Your Job Through the Lens of Artificial Intelligence

Over the past few months, journalists have been raving about AI-generated art. They talk of the magic of machine learning, and of companies launching tools that create images based solely on textual descriptions given by the user. MidJourney is one such tool, and it works wonderfully. I’ve been able to create images of futuristic cities, unicorns in the clouds, and french philosophers boxing one another.

We however cannot forget that artificial intelligence offers only a reflection of the data it’s been fed. And that got me thinking: can we learn about the way we see jobs by studying what AIs think those jobs look like? I opened careerprofile.info, fired up MidJourney, and got to work.

Below are the images algorithmically created by entering the name of major roles. They have not been retouched, and appear as they did in the MidJourney tool. You may however not be able to exactly reproduce the results as tool is continuously evolving.

“Registered nurses”

Unsurprisingly, the algorithm sees nurses as (white) women. I’ve also made a search for “doctor”, and the images created all came up masculine (or with blurred faces). This is in line with how society views these roles.

Gender biases are important to acknowledge as AI is increasingly used in healthcare; a symptom assessor might make a mistake on a diagnosis if a black male nurse uses it, for example. Most of the data the algorithm would have been fed would not apply to him.

The red uniforms are particularly interesting as they also appear in variations of this search. In the UK, the more senior nurses wear red. It’s a helpful hint as to where the algorithm may get its “inspiration” from.

“Customer service representatives”

The algorithm tends to struggle with faces when they are not textually described. I however find it fascinating that the people represented here are mostly faceless, whereas nurses were not. This is probably because their job is a lot more personable. It’s an accurate representation of how we often picture customer care representatives: faceless.

We also see mostly women being represented. MidJourney manages to link the job description to phones, which is a nice touch. Finally, the clothes pictured are fairly uniform, which also underlines how we see the brave people answering the phone for companies.

“Truck drivers”

These are fairly nice, detailed pictures, even if the trucks are looking a bit odd. The figures pictured seem to be mostly male, in line with the fact that most truck drivers are indeed male. They are shown with their trucks outside, on what looks to be rest areas after a rainfall, which is interesting. None of them are pictured driving the trucks.

What I find most fascinating, however, is that the algorithm decided to picture these figures as fairly large. This is in line with how I imagine truck drivers, showing that AI does indeed reflect our imagination.

“Executive secretaries and administrative assistants”

This search is fascinating, because the AI got a bit confused. If you type “Secretaries and administrative assistants” or “secretary”, the usual stereotype of a woman by a cluttered desk in an office appears. And we see a bit of this here. However, it seems the word “executive” made 2 things happen : more men appear in the image, and all the people are represented wearing a suit an tie.

This is a lesson that all machine learning aficionados learn in time : language is difficult, and algorithms tend not to understand all of its nuances. English is simple — who knows what such results may yield if we create algorithms that try to understand more complex languages. Language matters.

“Elementary School Teachers”

The fact that this, too, is a gendered role is not surprising (even though 33% of primary school teachers are male). A “Professor” search, on the other hand, only shows males. As with previous pictures, facial features are not very detailed.

There are a few things to appreciate here. Firstly, the algorithm clearly understood that a teacher needs a classroom and students. It’s contextually very good. It appears that the students/teachers may be wearing some form of uniform. This, once again, brings the UK to mind.

“Maintenance and repair workers”

MidJourney is good with blue-collar workers, that much is true. Their tools? Not so much. I don’t know what they’re carrying, but I’m pretty sure you won’t find a nondescript rod in any toolbox, orange or not (seriously, how much does AI love the colour orange?).

There is a good level of detail : I would expect such jobs to require high-visibility jackets and hard hats.

What I did not expect, however, is the look and feel of the room the workers are in on the first two pictures. It’s nearly identical to the room the elementary school teacher and nurses are in. Maybe the algorithm is not as imaginative as it seems? A fun question to study further.

“Accountants and Auditors”

I’m an office worker. Office workers have no face. I’m OK with that.

Not much else to say here. Accountants need paper. They often travel in pack, like gazelles. They wear ties and don’t like to be startled. Here’s a picture of a gazelle wearing a tie.

“Carpenter”

Carpenters work with wood. That’s accurate. I however don’t know if that necessarily needs to happen in a post-apocalyptic slum. The AI seems to think so. Who am I to argue. It gives the whole thing a great “last man in the world” feel.

An interesting comparison with earlier pictures : carpenters wear hard hats, but no high visibility jackets. Theories as to why are welcomed.

“Lawyers”

I have to say I laughed out loud. What are the odds that Trump has had so much legal troubles that when an algorithm thinks of lawyers, it also sees messy blond hair and a red power-tie? Unlikely, but amusing nevertheless.

These specific lawyers are all wearing robes. Attorneys only wear robes in criminal proceedings and in all proceedings before the Constitutional Court, the Supreme Court, and the Supreme Administrative Court. So I guess MidJourney isn’t messing about with the law. Except when it comes to images rights…

“CEO”

Speaking of Trump… is that him on the bottom left? Or have I been reading too much news lately? In any case, the business card is a nice touch.

Regardless, there is plenty to hate about how AI views CEOs. They’re all clearly male, and literally tower over the world. The CEO pictured at the top left even has a group of followers behind him. They are no doubt waiting for his insights.

If that’s how AI sees CEOs, we may need to throw the whole thing away and start again.

“Entrepreneur”

AI-generated entrepreneur pictures are very similar to the CEO ones, but the differences are telling.

They’re clearly high up in the world (in a skyscraper, in the clouds…), but are not necessarily towering like giants. Most interestingly, they have a halo, which says a lot about the cult we dedicate to entrepreneurs in Western society.

Most entrepreneurs, however, are small shop owners, or run a small service company. There are 31 million of them in the US, and 27% are women. The AI is not accurate, even as it reflects our collective imagination. This is a lesson we should keep with us as we think about the role of AI in our lives.

“Electrician”

Once again, we see that blue-collar roles are a lot more detailed, as they tend to be easier to imagine. Electricians work with electricity, and that’s what we’re seeing here.

The hard hats have made a comeback… and one seems to be protecting a man with a light-bulb for a head. Which is new. You truly are what you do, I guess.

“Recruiter”

Hey, recruiters… Are you guys OK?

The 13 jobs above represent 47 million people. What did we learn? AI loves orange, all rooms are the same, blue-collar jobs are easier to represent? Sure.

But there is a deeper truth here. AI fully represents our vision of things, not the statistical reality. This may quickly become problematic as tools such as MidJourney becomes more widespread.

At the end of the day, technology merely holds a dark mirror to society, its triumphs and its inequalities. Maybe, just maybe, the best thing to come from it isn’t a better understanding of our engineering prowess, but rather a better understanding of ourselves.

Good luck out there.

Also Published Here.