HR with this?
44:00
If this is a huge problem, there must be and there are like, you know, sort of in background checks, you know, there are companies that help hiring companies with that.
They do the background check.
Make sure that you actually did work there, right?
Like some of you maybe have gone through that and they’re like, well, it’s really expensive.
We don’t have a way to really do this and on mass and it’s another entrepreneurial opportunity there.
44:21
This is a huge problem of companies and there’s very few solutions.
Yeah, I mean, it is very difficult and whether you’re on the hiring side, so you’re the company and you’re trying to protect yourself from candidates using AI, whether it’s creating the resume, whether it’s sending in an avatar to go and do your interview for you or getting ChatGPT to do your homework for you, or you’re using bias algorithms to screen out like, you know, it’s made the best AI win.
44:48
I think it’s very, very difficult.
He’ll confirm all the work that you’ve done.
I mean, first of all, just so grateful for this work because it’s just so important to help inform both in industry as well as workers out there.
You know, this is one of the biggest use cases of AI.
I would say.
Rosen are obviously very passionate about companions, but actually AI in the workplace is probably the most important use case in terms of just how many people effect.
45:11
It would be great if you could just share some advice for employers and employees as AI becomes increasingly ubiquitous in the workplace.
Yeah, I think for employers, I hope that they do their homework and start to be a little bit more skeptical of these tools.
45:27
And like, you know, there’s sort of like a list of questions that you can ask, like how like how did you get to this accuracy rate?
How is the validation in any vendor?
You know, they sometimes call it technical report or validation report.
Like if a vendor doesn’t even have that, how they’re validated the tool, they at least can say here’s how we tested the tool, that the tool works for what we say it works for.
45:46
You should like run the other way.
But then you should also find someone who can discern that technical report for you.
I luckily had some experts who helped me with that because I’m not an expert in that.
And some of them were like, wow, this is like 120 pages to sort of hide the fact that the findings were not statistically relevant.
46:05
They were under .3, which is sort of a statistical relevant number, but they sort of gave this whole feel and like how this was tested and a lot of yeah, yeah, all the literature.
And so that can be really helpful to discern this also, Like, sometimes it’s like, so easy, even for me.
Discernible, right?
46:20
Like when I looked at some of the technical reports, like if the tool is only tested on college students between the age of like, 20 to 27, like, I don’t understand how anyone can say this is generalizable to anyone.
That seems wrong for anyone.
And I think anyone in HR could find that, right, like, pretty easily.
46:38
But then, you know, I think we do have a duty if we use these tools to make sure that they work and don’t discriminate.
So, you know, a lot of HR companies or folks in HR rely on the vendor they often offer to do this kind of analysis.
And we’ve seen this maybe most recently in the recession about a decade or so ago, where companies paid the rating agencies to rate their mortgage-backed securities.
47:01
And for some reason, even the most junk mortgages were rated AAA that caused the whole economic collapse.
So I’m saying like, I’m not saying that this might happen here, but I think what I’m saying is like the incentives are not aligned here, right?
Like the person who builds the software shouldn’t be the one that assesses the software.
47:16
That’s not a good idea.
Agreed.
For some reason they’re always come out on top is a mystery.
So like you wanted to have some independent monitoring or really, really have somebody in house who can do this and is empowered to be very critical and also use your kind of data for this, right?
47:31
Like do a sandbox approach where we maybe you pilot something in the background and like really use it against existing technology.
I also wish I’ve never had a company who does any longitudinal study on this, right?
Like looking at quote, UN quote traditional hiring and then new AI tools and like doing this over a long time.
47:47
And then actually compare this applicant who was labeled green, which is, you know, usually higher or whatever way of labeling.
Did they actually turn out to be a good employee?
And obviously that doesn’t work with one or two people because people leave jobs or, you know, there’s many reasons like you have to do this on mass, but we don’t really see that.
48:06
And I wish companies would be more sort of aggressively wanting to know, does this software work that we are spending so much money on?
And for applicants, unfortunately, I think the first step is to know like, what is happening here.
And then there is a way to really think through how can I get my resume more machine readable, right?
48:25
You know, the old tip was always like, you know, stand out, like have columns and colors and like all this cool stuff.
So human would be like, oh, this is cool.
I want to look at this resume.
It’s not working the HFAI and the likelihood that human recruiter looks at your resume early on is highly unlikely.
48:44
Maybe a non for profit of three people that hires like one person every three years.
They might not use AI, but any medium large sized company does use AI for resume screening.
So you want to make sure that the tool is actually like short crisp answers.
Anything that can be quantified, quantify.
49:01
Like you didn’t say if you company millions of dollars, you save $5,000,000 in this way.
All those kinds of things are really helpful.
And then there’s actually AI tools that job seekers can use to upload the job description and their resume.
