Transcript

Would ChatGPT hire you? Your age and gender matter

From nature video · Episode page · Watch on YouTube

October 16, 2025 · 11 min · 1,823 words · 18 timestamped sections

This is the full, timestamped transcript of "Would ChatGPT hire you? Your age and gender matter", an episode of nature video published October 16, 2025 (11 min, about 1,823 words). It mentions 5 people, works and ideas, including Douglas Guilbeault, Shamini Bundell, Google Images, IMDb, US Census.

What this episode is about

New research reveals a significant stereotype online that portrays women as younger and men as older, influencing perceptions in hiring practices and contributing to the gender pay gap. This study, led by Douglas Guilbeault, highlights the distorted representation of age and gender in digital platforms, using analysis of 450,000 celebrity images from IMDb to showcase these biases in AI training data.

Key takeaways

The internet ages women prematurely, with female celebs averaging 6.4 years younger than their male counterparts.
Search results subtly shape societal perceptions, with women linked to youth and men to authority—it's a digital dichotomy.
Job titles evoke age stereotypes; 'chef' feels youthful while 'CEO' suggests older—this skews opportunities for women.
Even language reflects bias; occupations dominated by women correspond to younger descriptors, revealing hidden ageism.
AI learns from biased data, perpetuating gender stereotypes—it's a tech-driven game of chronological cat and mouse.

Who speaks in this episode?

Shamini Bundell (Host) · 7 min speaking
Douglas Guilbeault (Guest) · 5 min speaking

What is discussed, and when

0:00 – 3:00: Douglas Guilbeault, Shamini Bundell, Google Images, IMDb
3:00 – 6:00: Shamini Bundell, Google Images
6:00 – 9:00: US Census

Notable quotes from "Would ChatGPT hire you? Your age and gender matter"

Let’s start with IMDb, the Internet Movie Database, a website filled with photos of crew and performers in film and TV shows.
about IMDb · at 1:32
DOUGLAS: We wanted to get the internet's depiction of regular folks too.
about Douglas Guilbeault · at 2:52
SHAMINI: Your job is being an assistant professor.
about Shamini Bundell · at 3:43
Not to mention of course if you look at just the raw census data on how long women and men live, women actually systematically live longer than men so on average they’re older.
about US Census · at 7:02

Who and what is mentioned in this episode?

People

Douglas Guilbeault — Person · Main guest · 1 mention · first at 0:57
Lead author of the discussed research; explains methods and findings about age–gender stereotypes in online data and how they propagate into AI systems like ChatGPT and affect hiring/pay outcomes.
Shamini Bundell — Person · Host / interviewer · 2 mentions · first at 0:00
Introduces the episode’s premise, frames the societal implications of the research, and interviews Douglas Guilbeault to unpack findings about gendered age stereotypes online and in AI.

Companies & orgs

Google Images — Company · Data source · 2 mentions · first at 2:11
Used to gather images for analyzing age and gender biases in search results.
SHAMINI: The team scraped Google Image search results for different words and then hired people to estimate how old the faces in those images looked.
IMDb — Company · Data source · 1 mention · first at 1:32
Provided data for the study on age and gender biases in online representations.
US Census — Company · Data source · 1 mention · first at 6:47
Provided real-world data to compare against the biased representations found online.

Full transcript with timestamps

0:00 SHAMINI BUNDELL: Hello! I'm a woman on the internet and it turns out that the internet, and AI, have some strange ideas about what that means. According to new research there’s this pervasive stereotype online that women are younger and men are older. It pops up everywhere from age gaps between celebrities to search results for words like ‘person’ or 'neighbour'. It’s in images, it’s in text, it’s in Chat GPT’s training data and these experiments show that it could be impacting the gender pay gap and your chances of getting hired.

0:41 Welcome to the internet. You may have been here before, and you may have noticed that it doesn’t always accurately represent reality. I’ve been reading a new study which reveals some striking biases in online data and in AIs like ChatGPT. Douglas Guilbeault is the lead author of the paper so I gave him an online call to find out more. DOUGLAS GUILBEAULT: Increasingly the internet is actually one of the main ways that we understand our social world. And so we started to become very interested in the ways in which the internet was not necessarily reflecting reality, but maybe providing a distorted picture of who we are in the social world.

