This article is a continuation of our series of profiles on women who have shaped AI technologies and systems. For more Herstory profiles like these, subscribe to She Writes AI (free).

Subscribe now

One of the most significant advancements in modern AI was relegated to … a poster. That’s right. At the leading computer vision conference, Fei-Fei Li intended to unveil her massive labeled image set known as ImageNet. It would eventually become the foundation of modern AI. She was not granted a presentation slot but was offered a table tucked way in a corner of the Miami Beach convention center flanked by a few posters.

In 2006, as an assistant professor in Electrical and Computer Engineering at the University of Illinois Urbana‑Champaign, Fei‑Fei Li began conceptualizing what would become one of the most transformative projects in AI history: ImageNet. Fresh off her PhD from CalTech, she had a simple, yet radical realization. The explosion of images online could be harnessed to create a massive, organized visual dataset modeled on human cognition. At the time, computer vision relied on tiny, hand‑curated datasets and hand‑engineered features. The field had not yet considered scale. Li’s idea challenged the prevailing wisdom: even the most sophisticated algorithm couldn’t work well if the data it learned from didn’t reflect the complexity and diversity of the real world.

When Li moved to Princeton in 2007, a pivotal moment arrived. She met Christiane Fellbaum, a linguist and director of WordNet — a comprehensive lexical database mapping relationships between words. Fellbaum had imagined adding reference images to WordNet entries. Li imagined something far more ambitious: a massive‑scale training dataset with thousands of labeled examples for each category, something that could fundamentally transform how AI learns to see. This exchange between two women researchers resulted in a revolution.

The scale of Li’s vision struck many of her colleagues as impractical, even foolish. One of her mentors, Jitendra Malik, warned that she was leaping too far ahead of the field. As her PhD student Jia Deng recalled, skeptical researchers asked, “If you can’t even do one object well, why would you do thousands?”

Li persisted. She assembled a team at Princeton and led a massive labeling effort using Amazon Mechanical Turk, ultimately cataloging over 14 million images across 22,000 categories. When her team published ImageNet in 2009, it drew little attention.

What she did next was a stroke of marketing genius. She created the ImageNet Challenge — a yearly competition to build the most accurate image recognition model — turning her dataset into a focal point for the entire field.

In 2012, Geoffrey Hinton’s team won by a dramatic margin with AlexNet, igniting the modern deep learning revolution.

Li’s bet proved correct: massive, high-quality datasets became one of the three pillars of modern AI, alongside neural network architectures and GPU compute.

That data‑first philosophy underlies modern generative AI. Today’s language models all trace back to Li’s 2006 bet that with enough clean and diverse data, machines can learn in surprisingly human ways. Fei-Fei reoriented the field around data.

Beyond ImageNet:

Li has been a tireless advocate for equitable, inclusive, human‑centered AI. She co‑founded AI4ALL to expand diversity in AI education and launched Stanford’s Human‑Centered AI Institute (HAI), urging the field to consider not just what AI can do, but what it should do and who it should serve.

Big Tech eventually embraced her data‑centric paradigm. It’s not surprising they appreciated the brute‑force scaling potential it unlocked.

Li is proud of her contribution, but it’s telling that she has consistently warned against hype, bias, and misuse. Her own work has evolved. She founded World Labs that represents a shift toward spatial intelligence and world models that move beyond pure scale toward more grounded, efficient, cognitively inspired systems. The data revolution she sparked was never meant to be the final chapter. It was an essential step that is beginning to show its limitations. Most researchers no longer believe scaling current systems will get to AGI.

Fei‑Fei wanted AI to see the world more accurately. She did not intend to spark trillion‑dollar compute races and hyperscale wars.

Her contribution was foundational. The use, weaponization, and commercialization of that foundation were choices made by others.

Subscribe now

Share

Leave a comment

If you enjoyed this post, you might like AI Everywhere, Volume 1: How Women Are Changing The World With Artificial Intelligence. Celeste is a participating essayist covering the forgotten women of early tech and AI.

Buy Here

Also available by Celeste Garcia:

A MOTHER-DAUGHTER STORY WITH
APOCALYPTIC STAKES

Buy Here

SOURCES:

As always, the ideas and writing are mine. ChatGPT, Notebook LLM, and Copilot helped with research and accuracy.