And it will say, like, you know, we don’t know exactly how every tool is calibrated, but in general, probably you would get this kind of overlap.
49:21
And I think that can be really helpful.
I’ve also had job seekers then say, like, hey, I really, you know, some people meticulously track their applications and they sort of realized in their Excel sheet that if they just send in a resume and cover letter, they weren’t getting any traction.
49:37
And then they started to see, OK, this is this recruiter and reached out to them on LinkedIn and send them their resume.
And that’s how they got started to get traction, actually get interviews.
Some recruiters really don’t like this.
But if you don’t get any traction by just sending your resume might be a way to to think of that.
49:53
Look in LinkedIn and see if any of you people that you know, know anyone inside the company and can do a referral right?
Because I’ve now learned that more companies do that.
Also, usually a referral gets to a recruiter so it bypasses the early rejections.
They will take a look at it.
50:08
I’m not saying that’s unbiased, but often it really is a numbers game and it takes enormous amount of applications to get through these days.
And you see this like companies often get so overwhelmed with application that they close an open job after 24 hours or so because they got already so many applications.
50:26
So you see how fast this goes.
So it’s probably not you, it’s probably the algorithm.
So like, don’t lose faith in yourself.
Just know it takes hundreds if not thousands of applications to get through, especially now in an economy that we in right where there’s like fewer jobs to more applicants.
50:45
Like it’s already stacked against applicants and now that people can apply en masse, it just makes it even harder.
And companies don’t know how to go through all of these applications.
You might just get rejected even though you’re the most qualified candidate.
Sounds like there are a lot of unanswered questions for making this all work in the age of AI.
51:03
And it seems, unfortunately, to some extent, that workers who are against AI system must use AI tools in order to work with that system.
Yeah, it’s, it’s it’s really unfortunate.
I mean, I sort of call job applicants or people in that space.
51:19
I call them like force consumers because if you want the job, right, you get an e-mail with a link and saying like, you have to take this one way video interview in 48 hours, otherwise we’ll reject you.
I mean, I’m going to say no, most people don’t.
They want the job, right.
So they’re going to go with whatever the company asks them to do.
51:35
You know, I think under the American with Disabilities Act, you do have to right in the United States to ask for a reasonable accommodation if you have a disability.
A lot of people who have a disability don’t want to disclose that, right, Because they don’t want to, you know, sort of an imaginary, but they feel like, oh, I’m going to end up on pile B that no one wants to look at because that’s where all the people with disabilities are.
51:55
Not sure that’s always true.
And I’ve also heard that not getting a reasonable accommodation is also harder knowing the age of AI because there isn’t necessarily a human in the loop who sort of sees your voicemail or looks at the emails that are being sent.
But we do have that, right?
And so usually what happens, they will give you a different way to interview maybe with a human, maybe something that you prefer to whatever the company puts out.
52:19
And then I think what’s also really helpful is like, we talk a lot about bias hypothetically in these tools, right?
Or in general in AI.
But I really felt like I wanted to have concrete evidence.
How do these work or not work?
Really wanted to know that.
And I think that is helpful.
We have that evidence.
52:34
You can go to your representatives and lawmakers and sort of be like, hey, we know this is not working in this way.
There needs to be a change here.
And we’re not seeing the change originate with companies because of all of the other economics incentives that push them towards these tools.
52:50
You know, as we said, like hiring is really hard.
Hiring by humans is really hard.
So, you know, it’s really hard for companies to sort of understand like with this work or not, they sort of often believe what the marketing department of the vendors tell them, right?
There isn’t a whole lot of easy benchmarking that you can do in other fields with AI because either you hire me or you don’t.
53:10
There’s no Hilke clone and you can see like, oh, the Hilke clone, but actually was more successful over there.
So we shouldn’t that doesn’t exist.
So it’s actually really hard to benchmark.
I mean, it is doable.
It’s just much harder and very labor intensive to do so.
So there’s no lot of incentive to do that.
53:25
And I think there’s also a shield from liability, right?
Like if the vendor doesn’t know how somebody was scored and the company doesn’t know who somebody was scored, maybe they hope that they will be immune to a lawsuit because nobody knew what was happening.
But it’s, it’s also really hard for applicants to start a lawsuit because most of the time you need to show that you were harmed and just getting rejected is possibly not.
53:48
I mean, it’s not evidence of harm, right?
Everyone gets rejected and hiring.
So we’ve seen a couple lawsuits, but very little.
And it’s just really hard to make a claim.
Yeah, I like the stance of giving people the information and tools to advocate for the ethical hiring process that they want and they deserve.
54:07
And with that in mind, I both on the workers getting hired side and the companies doing the hiring through AI tools, what they all need to do is go read your book so that they can know where is it not working, where are the gaps?
What are the things that people are calling for?
54:24
What works in hiring with AI and without AI?
And what does good things with hiring with AI look like in the future?
