Discussion keeps the world turning.
This is Roundtable.
Hello everybody, welcome to Roundtable.
Coming to you from Beijing, I'm He Yang.
Good as always to have you join us on today's show.
We discuss the role of AI -enabled hiring in the recruitment process around the world.
AI, like humans, is susceptible to biases.
Underscoring the importance, of Vigilant Oversight.
Together, let's navigate the intricacies of AI hiring bias and contemplate more equitable recruitment outcomes.
And should grass be walked on?
We take a look at the shifting attitudes towards the enjoyment of Chinese parks and public green spaces.
For a cross -border perspective, in many countries, walking on grass is seen as the natural part of outdoor leisure.
For today's program, I'm joined by Lai Ming in the studio and Josh Cotterell on the line, first on today's show.
Ever since ChatGPT was released into the wild in November 2022, there's been a lot of both exuberance and hand -wringing about how generative artificial intelligence tools, like chatbots, could change work as we know it, starting with how we get that job.
A survey from Resume Builder this year found that almost half of U .S. job seekers were already using chat GPT to generate resumes and cover letters.
And those who did reported getting interviews at a higher rate.
And the trend seems like a win for job candidates.
But it could further complicate a hiring process in which employers are solving hiring challenges exacerbated by AI, likely involving more AI.
Bots already play a big and sometimes discriminatory role.
A recent Bloomberg experiment in the U .S. found bias against job applicants based on their names alone.
So, Lai Ming, could you tell us what happened?
Well, so we learned that artificial intelligence can help us do a lot, like picking up calls and chat with users of any platform, and they can even write your essays.
But apparently, many employers are also using AI like GPT produced by OpenAI to help them sort out the resumes that they're getting and to screen the kind of candidates that they would like to potentially employ.
And when it comes to recruitment, we know it's competitive and everything that a candidate can hope for is for them to get fair treatment, to compete on an equal playing field, on a level playing field.
so this is actually a rather sensitive matter and luckily Bloomberg, like you said earlier, looked into the issue, they've done some experiments with the technology, they used GPT to generate eight different resumes and then edit them to have the same level of educational attainment, years of experience and job title and the only difference is the name of the fictitious candidate and whether it is statistically associated with men or women who are either black black, white, Hispanic and Asian.
So these are all race related.
And those demographically distinct names were randomly assigned to each of the eight resumes.
And then reporters prompted GPT to rank the resumes to identify the most qualified candidates for a real job opening at a Fortune 500 company, including HR, business partner, senior software engineer, retail manager manager and financial analysts.
So there are different posts and there are different candidates with different names applying, presumably in the experiment, applying for different jobs.
And they actually did about a thousand times with the experiment, using hundreds of different names and combinations of resumes and educational backgrounds to see whether AI can express certain biases that may or may not already exist in our society, when when humans recruit humans.
And their finding is not, well, encouraging.
So the result is, if GPT treated all the resumes equally, each of the eight demographic groups would be ranked as the top candidate one -eighth or 12 .5 % of the time.
However, GPT's answers displayed disparities.
Resumes with names distinct to Asian women were ranked as the top candidate for the financial analyst role, more than twice as often as those with names distinct to black men.
Resumes labeled with names distinct to black Americans were the least likely to be ranked as the top candidates for financial analyst and software engineer roles.
And GPT seldom ranked names associated with men as the top candidate for HR and retail positions, two professions historically dominated by women.
And GPT was nearly twice as likely to rank names distinct to Hispanic women as the top candidate for an HR role compared to each set of resumes with names distinct to men.
So there are social, racial and gender factors being at play here that may skew the ranking of an artificial intelligence.
But remember, this is one type of artificial intelligence technology, and it doesn't represent every possible AI technology that we already have or that we may come up with in the future.
Yes, and currently there are many job screening AI tools out there, and ChatGPT is merely one of the many available out there.
Josh, what was your initial judgment after listening to the results and findings?
things well sadly i'm not really that surprised i think that biased training with ai systems like this is pretty evident in a lot of different industries and i think that when it comes to recruitment it makes a lot of sense because recruitment takes a long time it's almost it's a massive industry in itself actually right the recruitment industry a lot of people work full time as recruiters and obviously internally within companies this takes up a lot of resources um and there are a lot of people being paid to do this.
