25 women pioneers who shaped AI's evolution

1840s

Augusta Ada King, Countess of Lovelace, also known as Ada Lovelace, published the very first algorithm while organizing an extensive set of notes for Charles Babbage's vision of an Analytical Engine, a steam-powered computer that was never built because Babbage ran out of money. King also found other ways to share some of her ideas, such as padding her translation of a mathematics paper by Italian mathematician Luigi Menabrea until it tripled in size. She waxed poetic about how machines might one day generate music, long before multimodal GenAI.

1940s

Jean Jennings Bartik, Betty Snyder Holberton, Kathleen McNulty Mauchly Antonelli, Ruth Lichterman Teitelbaum, Marlyn Wescoff Meltzer and Frances Bilas Spence translated a poster-size wiring diagram into what was arguably the first computer program for a general-purpose digital computer. During World War II in 1945, the U.S. Army needed experts to program the most sophisticated electronic calculator of its day, ENIAC, for predicting artillery trajectories. The Army turned to its pool of human computers, women tasked with performing this complex and demanding process manually even though it was considered clerical work at the time. In the process, these six women, known as the ENIAC programmers, invented direct programming, looping and conditional logic. When the Army finally unveiled the classified project, the women were depicted in newspaper photographs without credits, so readers assumed they were models. They weren't forgotten, however. Researcher Kathy Kleiman assembled and promoted their full story 40 years later.

1950s

Grace Murray Hopper introduced a method to translate mathematical code into machine-readable instructions in 1952, a feat her colleagues considered impossible. The A-0 System compiler laid the foundation for all future programming languages. Hopper believed computer programs could be written in something more akin to human languages than mathematical equations. This insight paved the way for abstract and human-interpretable techniques that have evolved into agentic coding and prompt engineering. She also wrote the first computer manual, A Manual of Operation for the Automatic Sequence Controlled Calculator, which described how to operate the Mark I calculator and was considered the first extensive treatment of how to program a computer. Hopper joined the Naval Reserves during World War II and eventually achieved the rank of rear admiral. She was posthumously awarded the Presidential Medal of Freedom in 2016.

1960s

Frances Allen built on Hopper's early work, laying the foundation for optimizing compilers and automatically executing code in parallel -- for example, graph structures representing complex code execution flows. Allen's ideas support the code compilation and parallel execution required to scale AI across larger data centers. She became the first woman to receive the Turing Award in 2006 for her body of work.

Jones planted the seeds for the automated search engines that drove the growth of the internet.

1970s

Karen Spärck Jones created early natural language processing techniques in a landmark 1972 paper on the significance of statistical analysis for concept mapping -- planting the seeds for the automated search engines that drove the growth of the internet. Her work also demonstrated how statistical methods could extract meaning from text, which helped bridge the traditionally separate fields of linguistics and computing. Later work focused on automatic text summarization. Jones also explored the limits of purely statistical approaches to interpreting information, foreshadowing today's concerns about overreliance on ungrounded data ingested by large language models (LLMs).

Dana Angluin introduced what became the field of inductive learning in her 1976 University of California, Berkeley doctoral thesis, which demonstrated how machine learning can make sense of complex systems through guided trial and error exploration. This approach also plays a key role in finding helpful patterns buried in noisy data. Angluin later helped start the Workshop on Computational Learning Theory in 1988 to guide AI research long before recent breakthroughs in neural networks. The core algorithms are widely used today in interactive learning, active learning and human-in-the-loop AI systems.

1980s

Ruzena Bajcsy coined the term active perception, an essential element of embodied AI and autonomous systems research. At the time, most machine vision and robotics systems used sensors that passively received data. Bajcsy's active perception research found better results could be achieved by teaching the sensors to actively look for salient information to build a worldwide model. A similar approach shapes ongoing research into how active exploration might improve AI by mechanically dumping in data sets and hoping for the best -- a distinction that matters most in safety-critical systems like surgical robots and autonomous cars.

1990s

Isabelle Guyon and associates invented support vector machines to accurately and mechanically process large data sets. These techniques are widely used on their own and still play a critical role in organizing data for training LLMs. Guyon also played a leading role in developing feature selection techniques for identifying which variables matter most to a given question buried in the data. Her 2003 paper on variable and feature selection, with more than 15,000 citations, observed that the quality of data fed into an AI model is just as important as the model's architecture.

Corinna Cortes and colleagues in 1995 worked to extend the use of support vector machines to support vector networks, which proved a better match for noisy data. This work introduced the concept of soft margins that could better account for the various problems that arise when collecting real-world data. It also created avenues for research in handwriting recognition, medical AI and text classification and proved foundational in data mining aspects of most modern AI pipelines.

Rosalind Picard coined the term affective computing in her 1997 book of the same name, introducing many ways AI tools could recognize, interpret and simulate human emotions. Her early work focused on improving human-computer interaction. These techniques today support AI's ability to infer emotion from what people write, how their faces move and the quality of their voice. They're widely used to improve conversational AI systems and engagement.