So to all of our listeners, please go take a look at the algorithm.
Go take a look at Hilke’s other work because this is really important, especially for knowing how to move forward and being informed and intentional with AI at work.
54:46
Again, don’t get discouraged by all of the rejections as much as you can, but certainly advocate for what you want to see.
So Hilke, thank you so much for chatting with us today.
We really appreciate you being on the podcast.
So please, everyone, go check out Hilke’s work and thank you so much for being with us today.
55:02
Thank you so much.
Thank.
You both thank you.
OK.
So I really liked what we covered in the episode.
I think we got a really good sense of how AI is used in hiring.
We didn’t really touch on too much how AI is used in the monitoring process.
55:20
So Filka did mention at the beginning how AI can be used to monitor employees, see how much time they’re taking off, how many restroom breaks are taking, etcetera, etcetera.
So we didn’t get into that.
But what I was interested kind of to hear more about is the sort of information that AI picks up on from the resumes, CVS, cover letters and interviews that are being used as proxies.
55:47
One thing that Hilke mentioned was maybe an AI algorithm is used to sort through resumes, and the pile that it comes up with of people to interview is a bunch of people named Thomas.
Yeah.
And then there is also the example of how an AI algorithm might look at not people’s names.
56:04
Maybe it’s this algorithm that’s being applied and the recruiting process is told or not given people’s names or anything like that that’s supposed to be identifiable about their gender or whatever.
But then the algorithm might use for example certain hobbies as proxies for what that applicants gender may be.
56:20
And again, he’ll could give an example of if someone’s hobby is softball versus baseball, the algorithm is more likely to say you should interview the applicant with the hobby baseball because historically, given historical data of who gets hired, it’s the person who is a man who is more likely to have a hobby that’s baseball rather than softball.
56:41
So I think those are some things that I would have liked to go deeper into or just I just want to know more about.
But I recognize that maybe there’s just not a lot of research on it.
Well, I think what the challenge is there is it’s a little bit like the LLMS in the first place.
In the past, we would have spoken about it.
Why does an LLM, when you ask it for pictures of CEOs, does it give you a white male CEO?
57:01
It’s like, oh, well, that’s bad, right?
And it’s like, well, it’s because it’s based on historical called pictures and that’s predominantly what that job has been made out of.
Fine.
So then should we be changing future pictures so that we base it on the future that we want to have or to represent the past that we’ve come from?
57:21
And and so those are like some of the very early questions around bias in algorithms of other lens.
I mean, I’m not telling you something you don’t know, but I remember those are some of the early conversations around image databanks, right.
And so I guess the question here is you trying to create these algorithms for hiring and you’re saying we don’t want to have a bias around gender, we don’t want to have a bias around race, We don’t want to have a bias around where people live like a whole lot of protected variables that we want to then take out.
57:50
And so all the algorithm is going to do is it’s going to look for other overlapping variables and it’s going to then start coming up with quite random things, but it is going to find the closest proxy for the predictive variables, right?
58:05
And what I find interesting then is what it’s not doing is are the people who score the highest in map for the engineers?
But maybe it is.
Maybe that is one of those variables, but the humans who are auditing or vetting these tools are just unaware that that is what the algorithm is using.
58:22
I think there are ways to test whether or not certain proxies are what the algorithm is picking up on.
For example, putting in identical CVS that have baseball versus softball.
And then if you see a discrepancy in which baseball or softball are more likely to get filtered into the interview process, then you can at least say with some level of confidence that, yeah, the algorithm is picking up on the hobbies, but there might be something else going on to a greater degree that is not tested.
58:54
So it does require, at least at this stage, given the black box understanding of algorithms and how they work, we have to specifically look at certain things and reduce the amount of variables that we’re looking at.
And we’re making choices about what it is that we think the AI might be using as a proxy, but that doesn’t necessarily cover what it’s using most.
59:15
So I think that makes it difficult.
What do you think we need to do next steps wise?
It feels very unclear, but especially with your experience hiring a lot and being embedded within an organization, what do you think would be the most walkable path forward and most optimal path forward?
59:33
It was very interesting getting to that point where Helka was talking about, wouldn’t it be better if the AI tools were used to then coach the managers, to coach the hiring managers?
The kind of pressure points that exist come even earlier where hiring managers aren’t clear about even the kind of person they need.
59:53
You know, it’s very difficult to come up with the right job description if you don’t know the kind of person you need.
So if these AI tools were then more as like copilot coaches, and I’m not saying we should just be relying on AI, but we’ll come back to a second about why that’s almost inevitable to some extent in big companies.
1:00:08
But if they were, then like, well, do you actually need someone on your team?
How is your team operating at the moment?
What role would they actually?
To be fulfilling.
And then it’s like, who would they report to?
Are you sure that person’s going to be supporting them?
What is their onboarding going to be like?