So, if you can save a lot of time and then make more and then make effective recruitment decisions through AI, it's quite understandable.
But then from a moral perspective, it really does seem pretty unfair.
And we know that there's a lot of algorithmic bias in AI generally.
These algorithms are flawed. We know that they can be flawed and that they do have a moral basis, right?
That's the problem with AI.
One of the the main issues with it something it can't do is have a morality right and one of the reasons for that is because we have different moralities right uh on an on individual basis but when it comes to something like this it does seem to be quite unfair and if somebody maybe is getting cut out of um say an employment process simply because of um you know some sort of statistic or because they didn't go to say a particular i don't know university i mean or something like this what what chance do you have?
Right. I mean, if this becomes the norm for all industry across the board, it's quite scary because it seems as though you're immediately debilitated.
Right. Yes. And well, I suppose technology, especially AI, is only as good as the data that it's been trained on and also how we use technology and the results and consequences from the usage of technology often reflects the kind of world that we live in or the kind of situation a particular economy is in.
And in China, there are also media outlets who've done similar hiring experiments.
What are the findings?
Well, another experiment randomly generates various types of resumes featuring candidates graduated from prestigious Chinese universities, including Tsinghua University, They have former work experiences in big Internet companies, including Tencent, Alibaba, Baidu and ByteDance.
And the experiment shows that AI has a ranking system for the universities and companies during the resume sorting.
And in the eye of ChatGPT, Tencent ranks first, followed by Alibaba, Baidu and ByteDance in terms of work experience.
And considering the universities, Tsinghua University ranks first, followed by Peking University, Zhejiang and Fudan and Shanghai Jiao Tong Universities.
So we don't have the same kind of issues that Bloomberg has identified with their experiments pertaining to race and gender, at least according to the experiments that people have done here.
but there seems to be certain preferences over particular schools and workplaces and I don't think that is an issue because honestly if a human were to screen the resumes they would have some kind of metrics they will have to have their own preferences as to what kind of qualifications they're looking for and what kind of talent they need so I'm not sure if that is an issue here also pertaining to what Josh mentioned earlier about AI having a moral basis.
I'm not sure technology is capable of having a moral basis.
It only reflects the kind of data that is being fed.
It reflects the kind of moral attitude that human beings, those who design the software, can have. And I'm not sure if that comes off as a flawed technology because it simply reflects the kind of reality and the social reality that we are now living in.
And what happens when the social reality might be flawed, right?
But we'll get into that a little bit later.
And Josh, here is an interesting, maybe difference in opinion or difference in perception.
Because here in China, a lot of people, and I read some of the online comments of people just sharing randomly what they think about this Chinese experiment.
And then a fair number of them kind of stand in a camp similar to what Lai Ming just iterated.
That is, why is this bias?
If you've actually been through the ringer of Gaokao, and that's the college entrance examination in this country, and you've come on top, you've entered a top university, and therefore you should have a better chance in getting employed.
And also, you've been recruited by one of the best well -known companies in the tech world in China, then of course you get a leg up in being selected for your next job.
So what do you think about this?
Is there a bias towards maybe better education?
Also, you know, better career achievement already?
What do you say? Yeah, well, I think the issue that you're talking about is is a greater problem with society as a whole, in my opinion, in that we value for some reason, maybe obvious reasons to some people, but we value, you know, having gone to a really good prestigious university and we somehow equate that to being competent and intelligent.
I personally think that just because you went to a good university and a good school does not mean you're intelligent at all.
And I think that that's something that we have to try to change.
But I I mean, it's true what you say.
I mean, if AI can give us the best people for the job, is that maybe even more fair, right?
Is that maybe even more fair and more objective than say having a human evaluate this kind of thing?
The problem again, as I say, is actually I think AI maybe has the ability to make things more fair, but it doesn't appear as though it's doing that right now.
I think that actually AI has the ability to combat itself in this way with things like algorithmic fairness tools right so i think that it can be used to implement fairness and at the same time find the right candidate right because that's very subjective but when it's simply cutting people out because of something like an educational thing like um you know you've gone to a specific university or something like that um or if you go all the way back to gaokao that's i mean that's pretty scary that's even before university and we're we're so young when we do those university entrance exams. Does
it mean that if you mess that up when you're 16, 17, that your life is suddenly defined by that?
