AI's Gender Gap Is Becoming an Ownership Gap

Between 2002 and 2018, the share of US patents involving artificial intelligence more than tripled. The USPTO's AI Patent Dataset, in its 2023 update, classifies 15.4 million US patent documents from 1976 through 2023 to identify which contain AI-related technology. AI patenting is one of the most consequential and rapidly expanding patent terrains of the 2000s-2020s.

Women's participation in this engine lags. USPTO researchers have characterized women's representation in AI patenting as "notably low" relative to women's already-low baseline representation across all US patents. Cross-country evidence from China's WTO accession shows that the gap is policy-responsive, but in the absence of policy levers it compounds.

That compounding has a name worth being precise about. The gap in AI inventorship today translates into a gap in AI patent assignment tomorrow, which translates into a gap in AI commercialization wealth over the following decade. The gender gap in AI inventorship is becoming a gender gap in technology ownership.

This article presents published research data for educational purposes. It is not legal, policy, or investment advice. Patent gender statistics depend on name-based inference and methodology choices: see How we measure.

The AI Patent Dataset and what it measures

The USPTO's AI Patent Dataset (Giczy, Pairolero, & Toole, Journal of Technology Transfer, 2022; 2023 update) is the canonical source for AI patent data. It applies machine-learning classifiers across eight AI subcomponent technologies (machine learning, natural language processing, AI hardware, computer vision, knowledge processing, AI planning, evolutionary computation, and speech) to identify AI content. The 2023 update covers 15.4 million US patent documents from 1976 through 2023.

Subsequent USPTO research extended the dataset to study inventorship dynamics:

• Giczy, A. V., Pairolero, N. A., & Toole, A. A., "Discovering value: Women's participation in university and commercial AI invention," Nature Biotechnology 42: 26-29 (February 2, 2024).

The Giczy, Pairolero, & Toole (2024) finding most relevant to ownership: women's participation in patenting (both in AI and in non-AI fields) is associated with patents of higher economic value and with more diverse teams. The implication runs in both directions: women-involved patents are disproportionately valuable on average, and that disproportionate value is itself part of what the system is leaving on the table when women are under-represented.

The AI participation gap

Published USPTO research characterizes women's AI patenting participation as growing but uneven, and lower than the already-low US baseline rate (12.8% for all US patents in 2019). A precise side-by-side comparison ("AI WIR = X% vs baseline WIR = Y%") is not cleanly stated in any single USPTO publication we could verify; the most precise public statement is in the USPTO's February 2024 press release for Giczy, Pairolero, & Toole, which describes AI as "a sector where more can be done to diversify, especially in biotechnology AI."

Several structural reasons compound the headline gap:

1. Pipeline composition. Women earn approximately 21% of US computer science bachelor's degrees and a smaller fraction of computer science PhDs. AI patenting depends heavily on doctorate-level researchers; the pipeline is thinner than for general patenting.
2. Capital concentration. AI research is concentrated in a small number of large industrial labs and well-funded universities. The same labs that have historically had lower women representation in senior R&D roles are now driving most of the AI filings.
3. Employer-assignment dynamics. Most industrial AI patents are assigned to the employer at filing. The wealth created by AI patents flows to the assignee, not to the inventors. A gender gap in AI inventorship therefore translates more directly into a gender gap in career advancement and future inventor designation than into immediate personal wealth, but the long-run aggregate is the same.

The China case: evidence that the gap is policy-responsive

The strongest cross-country evidence that the AI gender gap can move quickly under policy pressure comes from Agrawal, Rathi, Chatterjee, and Higgins (NBER Working Paper 32547, 2024). They studied the effect of China's 2002 WTO accession on Chinese AI inventorship by gender.

After WTO accession, relative to comparison countries:

• Chinese patents with at least one female inventor rose 95%.
• Individual female AI inventors in China rose 111%.
• Gains were concentrated in less-complex AI applications rather than at the frontier.