1:20 SHAMINI: Douglas and his team had previously looked into the distorted over-representation of men in online text and images. Now they wanted to add age into the mix. They collected a huge amount of data from all sorts of different sites. Let’s start with IMDb, the Internet Movie Database, a website filled with photos of crew and performers in film and TV shows. The team cross-referenced 450 thousand photos of celebrities with real information about those people. DOUGLAS: We can get their actual age and their self-identified gender.

1:51 And crucially, we could also see the time stamp of when the image was taken that's put on IMDb so we actually know exactly how old they were in that image at that time. The results? Female celebrities were on average 6.4 years younger in their photos than the male celebrities. The same kind of pattern popped up on a database of celebrity photographs from Google Images. Here’s a graph showing the number of photos of women at different ages compared to men. The most common age for female celebs was 25 whereas for men it was 44.

2:27 DOUGLAS: These are people who are influential, these are the idols. And so seeing this age distribution in such an impactful population of people is extra concerning from the point of view of the possible consequences of these stereotypes. SHAMINI: Now you might have noticed that the entertainment industry isn’t always a bastion of age and gender equality. But Douglas and his team weren’t done yet. DOUGLAS: We wanted to get the internet's depiction of regular folks too. And so we looked at a comprehensive set of what we call social categories which includes occupations, so like, banker, doctor, carpenter, baker, but also everyday social relationships like friend, colleague, and neighbour.

3:16 SHAMINI: The team scraped Google Image search results for different words and then hired people to estimate how old the faces in those images looked. And it turns out it’s not just celebrities. Whether the search was for neighbour or scientist, writer or colleague, the women in the photos were constantly estimated to be younger than the men. In every case, the internet has a specific idea of what someone in a particular job should look like. SHAMINI: Your job is being an assistant professor. Do you think if you Google ‘assistant professor’, do you think you look like what the internet thinks an assistant professor should look like?

3:53 DOUGLAS: In some ways no, I can say like, personally, when I go to the cafeteria here, they often ask me if I'm a student. Maybe that's my inability to grow a full beard. SHAMINI: You need a mustache. DOUGLAS: I can grow a mustache. I don't know if the world wants that. So maybe I'm a bit of an outlier, but I will say though that there is a sort of stereotypical image of what a professor looks like. And of course this can vary, but the sort of archetypal image is an old white guy, like with a corn cob pipe and pads on his sleeves, right?

4:31 But that's changed a lot. It does not at all reflect the actual distribution of professors today, but it was historically the case that it was older men, and there's still a legacy of that stereotype impacting our understanding of what it means to be a professor. SHAMINI: These old man stereotypes are still popping up all over search results and on websiteslike Wikipedia, Flickr and YouTube which the team also looked at. But is this just a perception problem? You might be thinking, maybe women just look younger. As a woman, I certainly feel like there’s a bit of pressure to appear youthful.

5:08 So the team decided to move past photos and dive into the text of the internet itself and they were looking for patterns of word associations. What kind of jobs are associated with what kind of words? DOUGLAS: The gender pronouns like he, she, him, her, which tells you whether a given category on average is being described as or is related to a man or a woman. And then with age it was basically categories like young, old, child, adult. SHAMINI: Once again the team were looking at social categories like jobs. Some jobs were often described with hes and hims and words like old or senior.

5:46 DOUGLAS: You have like chairman of the board, chief of staff, director of research. SHAMINI: Other jobs were linked to she/her pronouns and words associated with youth. DOUGLAS: Cook, secretary, intern, even nurse. But importantly, our results actually are over thousands and thousands of these categories. And what we see is just a very strong correlation. The more a category is associated with women, the more it is associated with these younger ages. SHAMINI: The textual data that they used here is the same data that’s previously been used to train large language models like ChatGPT.