You know, I mean, to some degree, it is already a bit like that in most societies, right?
Like, if you don't have a degree, it's really difficult to even be considered for most jobs in certain industries, right?
And in my own country, in the UK, we decide our subjects, you know this, Hayoung, you decide your subjects really early, right?
You choose three subjects when you're 17, depending on which school you're in.
And you're sort of defined by those studies really, really early on.
And there's problems with that already.
But I agree with your point.
But I do think that AI right now is not being used very well, but it has the potential to be used really well if it employs algorithmic fairness tools.
I'm not sure if we can ever help the AI become fair in real sense, because I don't think the real world is fair.
I mean, despite what we want, what we all want in recruitment or in anything that we encounter in our lives, there are measures that's put in place for the sake of political correctness.
Because if the reality is such that a certain group of people don't necessarily receive the kind of education as they do as the other group of people, and just they they don't turn out as qualified or as skilled as the other group, then that is the case.
I mean, certain, for instance, here in China, there is not the element of race here, but there is the element of socio -economic development and access to resources, educational and social, and that would affect the development of a student growing up.
And if a student is exposed to better quality education in urban areas, metropolitan cities, then they are more likely to succeed.
In order to make up for the shortfall for students that grew up in remote and less developed regions, there are certain policies, preferential policies, awarded them so that people from, for instance, certain ethnic minority autonomous regions, they are given some extra quotas when it comes to the very competitive college entrance examination.
This is artificial, an advantage that's awarded to them given the fact that we might need these people to get opportunity for further education so they become the kind of talent that their region might need if they would like to have a stability and prosperity going forward this is a long -term strategy that this country has decided on and our population has agreed upon so there is no fairness in this regard there's no absolute so I'm not sure if we can change that with the AI technology, because there's always a need for consideration in the long term or in a larger picture, as compared to this particular
recruitment session or this particular college entrance examination test. The world is much more complicated than just one recruitment session.
Yes, and that's a lot to take in, but i'm digesting and certainly we're seeing already in the tech world some good intention people are trying to mitigate the situation and it's complicated i'll give you an example so ai text to image generators that thing we're kind of familiar nowadays as well and it have a well -documented bias problem especially towards race and sometimes towards gender and AI models are trained on images from the internet.
So bias in, bias out.
A 2023 experiment also from Bloomberg on the image generator stable diffusion found that AI portraits of architects, doctors, and CEOs skewed white and male while images of cashiers and housekeepers skewed towards women of color.
And solutions to this AI image bias raised their own ethical questions too.
So text -to -image AI tool Firefly has deployed a de -bias algorithm to tackle the racism issue in its output.
In Firefly's world, about 14 % of doctors should be Black, and the same percentage as the Black population in the US.
But in this messy, unequal real world in the US, only 6 % of doctors are Black.
So should AI images depict the world as it is or as it should be?
And I guess a similar rationale could be a question for the hiring process as well.
Josh, what do you think?
Yeah, it's a real issue here.
And I think that one thing is that when we ask AI questions, often it gives us back sort of singular answers or singular images right now.
And when it comes to representing or articulating something as nuanced and diverse as the human population and also the human experience, human history, which is different, you know, for different people from different perspectives.
When it comes to representing bias and discrimination, all of these things, it really struggles.
And to be fair to AI, I personally don't think that's a very easy job anyway to do.
do. So it's an issue.
And I think to some degree, maybe this is a good thing because what it's doing is, in a positive way, is it's aiding discussion.
And it is forcing us to take a look at our own bias, our own biases, and think about how we are represented.
And, you know, things like the topic we're talking about right now, right, recruitment.
For a long time, there's long been discussion about how the recruitment process has been pretty corrupt and easy to manipulate and nepotism and bias and things like this being at play and people from certain backgrounds, people from certain educational institutions, ethnicities, etc, have been preferred in certain industries.
