Women are swimming in the slow lane, laps behind men in the AI race

How can we support a rising tide to improve adoption?

AI isn’t the next big thing.

It’s today’s big thing.

It’s the knife cutting the shapes of entire industries and job roles today.

We hear these dual narratives on heavy rotation:

AI is the cure for our hopeless productivity at work (capitalist silver bullet meets Jack and the Beanstalk) or
AI is coming for our jobs and yours could be next

Both are deeply disempowering. And yet both are concurrently true.

For International Women’s Day, I return to a research topic from the AI tech hinterlands of mid-2024:

Is AI empowering women to level up, or a pin poking holes in existing inequalities?

It was apparent from my research that the future was not evenly distributed. Women were lagging behind the opportunity, like many prior waves of tech adoption. I wanted to understand why that may be. Pulling at the threads, I wrote 7 Gender Gap (and how to fix one of them) and these social and video resources.

Seven months later, we’re living in a parallel universe. DEI initiatives have been thrown off the table for many US corporates as boards desperately want (or need) to curry favour from the autocrat a l’orange.

Big tech firms developing AI like Meta’s Zuckerberg bemoan a dearth of “masculine energy”. Google scaps its DEI goals, then Sergey Brin, *checks notes* the world’s seventh richest man, champions a 60-hour work-week for developers at *checks notes* the world’s third-most valuable tech brand, to develop AGI (artificial general intelligence). This futuristic tech would *checks notes* make many roles obsolete - starting with you, developers. Buckle up, shirkers!

None of these relentless and exhausting narratives or power grab by the brolicarchy (tech oligarchy) is good for equity or supporting women in tech’s careers. And building equity in AI is a challenge beyond today’s current friction.

I’ve updated the data to check women's progress in AI adoption, knowing the education's leaky bucket will take longer to fix.

Good news: There’s more knowledge now about the uptake and the impact of intersectionality on AI adoption.

Bad news: Women and other minorities are swimming in the slow lane. The negative impact AI now has on women’s lives and careers is deepening.

Last week I gave a careers talk for four classes of children aged 8 to 12. My goal: to convince them a tech career is possible even if you aren’t good at science and maths and like to wear a garish pink hat. More on that shortly. I asked the girls in each class if they saw tech as a job for them. The 8- and 9-year-olds were open to it. But after 10, more saw tech as a “boys’ thing”. We must start early.

Back to the gaps and what we can do to reduce them.

Gap 1: Technology education

The first hurdle is access to technology. In low-income countries, only 20 percent of women have internet access. Those women online lag behind men in digital and AI skills.

The gender gap in tech starts early. World Economic Forum research shows four times more men than women graduate with ICT degrees.OECD data shows twice as many young men (16 to 24) can program compared to women.

In higher education, the share of women earning AI and computer science PhDs in the US and Canada is stuck at around 20 percent. That means fewer women in the pipeline for high-level specialist roles. Some markets are dialing up support: Rwanda run initiatives to improve access to ICT and digital for women.

Gap 2: Working in AI roles

In the workplace, the gender gap persists. UNESCO’s latest data shows that women make up just 20% of technical roles in major machine learning companies, 12% of AI researchers, and 6% of software developers.

The Alan Turing Institute calls it bleaker: only 22 percent of AI and data science professionals are women, and they’re more likely to be in lower-status jobs.

In 2019, women made up just 18% of C-suite leaders in AI companies and top startups. This ties to the narrow education pipeline, but the slow progress of women already in AI towards leadership is a concern. There are green shoots: Women's representation in AI is growing faster than men’s, albeit from a low basis.

Using R&D publishing as a proxy for talent, the gaps are clear. Men produce five times more sole-author papers (55%) than women (11%). Women in the Global South are leading the way - Brazil, India and China rank among the top countries for AI papers with at least one female author.

This mirrors a broader trend. Edelman’s Trust Barometer 2024 shows AI adoption is highest in China and India and across Africa and Asia. The most forward-thinking action is happening where many global leaders and investors aren’t looking.

Gap 3: Stunted progress for women in AI

Congratulations! You’ve pushed through unconscious and systemic bias to land an AI role. But the battle isn’t over.

Women in AI and data science have higher turnover and attrition rates than men. They’re more likely to leave their job or the industry sooner. Then there’s the (so-called) imposter gap. Men self-report more AI skills on LinkedIn than women. They’re more likely to be confident and get hired. 🫨

The pandemic made inequalities worse. In 2021, women in tech were twice as likely as men to have lost their jobs. Many took on the bulk of household work.

More than 70% of women say they’ve worked at a tech company where ‘bro culture’ dominates. Zuckerberg’s “masculine energy” vibe is triggering for women living through this. And don’t get me started on the tired argument that you’re not a woman in tech unless you have a STEM degree and you’re in a technical role. You can be. You are.

Gap 4: Reduced access to services through AI

Here’s a big cauldron of chaos: gender representation in AI data. Here are a few examples to illustrate why AI is undeserving women.