6:26 DOUGLAS: The fact that it showed up in every single language model that we analysed tells us that it's not about a particular data set. It's not, oh it's just about Twitter, or it’s just about Reddit. It shows up everywhere once you have enough language data from the internet. SHAMINI: And a key thing about these stereotypes in images and text is that they’re not true. DOUGLAS: We then look at US census data on occupations and industries. And what we find is actually, especially in the last decade or so, there is no real age gap between men and women.

7:02 Not to mention of course if you look at just the raw census data on how long women and men live, women actually systematically live longer than men so on average they’re older. SHAMINI: I'm gonna beep out the data, but I wanna know who is older out of us, as an ‘n equals one’. We have a woman and a man on the internet. I was born in [BEEP]. DOUGLAS: I was born in [BEEP]. SHAMINI: The stereotypes have been smashed. DOUGLAS: That's right, that's right. SHAMINI: I am an old woman on the internet. DOUGLAS: You wouldn't know it.

7:33 SHAMINI: And you're a young, youthful man. Just goes to show. DOUGLAS: No, I know, these patterns are, they're pervasive. SHAMINI: And these pervasive patterns could be part of why there’s a gender pay gap; why research shows that women across a range of industries earn on average less than men. Because the jobs that showed a big age gap online, also had a bigger gender pay gap in real life. And the stereotypes aren’t just staying online. Douglas and the team did another experiment that showed that getting people to Google Image Search particular jobs influenced their beliefs about those jobs.

8:08 They asked people to estimate the average age of workers in that profession. The ones who’d Googled it first thought that people in male-dominated jobs were on average older. Douglas calls a self fulfilling prophecy. DOUGLAS: If the stereotypes are sort of what we think the world should be like, then we interact with the internet and we start to increasingly believe that our idea of what the world should look like is actually what the world looks like. SHAMINI: Which brings us on to Artificial Intelligence and those huge models that take the stereotypes we publish online and feed them back to us.

8:44 So let’s talk about ChatGPT and one way that it’s commonly used: to sort through job applicants. DOUGLAS: Organisations are using ChatGPT to, say, identify the top ten percent of resumes to then expedite their own review process. And then there’s a question, which resumes actually get into that top ten or five percent, right? And what we see is that ChatGPT actually gives the highest scores to older men. SHAMINI: Here’re the results of Douglas and his co-authors’ experiment with ChatGPT using fictional resumes. For older applicants men have a clear advantage.

9:19 But, surprisingly, for younger applicants women actually rank higher than men. DOUGLAS: It also reflects a known pattern in systemic inequalities in labor markets and hiring where increasingly there actually is a preference towards hiring younger women. However, in terms of research on what's called the glass ceiling, there's also the case that younger women are less likely to be promoted into higher status positions than younger men. SHAMINI: Age and gender are relatively simple things to look at: you can quantify them.

9:51 But there’s yet more factors associated with both of these that’re harder to pin down. Who in society has authority, wisdom, experience, leadership skills? And like Douglas said, and like this research shows, these patterns are pervasive. DOUGLAS: Even something seemingly simple like the gender and age of an occupation, we actually have a kind of mythology around this. And this is concerning because the AI revolution and these large language models rely on learning from all of this data that humans produce. And so what we're doing is creating a system where it's basically learning from our stereotypes and our mythologies about how the social world works.

10:35 And so it really raises this question of how can we design tools and build culture in a way that actually helps us better calibrate our model of who we are, what society is that can actually eliminate some of these statistical biases, not just for our own behaviours and beliefs so that we can be agents of equality, but increasingly so that the AI systems that are gaining ever more power can actually also build an accurate model of the world and stop perpetuating and reinforcing these biases and can maybe even help us embody the better angels of our nature.

Questions about this episode

Who speaks in "Would ChatGPT hire you? Your age and gender matter"?

Shamini Bundell (Host), Douglas Guilbeault (Guest).

What is "Would ChatGPT hire you? Your age and gender matter" about?

What is mentioned in this nature video episode?

5 people, works and ideas are mentioned, including Douglas Guilbeault, Shamini Bundell, Google Images, IMDb, US Census. Each one on this page links to its own page with every episode that mentions it.

How long is "Would ChatGPT hire you? Your age and gender matter"?

The episode runs 11 min; the transcript is about 1,823 words in 18 timestamped sections.