So they have like everything that they need to succeed.
1:00:26
So you’re also thinking about the downstream impact for that person that’s coming into the team.
So there’s a lot of coaching embedded into this whole system.
Now.
That’s almost like a CBT tap model.
So that’s easy to build and that’s not difficult to build in.
And so that’s kind of like the Socratic conversation that happens at the time of I want to hire somebody.
1:00:45
And then you kind of build that.
Well, what are we actually looking for?
And then you can start kind of coaching the people that are looking at the CVS.
What are you really looking for?
Are you sure what’s going to work for the team?
And almost like a coaching through the whole process, coaching the team that’s looking at the people.
1:01:02
So I think that would be optimal and and then that kind of teaches people to think more critically.
Going back to a lot of the conversations we’ve had around is AI cause and cognitive deskilling.
Well, actually it could cause cognitive upskilling if used in that way.
1:01:17
And there, I think, is some real gain to be had.
It would just be so valuable to have clarity on what does a successful hire actually look like in terms of the person who is fulfilling that role and the process to get them in that role.
1:01:34
I think that it would be very useful for companies who are considering or are implementing these AI tools is to figure out what really makes a good applicant and what are the ways that these AI algorithms can help with that versus take away from that.
1:01:51
Because right now it feels as though applying AI tools for hiring is so much about not having enough personnel and not having enough time to get that outcome that you’re seeking.
But when that is the mainframe as the reason for applying these AI tools, it kind of forgoes the actual main purpose and the main benefit, which is finding someone who is a good fit for that role.
1:02:15
What I think is interesting is I don’t think that’s that difficult to solve.
So again, you can debate whether or not you think these are good measures.
And I’m talking like in very big companies where you are hiring huge numbers of people.
But in those companies, they’re also doing a considerable number of performance reviews where they’re using very clear metrics.
1:02:36
Now again, I’m not saying whether the metrics are right or wrong, but they are.
They are creating metrics.
So it would not be beyond the realm of possibility where you have these 10 candidates that came in, they were rated fours and fives by the algorithm right when they came in.
1:02:52
You would be able to measure them over the next three years and measure their performance rating over these three years and correlate the extent to which are fours rated fours in their performance view.
Or is there no relation between how the algorithm rated them when they came in and their performance reviews over the years 1-2 and three?
1:03:11
Do we have any relationship between their performance reviews or retention and the people that came in or not?
So I don’t think this would be that difficult to measure.
I think it goes back to what Hilco was saying, that these companies have the resources to do that research.
I’m talking very big companies.
1:03:27
Yeah.
I’m also thinking, though, if there is a bias, well, there is a bias in how the algorithm filters out who gets a callback and who might, might end up getting hired.
But there’s also so much to be said about an employee’s success when they fit in with the kind of culture of the organization or when they feel at least as though they fit in maybe by virtue of seeing themselves represented on their team.
1:03:52
Right?
So there’s already bias in terms of who will succeed and there’s bias in terms of who gets hired because the metric of being set up for success is that you’re similar to those in your organization.
So it’s like with your point of how do you bring in more diversity of background, of voices, of experience, of skill into the workforce and make it such that people can really succeed in that environment.
1:04:19
Am I having some sort of setup of being like, listen, having a bunch of different voices in the room is actually really good for your well-being and beneficial for the productivity of the company.
So this is what our status quo is.
This is what we are moving forward.
1:04:35
I think that would be just so helpful to have this very intentional reframing and you know reinvigorated push for diversity that we are not seeing like you said with AI application, we’re reverting back.
So I don’t know how do we get that done.
1:04:50
Yeah.
And I think it takes effort.
I think your point around being intentional is exactly the point, because it takes real consideration for people not just to embrace diversity at a hiring perspective, but then to your point, to create the environment where people can thrive and learn to respect and get the best out of each other.
1:05:11
That’s not going to just happen because we hired for diversity, But to actually create a sustained culture where people can celebrate and enjoy a diverse working environment and get the best out of each other, that’s another piece of work that takes a lot of attention to your point.
1:05:27
And takes a lot of humans in the loop once again.
Again, listen.
To invest more time, effort and care at a foundational level for AI to be good in the workforce, but also for humans to be good to each other in the workforce.
1:05:43
So I guess that’s a good theme to end on, right?
Yeah.
Angie and I always go back to it, bring back the human, Let’s get back to human.
But that doesn’t mean no AI at all, right?
That’s not what we’re saying.
Yeah, it’s just how we use it.
And that’s why you’re seeing Angie and I talk to each other on a podcast rather than listen to a notebook LM generated podcast because we could do that, but that would forego the main point of this podcast, which is US speaking to you from our own experience and keeping the human in the loop.
1:06:12
Exactly from different sides of the world.
So bye from Angie.
And bye from Rose.
We’ll see you next time.