And now AI is being used, although it's not doing a very good job right now, at least we're able to see it's already more transparent and this has been something that we've wanted for a long time in a lot of industries is transparency transparency on things like salaries dependent on you know um the the male female gender pay gap right this is something that is needed for that is transparency or with salaries and stuff like this and also with recruitment systems ai at least is becoming a bit more transparent hopefully i mean i'm sure there are going to be ways in which it's not going to be transparent
but so yeah uh it's difficult for me to give a straight answer to your question but um i think that it's quite difficult um to accurately you know represent these images and stuff like this but i do think there are some positives to be taken away and much like what we do in our real life given certain advantages and preferences to particular group of people for the right reason we can somehow change the metrics that we put in artificial intelligence technologies to try and reach a point where we want to be in terms of giving equal opportunities to different groups of people, and that's for sure.
But I'm not sure that this is a very easy job, because there are certain discriminations that are so entrenched, that are so hidden, that we couldn't name them all.
We couldn't change them all, because, for instance, certain discrimination in the Western world is encoded in zip codes, in education backgrounds, in life experiences, in the kind of habits.
For instance, certain group of people would like to use cash more than the other group of people.
So if AI were to catch on on these metrics and use them to skew the results, then I don't know how we can go about changing these things.
Also, with regard to whether AI is enhancing existing biases that we already have in our reality i'm not sure how to look at this on the one hand i think it only reflects the kind of reality that we live in but on the other hand in keeping the way that society operates and also in allowing human beings to trust ai if we were to trust them and full -heartedly maybe we neglect the kind of manual inputs that we often introduce in as when humans conduct these businesses and if If we put our whole trust or all of our trust in our AI, we forget about these things, then the biases can be entrenched further
on. And this is problematic.
It is. And it's interesting you pointed to that, Lai Ming.
And with proper development and oversight, AI algorithms can enhance fairness by standardizing evaluation criteria.
Implementation of AI can also streamline processes, leading to more efficient and consistent hiring decisions.
But also, I wonder if you are just that unique, bright firework.
You are the outlier.
Do you ever stand a chance in the world of full AI and technology if we just, you know, push the envelope and make it really extreme?
Because I feel with technology, because everything is supposed to be standardized, and therefore, if you're just that unique talent that is different, so maybe in some criteria you're like subpar, you're weird, and then you might not get a chance.
But if you're talking to a human being, let's say in a job interview, and maybe your unique quirk or something special can actually catch the eye of another fellow human being.
But if you're an outlier, maybe you just get the door shut on your face and you don't stand a chance.
Or am I too pessimistic about this, Josh?
Well, I really like that you brought this up because I guess that not just in this particular circumstance or with this topic, but AI, from what you're saying, I mean, AI really has the potential to suck the humanity or, you know, all of the little nuances and chances and opportunities and differences and things that make human life worth living.
you know i mean yes there's the opportunity here to make things really really fair and black and white and standardized right as you say but is that always a good thing i guess the the flip side of that is is that you know you don't get the opportunity to work the system and um make opportunity for yourself make chance for yourself and just as much as the system can you can see the system is working against you which i'm sure for many people it is more than others you also have have the ability to make the system you know maybe work in your favor right but it really depends here i mean we could talk
about some there's many things you could use your example to talk about i mean um dating for example right i mean ai hasn't really found its way so much into the dating world yet but i i imagine that it very soon will be there in finding you yeah i mean imagine all these dating apps all of the data in those dating apps if there's a new app that where you can allow Allow your previous dating history to be loaded up to it.
And then it finds you your perfect partner, right?
I mean, I imagine that love and romance, I hope, in my romantic mind, is too nuanced and difficult for AI to help with.
But I imagine AI can probably find you a pretty good match. But that's pretty scary and sad, isn't it?
I mean, it almost feels like it's taking some of the soul out of the human experience.
So I get what you're saying.
saying. But I must say, I think that probably if it's done in the right way, it's probably for the better that AI helps with things like recruiting.
But if it's done in the right way, I have doubts whether it will be.
Well, at the end of the day, it's up to humans to ensure that AI works for everyone, not just for a select few.
By combining the best of two worlds, AI's efficiency and our human empathy, can we build a hiring process that's fairer and more inclusive?
How is that going to happen?
Well, feel free to contact us.
You can send us an email or voice memo to ezfmroundtable at foxmail .com.