In 2019, Genevieve Smith discovered algorithmic bias when she received a lower credit limit than her husband despite her better credit rating and equal income. A 2021 study found gender bias in nearly half of analyzed systems, causing women and non-binary people to experience worse voice recognition and unfair resource allocation, including in hiring.

UNESCO’s latest study shows LLMs perpetuate bias, associating men with higher-status roles like doctors and women with lower-status positions like cooks, while exhibiting homophobia and racial stereotyping.

A 2024 analysis demonstrated that single mothers disproportionately face incorrect automated benefit decisions. Anna Dent from the wondrous Careful Trouble research studio explains this is a predictable mis-prediction: "Over or underrepresented groups in training data will encode the same bias into any system."

In theory, the now-in-force EU AI Act should subject these ‘high-risk’ designation systems to strict regulations in markets that interface with Europe, but it’s early days for implementation. European regulators may lack teeth for enforcement, especially amid growing anti-regulation sentiment.

Gap 5: Gender representation in GenAI

If you ask a GenAI tool to generate an image of a surgeon, it’s more likely to show a man, and a woman for a nurse. They do this because they mirror societal patterns in the training data. More photos exist of men as surgeons and women as nurses, perpetuating these stereotypes. Synthetic AI data offers one solution but risks undermining trust and authenticity - essentially "garbage in, garbage out." Concerningly, researchers predict tech companies may exhaust high-quality AI training data by 2026.

Proof in point: My Dall-E prompt 1: “Hospital operating theatre. There should be two people: A surgeon, who is a woman, and a nurse, who is a man.” The first set, epic fail.

Prompt 2: “The man should wear a nurse's uniform. Show the woman operating on the patient and the man assisting her.”

Here’s the best of the bad bunch it gave me:

1. The nurse assisting the surgeon equally.

Gender equity surgery attempt, created with AI using Dall-E

2. We finally see some men as nurses, but why are so many nurses conducting the surgery and no surgeon?

Gender equity surgery attempt, created with AI using Dall-E

We also see gender bias propagate in discrete but disconcerting ways. See Cornell University’s study "Kelly is a Warm Person, Joseph is a Role Model". The clue’s in the title.

Gap 6: AI is more likely to shrink women’s jobs

Looking to future opportunities (or lack thereof), Goldman Sachs's jaw-dropping 2023 report predicted 300 million US and European jobs are at risk of AI automation. Digging into the details, the odds are more stacked against women. 79 percent of women work in occupations susceptible to automation, like admin support (70 percent), healthcare (76 percent), education (73 percent), and community and social services (67 percent). Male-skewed jobs like construction particularly have low levels of automation potential.

Gap 7: Women’s access and uptake of AI at work

Women are underrepresented throughout AI - from education to technical roles to adoption rates. This creates a troubling cycle: fewer women influencing AI design leads to systems that potentially disadvantage women in services, representation and job security.

Where’s the olive branch? For knowledge workers, access to AI tools at work should be a level playing field, right?

Not quite.

Despite theoretical equal access, adoption still shows gender disparity. Deloitte bullishly predict GenAI adoption will reach parity by end-2025, though this contradicts wider global trends showing double-digit adoption gaps across 12 European countries. Women demonstrate higher "technology trust" concerns about data privacy.

Several factors drive this disparity.

MaryLou Costa (an excellent journalist I’ve worked with) reported for the BBC with first-hand stories about why women may not trust bringing AI into their work. Hayley Bystram, the founder of a match-making agency, rejected using AI to write members' profiles: "It would take the soul and the personalisation out of the process, and it feels like it's cheating, so we carry on doing it the long-winded way."

Psychologist Lee Chambers notes,

"Women are already discredited and have their ideas taken by men. Having people knowing that you use an AI might play into that narrative that you're not qualified enough."

This aligns with Microsoft's AI at Work survey showing that 1 in 2 people don't openly admit to using AI, fearing they'll appear replaceable. Another barrier is practical:

"Working women with kids don't generally have time to experiment out of purely intellectual interest; there needs to be a clear, practical application."

This was a CEO quoted in BCG’s survey of 6,500 tech workers. Role-specific patterns emerge. Senior technical women outpace men in AI adoption by up to 16%, while junior technical positions and non-tech roles show the reverse: women trail men by up to 21%.

Yet, women comprise only 28% of Coursera's global enrollments. Randstad’s study shows 42% more men than women report AI proficiency. Ageism is another factor with less experienced workers gaining greater access: just 22% of Baby Boomers and 28% of Gen X were offered AI upskilling compared to 45% of Gen Z and 43% of Millennials.

Women are 16% less likely than men to use AI tools in equivalent positions. Harvard Business School research confirmed that women’s adoption averaged 25% lower than men (10 to 40% depending on industry and country). Even with support, adoption gaps persist: In a Kenyan entrepreneurship programme, 13% fewer women used ChatGPT access when it was offered.