"Godmother of A.I." Fei-Fei Li on technology development: "The power lies within people" - CBS News

AlexNet and ImageNet: The Birth of Deep Learning | Pinecone

Fei-Fei Li - Where Did ImageNet Come From?

ImageNet: A Pioneering Vision for Computers - History of Data Science

BUY NOW

AI BABY is a humorous take on the AI apocalypse— when a “perfect” new girl arrives at an elite Seattle prep school & threatens Zoey’s dominance & Ivy League dreams, her mother Erica embarks on an unhinged search for truth that pulls her into the heart of an AI empire and a fight for her daughter’s soul.

BUY NOW

Early Reviews are In!

“AI Baby explores how far parents will go for their children in a world where intelligence can be engineered—and love may be the last thing that can’t be programmed. A must-read.”

- Dana Frank, National Best-Selling Author, business luminary, and philanthropist

“Compelling, complex characters and a strong story that could be less fictional than many think…I couldn’t wait to see how it turned out…Brava!”

- Karen Smiley, Founder, She Writes AI Media

“Smart, humorous and a somewhat terrifying look at life, where we are now and what might soon be our reality. I loved every minute of it.”

- Stacey R. Campbell, Author of the Highly Rated Lakeview Series

For More Information

Share

This article is a continuation of our series of profiles on women who have shaped AI technologies and systems. For more Herstory profiles like these, subscribe to She Writes AI (free).

Subscribe now

Kauna Malgwi is a clinical psychologist who spent years experiencing the industry’s devastating psychological harms as a Facebook moderator. She suffered PTSD after prolonged exposure to graphic warfare content.

A recent post by helped me see the practical application of Timnit Gebru’s vision for the Distributed Artificial Intelligence Research Institute (DAIR) that I wrote about in my last post. Gebru’s organization is helping Malgwi turn suffering into action.

Applying her own personal trauma and training as a psychologist, Malgwi has become a leading voice for “ghost workers” — the millions of hidden laborers who prepare the data AI needs to function. These workers are often paid less than two dollars an hour to tag images, transcribe audio, and filter out graphic violence so that AI models can “learn” to identify objects and remove toxic content. This invisible workforce is the essential foundation of AI; without them, AI models would not be safe, accurate, or reliable. With DAIR’s backing, she has moved from survivor to expert, leading a funded clinical program for the entire workforce.

Awomosu’s post captures the essence of this partnership perfectly:

Malgwi is “a clinical psychologist who co-researched with DAIR while witnessing the psychological toll of content moderation work; she founded the Digital Rights and Mental Health Initiative Africa to build (…) mental health infrastructure specifically for the people the AI industry processes and discards.” Awomosu notes, “She didn’t study the problem from outside it. She built the response from (…) inside.”

Gebru founded DAIR to do more than just write academic papers; she built an organization to provide the funding, legal support, and professional networks that allow independent researchers/advocates like Malgwi to secure resources. With DAIR’s support, Malgwi developed a pilot program designed to provide various forms of professional therapy to data workers, offering a direct response to a workforce where psychological injuries are a predictable result of the labor required to train AI.

Specifically, DAIR’s ecosystem supported Malgwi in:

• Building Mental Health Infrastructure: Through her non-profit, she is creating support frameworks for data laborers.

• Launching a Clinical Pilot: DAIR provided the research tools for Malgwi’s proposal to provide various forms of therapy to workers in need, and interventions designed specifically for the risks of digital labor.

This is ethical AI put into practice. DAIR, an independent institute, provides support for a clinical psychologist to build programs and even testify at the European Parliament. Her testimony in February 2024 directly influenced the negotiations for the Platform Workers’ Directive, a piece of legislation aimed at improving working conditions and clarifying the employment status of digital laborers.

Thank you to Abi Awomosu for the inspiration for this post about Kauna Malgwi. This is a great real-world example of how Timnit Gebru’s organization, DAIR is making ethical AI a reality, not just platitudes.

Subscribe now

Share

Leave a comment

References

See Abi Awomosu’s post: What They Call 'Niche' Is the Only Thing That Scales

Read Kauna Malgwi’s full intervention proposal here: Malgwi, K. (2025) A Mental Health Intervention for Data Workers. In: M. Miceli, A. Dinika, K. Kauffman, C. Salim Wagner, & L. Sachenbacher (eds.). Data Workers‘ Inquiry. Creative Commons BY 4.0. https://data-workers.org/kauna.