The mechanism the authors identify is institutional: WTO accession came with strengthened IPR enforcement, which incentivized university and SOE filings, which in turn brought more women into the formal patent record because those institutions had more gender-balanced researcher pools than Chinese private industry.

Two takeaways for the US context:

1. The AI gender gap is not destiny. Institutional levers (IPR strengthening, university filing incentives, public R&D funding patterns) can drive double-digit percentage gains within 5-10 years.
2. The gains tend to be at the less-complex end first. Frontier AI patenting (the most strategically valuable) is the last category to see gender rebalancing. This matches the US pattern, where biotechnology AI shows the most progress and frontier ML the least.

From inventorship to ownership

Patent inventorship and patent ownership are different things. Inventorship is the legal recognition of who conceived the invention. Ownership (assignment) is the legal entity that holds the resulting property right. For most industrial AI patents, inventors are employees and assignees are corporations.

The mechanism by which an inventorship gap becomes an ownership gap therefore runs through corporate compensation, equity, and seniority rather than through direct patent-revenue distribution. Three specific transmission channels:

1. Inventor designation drives career advancement. Being named as inventor on consequential patents is a senior-track signal at most large tech employers. Women under-represented as AI inventors are under-represented in the senior-track signaling pool.
2. Founder credibility runs through inventorship. Founders of AI startups disproportionately come from senior IC roles at large AI labs. Women under-represented in those roles are under-represented in the founder pool. EPO's March 2026 Observatory study found that only 13.5% of European deep-tech startups holding European patents have at least one female founder.
3. Venture capital favors patent-rich teams. AI startups with strong patent positions raise larger rounds at higher valuations on average. The cumulative effect over a decade is that female-founded AI startups raise less capital, hold less equity, and produce less downstream wealth.

None of these channels involves the patent system mistreating women. All of them involve the patent system reflecting and amplifying upstream gender dynamics in employment, capital allocation, and founder selection. The patent record is the audit trail of the underlying pattern.

Why this matters in the 2020s specifically

Two facts make this decade different from prior decades of patent gender gap analysis:

AI patenting compounds faster than baseline patenting. AI patents have stronger forward-citation networks, larger licensing markets, and higher commercial multiples than the average US patent. A gender gap in AI inventorship therefore translates into a wealth gap faster than the historical baseline rate would suggest. The 12.8% baseline US WIR took 50 years to reach. The AI ownership gap will set its structural shape in roughly a decade.

The corporate concentration of AI is unprecedented. A small number of large US firms (Microsoft, Google, Meta, NVIDIA, IBM, Amazon, plus a handful of recent AI labs) account for a disproportionate share of US AI patent filings. The internal pipeline dynamics at these firms therefore set the gender composition of AI inventorship at national scale. Closing the AI gender gap is, to first approximation, a hiring and promotion problem at roughly 20 companies.

What could close the AI ownership gap

The interventions that work for baseline patenting (covered in Closing strategies that work) apply with two AI-specific modifications:

1. Pre-populated invention disclosure forms. AI research teams are often large and cross-functional; researchers downstream of the core ML work (data scientists, evaluators, applied researchers, product engineers) are frequently overlooked as inventors. The Washington University WIT pre-population approach maps directly onto AI lab disclosure dynamics.
2. Inventor-designation audits at the company level. Large AI labs could (and several USIPA pledgees have begun to) audit their inventor-designation patterns to identify systematic under-naming. The Giczy, Pairolero, & Toole 2024 finding that women's participation correlates with higher patent value is itself an argument for the audit: under-named women researchers represent value the company is failing to capture credit for.
3. USPTO examiner training on AI patents specifically. The Pairolero RCT showed that examiner-side interventions can shift grant rates by 12+ percentage points. The PNAS 2026 study showed that the gap is concentrated on unconventional inventions and on inexperienced examiners (see The breakthrough penalty). AI patents are disproportionately unconventional (they combine ML methods with novel application domains), and the AI examination corps is disproportionately newer. The intersection is the highest-leverage examiner-training target in the US patent system right now.

Sources

More from the patent gender gap series