The business implications are massive. Organizations that scale responsible AI programmes outperform those focused solely on tool capabilities. Dedicated training and experimenting time could help close the adoption gap and improve business outcomes. Hooray for that.

Gap 8: Intersectional barriers to AI adoption

Here’s a new gap: there’s new research into the compounding impact of intersectionality – how a person’s identity layers up to deepen inequality. UNESCO's study on AI and gender shows that women of colour earn less than white women in tech firms. As end users, women with disabilities face a greater digital divide with cost and usability barriers with AI interfaces that aren't designed with accessibility in mind.

Pew Research Center's report on US workers' exposure to AI identifies rural women and those over 50 as the least likely to use AI. Access, training and confidence create barriers. Language plays a critical role too: non-English speakers face barriers as most AI systems perform substantially worse in other languages which have less training data.

These factors combine to create a ladder of access. Privilege at each level (gender, race, ability, geography, language, age) gives you a different start rung up the ladder. The further you are from being the Silicon Valley tech bros designing AI, the less likely you are to reap its benefits.

How can we improve women’s success in AI at work

Reviewing this research again, it’s bleak inside ten buckets of bleak. But I’m a bucket-half-full gal, and my mission is to make AI work for everyone by starting with business leaders. So, let’s get practical.

How you can improve women's uptake of AI depends on who you are.

1. You’re a business leader in a tech or people management role

When introducing AI tools to your team, don’t say, “Have a play around and experiment”. Do say, “We have a dedicated training and support programme, and we’re giving you time and access to experts to learn.”

It’s not a one-and-done session in a damp conference room on a Tuesday afternoon; it’s a continuous annual programme with refresher training and dedicated time for implementation and review. Skip the late-night scrums with beer and pizza. Make it inclusive. Bring nice biscuits.

They say if you want to get something done, ask a busy person. That person is often a woman. She’s busy because she’s taking on someone else’s work, a “side of desk” (😝) project AND a load of responsibilities at home.

Craft-based meme by Susi O'Neill. GenAI image on paper, tape and pen

This is too important an opportunity to miss. Once girls and women have skills to build from, they often outperform men. Don’t make systematic bias worse by letting her fall behind with AI.

Make AI training inclusive for everyone. But one-size-fits-all AI training programmes often fail women and diverse team members.Offer differentiated learning paths depending on role and experience, with practical application time. Peer mentorship creates a culture of psychological safety, encouraging experimentation within a safe environment.

To feed the talent pipeline, support initiatives to get more girls and young people into AI careers. AI4ALL is a US non-profit that increases diversity in AI by working with young people in under-represented communities. It offers some free high-school level teaching resources. In the UK, Stemettes offers courses and online inspiration for women and non-binary young people to build skills and careers in science and tech.

For professionals and teams, there are many marvellous trainers in this space, like Heather Murray from AI for Non-Techies, who delivers free online resources and affordable subscriptions for ongoing learning. Whatever your business size or challenge, there will be a trainer who can help.

The carrot: get it right, and your people will be happier and more efficient. 🥕

The stick: The EU AI Act mandates AI Literacy for any organisation using AI systems. Workers using the systems must understand the harms and risks. This is the base level of your broader AI education and academy programmes. I wrote previously about the Commision’s guidance on AI LIteracy best practices

2. What to do if you’re that busy woman

If your company doesn’t have a policy or AI tools, ask for it. If you’re feeling left behind and want to get started, sign up for resources like Ben’s Bites newsletter and learning community. For marketers, the Marketing AI Institute gives a slow drip feed of regular learning via webinars and reading.

Join the slow AI movement. We don’t need all the information at once right now. You need a baseline of knowledge about how AI works and how to use tools. You can deepen your knowledge with sector or role-specific training and networks.

Subscribe to Rethinking the Hype Cycle for a slower monthly digest of essential things happening in AI and tech.

If you feel you or your team are still stuck at first base, don’t panic. Most are still there, at best. You haven’t missed the express train. You’re just in time for the slower one going precisely to your destination.🚆

3. What to do if you’re an AI guru, tech bro or hustler selling an e-book of 10,000 AI prompts and the hacks to automate your job in 30 minutes a week

Put a sock in it. Please.🧦

Play the long game, not the slow game

The World Economic Forum (WEF) predict it will take 134 years to reach gender parity globally. And as I updated this, I was saddened but not shocked that the gaps have gone up by 3 years. Again.

As AI shatters and reforms our working world, if we don’t take action, women’s careers and economic opportunities will reverse. WomenInAI and AI Sweden’s report gives voices to women working in AI and practical advice on how to make AI teams more inclusive.

These are my eight gaps and one fix. What have I missed? Add your thoughts in the comments on good initiatives and how we can fix this.

EVA is working to make AI work for everyone by helping business leaders plan and communicate how to grow sustainably. Get in touch if I can help shape your thinking.