Data Workers’ Inquiry | DAIR

AI BABY by Celeste Garcia arrives March 24

A humorous take on the AI apocalypse — when a ‘perfect’ new girl arrives at an elite Seattle prep school & threatens Zoey’s dominance & Ivy League dreams, her mother Erica embarks on an unhinged search for truth that pulls her into the heart of an AI empire and a fight for her daughter’s soul.

This article is part of our series of profiles on women who have shaped AI technologies and systems. For more Herstory profiles like these, subscribe to She Writes AI (free).

Subscribe now

When AI is framed as an autonomous “super‑intelligence,” Timnit Gebru argues, then the humans who design, fund, and profit from it can absolve themselves from blame. Meanwhile, the hype — whether utopian or apocalyptic — acts as a marketing engine, distracting us from the very real harms occurring today: racial and gender biases in datasets, homogeneous design teams, rampant copyright infringement, labor exploitation, and the environmental costs of scaling these systems.

I was already planning to feature Timnit Gebru for my second Herstory post during Black History Month, but the latest onslaught of Tech Bro prophecies gave me additional inspiration. They seem determined to validate Gebru’s arguments.

At the end of January, Dario Amodei published a 20,000‑word treatise on AI’s catastrophic risks, calling it the “single most serious national security threat we’ve faced in a century.” Matt Shumer’s viral X post “Something is Happening” asserts that exponential AI progress will lead to massive job displacement. And we hear the same chorus from Mustafa Suleyman, Elon Musk, and Sam Altman. They warn of existential doom — but continue to ship the systems they claim might end us — while fortifying their bunkers.

I am not dismissing the risks, and I encourage public discourse. But isn’t it ironic that the same folks sounding the alarm are also building and releasing the products they say we should fear? Shouldn’t they be the ones ensuring safety, monitoring security, and designing systems with the highest possible ethical standards before unleashing them on the world?

I often find myself annoyed with the gloom and doom prognosticators — the ones who already built fortunes developing the technologies or the ones currently raking in their billions by recklessly releasing untested models. OpenAI releases mid‑generation model updates every 1–3 months. This from a company that went through a leadership crisis in late 2023, driven in part by internal concerns that Sam Altman was prioritizing speed over safety.

While researching Gebru and her assertions, I had that “aha” moment. What she was saying was what I had been circling. By stating that AI is all-powerful — in her words, a digital god — while simultaneously warning of an existential threat, these men are abdicating all responsibility for what they built and sell.

They act as if they are innocent bystanders. Why wasn’t safety and ethics prioritized from the very beginning? Timnit Gebru has been calling out these contradictions for years, long before most people had even heard of ChatGPT. As early as 2018, she advocated for adapting longstanding industry standards to AI development. And perhaps, if Google had listened to her warnings instead of dismissing her, we might be facing fewer of the “existential threats” today.

Timnit Gebru’s Path: A Timeline of Impact

• 2004–2013: Engineering — Nearly a decade in Apple’s hardware division, designing analog circuits and developing signal‑processing algorithms for the first iPad.

• 2017: PhD Stanford — Completed her doctorate in computer vision under Fei‑Fei Li, previoulsy earned BS and MS degrees in EE at Stanford.

• 2017–2018: Microsoft Research (FATE Group) — Investigated algorithmic bias and how homogeneous developer teams distort computer vision systems.

• 2018: “Gender Shades” — With Joy Buolamwini, exposed stark racial and gender disparities in commercial facial recognition systems, with error rates up to 35% higher for darker‑skinned women.

• 2018: Joining Google’s Ethical AI Team — Co‑led the group with Margaret Mitchell, tasked with studying AI’s social impacts and ensuring the technology served the public good.

• Fall 2020: Probing the Risks of Large Language Models — Co‑authored the now‑famous “Stochastic Parrots” paper, warning about environmental costs, data bias, and the illusion of machine “understanding.”

• December 2, 2020: Forced Out of Google — After refusing to retract the paper, she was terminated, sparking global backlash.

• 2021: Founding DAIR — Launched the Distributed Artificial Intelligence Research Institute, an independent, community‑rooted organization studying AI.

Building Something Outside the Machine

Gebru warned as early as 2018 that if models were trained on “unfathomable” amounts of uncurated internet data, they would inevitably absorb and amplify the biases embedded in that data. That seems so obvious, but Big Tech didn’t listen. She argued for scientific documentation standards, transparency about data sources, and a slower, more accountable approach. But the industry chose the opposite: hype, speed to market, and hyperscale.

