Women get about one in four of the new jobs in AI, where in other kinds of work they get one in two. The gap also shows in what they do once they are in: women in AI are more likely to hold low-paid work such as labelling data, and across AI jobs the typical man earns $45,000 more than the typical woman.
A new field has no credentials, so employers pick by stand-ins
Programming had the problem AI has now. Nobody had years of experience, so employers needed some other way to choose, and in the 1950s and 1960s they chose aptitude tests and personality profiles that screened out women. The effect lasted for decades. Women's share of computer science degrees rose until it reached 37% in 1984, and after that it was the only science field where women's share kept falling for years.
AI has no tests like that. Women who work in it say companies are hiring at great speed and finding people the way they always have, through referrals. A referral is a stand-in too: someone the company trusts vouches for a person. It works well, which is why it sticks, but people refer people like themselves, so every early hire becomes the door for the next one. If a quarter of the early hires are women, referrals will hold the share near a quarter for a long time, and the pay gap goes with it, because the best-paid jobs are the most lopsided.
Most readers will meet this choice inside their own company
You do not have to work at an AI lab to be part of this. One of the fastest-growing AI jobs is the forward deployed engineer, whose work is helping a business put AI to use, and it is now the third most common AI job in postings. Every company that moves a department to AI needs someone in that seat, whether or not the title exists.
Take a finance director at a hospital group who wants AI to check insurance claims before they go out. She needs one person to lead it. If she asks who is good with this kind of thing, she will hear the names of the people she talks to most. If she announces the role and asks every team member to bring one task they would hand to AI and say how they would check the answer, she gets a different set of candidates. The person who knows why claims keep coming back is often better at telling an AI what to do than someone who knows how models work, and a year later that person has a record in the kind of work employers now pay extra for.
Women hold most of the jobs that AI is likely to change first, customer service among them, so they also hold much of the knowledge those rollouts need. Leaving them out of the lead roles wastes that knowledge. It also puts them in the group that absorbs the change and not in the group that is paid for managing it.
Diversity programs used to be the answer to this, and many companies have cut them back. That leaves the choice with the person who picks.
If you are the one who knows the work, do not wait for the role to be posted. Write down one task from your week that AI could do and how you would check its answer, and take it to your manager. A manager who has that in hand has an easy name to give.