2000s

Latanya Sweeney was the first Black woman to earn a Ph.D. in computer science at MIT, where she explored the limits of what were considered anonymous data sets. She introduced a new mathematical framework, k-anonymity, which could statistically quantify the privacy implications of data sets when considered on their own or when combined. Her research, for example, revealed that 87% of Americans could be uniquely identified by a combination of their ZIP code, birth date and sex. Controversy ensued when she publicly deanonymized the medical records of the Massachusetts governor from what everyone thought was anonymized public data. Sweeney's work influenced privacy frameworks like HIPAA.

Barzilay discovered novel techniques to better analyze mammograms for earlier breast cancer detection.

Regina Barzilay wrote her 2003 doctoral thesis on a new approach to automatically summarize news stories, which shaped the development of text-generation techniques. She also discovered ways to use the extensive body of research on more common languages like English to infer the structure of less studied languages. Barzilay pivoted her focus to medical AI research after she was diagnosed with breast cancer in 2014 and discovered novel techniques to better analyze mammograms for earlier breast cancer detection. She also pioneered methods of using AI to discover new drugs, such as a novel antibiotic in 2020.

Daphne Koller and her colleagues built what became a Rosetta stone for mapping different approaches to probabilistic graphical models with the publication of Probabilistic Graphical Models: Principles and Techniques in 2009. Engineering, scientific, risk management and mathematical disciplines used graphical models but with different languages, approaches and use cases. The book's framework helps map probabilistic approaches like Bayesian networks and Markov models onto larger classes of problems across AI pipelines. Koller later founded Insitro in 2018 to commercialize emerging techniques for exploring how genomics data could predict and test the potential efficacy or harmful effects of new drugs earlier in development.

Fei-Fei Li was inspired by thenrecent progress in using machine learning techniques for automatically labeling text and wondered if they might help improve machine vision research. Machine vision was making minimal progress then, partly because no one had found a method to label numerous images. Amazon's Mechanical Turk service had recently gone live, so Li received a grant to hire a team to label over 14 million images. She then organized a contest, 2012 ImageNet, that inspired many entries, including one dark horse, AlexNet, that used deep convolutional neural networks in an entirely new way, dramatically outperforming the other contestants by reducing the error rate by nearly half. That planted the seeds for the deep learning revolution using neural networks that have reshaped the trajectory of AI research.

2010s

Maya Ackerman studied how emerging AI techniques could be used for generating text, music and images in a structured way in 2014. Around that time, variational autoencoders and generative adversarial networks first emerged a few years before diffusion models and the transformer architectures. Ackerman cofounded WaveAI in 2017, which pioneered tools for programmatically creating lyrics and melodies, some of which the company says fueled chart-topping songs. In 2025, Ackerman published Creative Machines: AI, Art & Us, which traces the history of creative AI.

Kate Crawford co-founded the AI Now Institute at New York University in 2017 to study the social implications of AI. She organized these insights into her 2021 book, Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence, which invited readers to look beyond the algorithms and consider the mines that produce raw materials for the chips, the workers labeling data in what she described as sometimes sweatshop-like conditions, and some of the privacy implications of large-scale data mining and emerging AI business models.

Margaret Mitchell thought AI systems should come with the equivalent of nutrition labels. She, along with collaborators, introduced a framework of model cards for capturing information describing support for different use cases, performance across different populations and known limitations. Mitchell founded and co-led Google's Responsible AI and Human Centered Technology (RAI-HCT) team to develop model cards. She departed Google in 2021 amid a widely reported dispute over the company's handling of AI ethics research. She's now chief ethics scientist at Hugging Face, focusing on improving ethical frameworks for the open source AI ecosystem.

2020s

Timnit Gebru co-led Google's RAI-HCT team, where she explored some of the ethical implications of emerging AI tools. Her earlier doctoral research at Stanford demonstrated algorithms that could infer a neighborhood's voting patterns from the types of cars visible in street-level map data. A widely publicized dispute ensued after she declined to retract a 2020 paper analyzing the potential social risks of LLMs. The paper analyzed environmental costs and the risks of bias that have grown more acute amid the GenAI boom. Gebru eventually parted ways with Google and later co-founded several organizations, including Black in AI and the Distributed AI Research Institute to foster AI research and collaboration.

Daniela Amodei joined OpenAI in 2018 and served as vice president of safety and policy, shaping the company's approach to AI safety. In late 2020, she and her brother Dario Amodei, along with five other colleagues, left OpenAI, and in 2021 founded Anthropic to pursue a different approach to AI safety and commercial deployment. She's now president of Anthropic, which has been making headlines lately for its Claude Mythos LLM as well as disputes with the Pentagon over the military use of Claude.

Mira Murati joined OpenAI as vice president of applied AI and partnerships in 2018 after a stint as product manager at Tesla. She eventually became OpenAI's CTO in 2022 and served briefly as interim CEO during the November 2023 board dispute that removed and rapidly reinstated Sam Altman. Murati played a central role in AI product development at the company, helping to steward products like ChatGPT, DALL-E, Codex and Sora. She left OpenAI in 2024 and launched Thinking Machines Lab in February 2025, which is developing tools to create custom frontier AI models.

George Lawton is a journalist based in London. Over the last 30 years, he has written more than 3,000 stories about computers, communications, knowledge management, business, health and other areas that interest him.