And because those incentives won out, many of the structural biases she identified eight years ago remain some of the most persistent challenges in today’s LLMs.

At the end of 2020, Gebru was effectively fired for doing her job when she published research on the environmental and structural harms of large language models. When leadership demanded she retract the paper that famously described these systems as “stochastic parrots,” she refused to compromise her scientific integrity. Her termination made one thing painfully clear: internal “ethical watchdogs” inside tech giants are designed to fail. You cannot hold power accountable when that power signs your paycheck.

And yet Timnit Gebru refused to be sidelined. Through the Distributed AI Research Institute (DAIR), she created an independent research space focused on investigating how AI systems shape society, documenting their harms, and developing alternatives that prioritize public accountability over growth at any cost. It offers a working model of ethical AI research that isn’t tethered to shareholder incentives — and a reminder that the direction of this field is still contested, not preordained by the companies that once dismissed her concerns.

Leave a comment

Share

Subscribe now

Subscribe to Getting Real About AI

Coming Soon…AI BABY, a debut novel by Celeste Garcia

A satirical treatment of the AI apocalypse—when a “perfect” new girl arrives at an elite Seattle prep school and usurps Zoey’s soon‑to‑be valedictorian status, her mother Erica embarks on an unhinged search for truth that pulls her into the heart of an AI empire and a fight for her daughter’s soul.

Sources:

“It’s All Marketing” - Dr. Timnit Gebru on the Smoke and Mirrors of AI Hype - Part I - YouTube

Timnit Gebru TED talk - YouTube

NAE Website - An Interview with . . . Timnit Gebru, founder and executive director, the Distributed AI Research Institute

DAIR Institute | DAIR

Gender Shades Study

Google AI researcher's exit sparks ethics, bias concerns | AP News

This article is a continuation of our series of profiles on women who have shaped AI technologies and systems. For more Herstory profiles like these, subscribe to She Writes AI (free).

Subscribe now

By any standard, Dorothy Vaughan was a brilliant mathematician and leader, and her accomplishments are all the more staggering given the racial and gender barriers she overcame. When I revisited her remarkable life with fresh eyes, something else stood out just as clearly: her vision.

At a moment when many are being told that if they don’t “upskill for AI,” they risk being left behind, it can feel like a uniquely modern problem. But is it?

More than sixty years ago, Vaughan (1910–2008) recognized the same pattern. The disruptor wasn’t AI, but equally transformative: electronic computing.

Long before she became widely known through Hidden Figures, Vaughan was doing something radical—beyond leading a team of Black women in a segregated corner of what would become NASA. She was preparing herself and her entire team for a massive technological transition she could see coming before most others did.

Before the Machines Arrived

Dorothy Vaughan’s brilliance was evident from an early age. She graduated from high school at 15, earned a degree in mathematics and education from Wilberforce University, and worked as a math and French teacher before joining the National Advisory Committee for Aeronautics (NACA, NASA’s predecessor) in 1943. She assumed it would be a temporary wartime role.

Instead, she entered one of the most technically demanding environments of the era, performing complex aeronautical calculations by hand, as a “human computer.” Because of Jim Crow segregation, Vaughan and her colleagues were assigned to the all-Black, all-female West Area Computing Unit, working in separate offices with separate bathrooms and dining facilities.

In 1949, Vaughan became NACA’s first Black supervisor and one of its very few female managers. Under her leadership, the West Area became known for its accuracy and productivity. She also had a clear instinct for nurturing talent and advocating for better conditions and pay. However, her most consequential leadership decision was still to come.

By the late 1950s and early 1960s, massive electronic computers—like the IBM 7090—began arriving at Langley. These machines could perform tens of thousands of calculations per second. To many, this signaled the end of human computing as a profession.

Vaughan saw something else. She understood that the real threat wasn’t the machines; it was not knowing how to work with them.

Without waiting for official retraining programs (which may never have arrived for her or her unit), Vaughan taught herself FORTRAN, one of the first high-level programming languages. This was not casual learning. Early FORTRAN programming was rigid, unforgiving, and constrained by the physical realities of punch cards and giant machines.

And she didn’t stop there. Vaughan taught FORTRAN to the women in the West Area Computing Unit, ensuring they could advance with the technology.

In 1958, when NACA became NASA, a formal memo dissolved the segregated West Area Computing Unit and reassigned its members to integrated divisions. Dorothy and her team could easily have been phased out. But because they were fluent in the language of the new machines, they were absorbed into the newly formed Analysis and Computation Division. Vaughan herself became a leading FORTRAN expert, contributing to projects including the Scout Launch Vehicle Program and calculations supporting early spaceflight.

Architect of the Tech Pivot and What We Can Learn from Dorothy Vaughan

Today, many professionals are told that AI will “replace” their roles—just as human computers were once told they’d become unnecessary. Vaughan’s response offers a blueprint that still applies:

• She didn’t wait or ask for permission. She took ownership of her learning.

• She shared knowledge freely and elevated her team.

• She recognized the speed of change; she was a first mover and capitalized on that advantage.

AI, like the IBM 7090, shifts where humans bring value. The work moves from raw computation to interpretation, validation, system design, and ethical judgment. Vaughan understood this instinctively.

Becoming Indispensable

Dorothy Vaughan’s impact cannot be overstated. With so many odds stacked against her, she not only prevailed but was way ahead of the curve in embracing emerging tech and how to use it.

She remains one of the most formidable minds in NASA’s history, but she also left something modern and enduring: a model for navigating technological upheaval with dignity, foresight, and care for others.

As we rush to prepare for an AI present and future, Vaughan’s story reminds us that the most powerful response to disruption isn’t fear. It’s learning early, sharing knowledge, and refusing to let new technology determine your fate.

That lesson has never been more relevant.

Share

Leave a comment

Visit Celeste's Getting Real About AI

REFERENCES

Research Partner: Google NotebookLM

Dorothy Vaughan - NASA

DPMA | Dorothy Vaughan

Dorothy Vaughan: A Hidden Figure, A Lasting Legacy

From Paper-and-Pencil Calculations to Space: How Dorothy Vaughan Became NASA’s First African American Programmer - Internet Socity NGO

History of Computer Girls, Part 3: Dorothy!

Memo from NACA Associate Director Floyd L. Thompson to all NACA Installations Regarding the Dissolution of the West Area Computers Unit | National Archives

NASA Hidden Figure Dorothy J. Vaughan (Narrated by Octavia Spencer)

This article is the first in our series of profiles on women who have shaped AI technologies and systems. For more Herstory profiles like these, subscribe to She Writes AI (free).

Subscribe now

The concept that computing machines could move beyond simple calculation to perform any abstract operation — the defining phenomenon of computer technology — dates back to the 1840s, thanks to Ada Lovelace, a pioneering mathematician and computer scientist.

Working with Charles Babbage on the theoretical Analytical Engine in the mid-19th century, Lovelace created what is considered the world’s first computer algorithm. She recognized that computation was not just arithmetic but the execution of step‑by‑step instructions — what we now call algorithms — a precursor to the general‑purpose computer. This profound realization — that numbers could represent entities other than quantity — launched the idea of universal computation, a century before it was realized.

Even more prescient, Lovelace articulated the limitations, arguing that “The Analytical Engine has no pretensions to originate anything. It can do whatever we know how to order it to perform.” Known as “Lady Lovelace’s Objection,” this philosophy is still at the heart of the AI debate.

Lovelace’s philosophical insight was so profound that when computers finally arrived 100 years later, researchers recognized they were still grappling with the exact question she had framed. In his famous 1950 paper “Computing Machinery and Intelligence,” Alan Turing addressed nine objections to the idea of machine intelligence — and Lady Lovelace’s Objection was among them, because her insight remained the most compelling articulation of why machines might never truly “think.” Her objection wasn’t outdated; it was ushering in the era of computation.

Today, modern AI systems produce outputs their creators never explicitly programmed. In 2016, Google’s AlphaGo defeated Lee Sedol, one of the world’s greatest Go players, using moves that stunned observers and were later adopted by human players — strategies no programmer had taught it. Yet all AI algorithms still require massive quantities of human-generated data and are designed according to principles established by human engineers. The debate hinges on how we define “origination.” If it means producing something humans didn’t program step-by-step, then AI already challenges Lovelace’s objection. If it means true creativity with intent and understanding — the capacity to anticipate genuinely new truths independent of human input — then her objection still stands. The question remains philosophically unresolved.