If you read my last article, AI Abundance is a Lie, you might think there is little room for optimism about our future.
Concentrated wealth. Captured politics. AI systems built by billionaires who openly oppose either empathy or sharing. Forty-five years of productivity gains that went to capital owners while workers got nothing. The path to structural change blocked at every turn by people who benefit from keeping things exactly as they are.
That was a lot.
But what I didn't say clearly enough is this: the system is powerful, but not inevitable. The people running it are not invincible. And history — including our own very recent history — is full of moments when one person showed something so clearly, so undeniably, that the conversation never went back to where it was.
This is about three of those people.
People, not "heroes". Not geniuses with special access to power. A novelist who went undercover in a slaughterhouse. A grad student who couldn't get a computer to see her face. A researcher who got fired for writing the truth.
They didn't defeat the system. But they changed what the system had to answer for. That is not nothing. That is, in fact, how change starts.
Let's be precise about the argument here, because it's easy to slide into something I don't mean.
I'm not saying one passionate person can change the world through sheer force of will. That's a motivational poster, not a theory of change. And it implicitly lets everyone else off the hook — if you're not exceptional, why bother?
What I am saying is something more specific: structural change often starts with a moment when a problem becomes visible to enough people that ignoring it is no longer possible. Someone has to create that moment. That person is almost always an individual who decided to expose a truth that others were either too comfortable, too cautious, or too powerful to name themselves.
The lever that moves society isn't heroism. It's reframing.
Before Upton Sinclair, people vaguely knew that industrial food production was probably not great. After Sinclair, they knew exactly what was in their sausage. The disgust was always there, waiting. He just gave it a target.
Before Joy Buolamwini's research, the AI industry could claim its facial recognition systems were highly accurate — and produce the numbers to prove it. After her work, a new question replaced that one: accurate for whom? You can't unask that question. It changed what "accuracy" means.
Before Timnit Gebru was fired by Google for publishing research they didn't like, most people outside the industry had only a vague sense that AI ethics might be a problem. Google's attempt to bury her work turned it into a cause — and her into a symbol of exactly the conflict she had been describing.
None of them changed things alone. Sinclair's book needed an outraged public to pressure Congress. Buolamwini's research needed journalists, lawmakers, and the ACLU to amplify it. Gebru's firing needed thousands of people to stand up and say this is not acceptable.
The individual act was the spark. The collective response was the fire.
This matters, because it means the story isn't just about exceptional people doing exceptional things. It's about what becomes possible when someone creates the conditions for collective action — and about the thousands of ordinary people who made that action real.
You might not be the person who testifies before Congress or goes undercover in a meatpacking plant. But you might be one of the 4,300 researchers and others who signed a letter in support of Timnit Gebru. Without them, her firing would have been a footnote instead of a reckoning.
The three people in this article each created one of those moments — in different industries, in different eras, in different ways. The details are worth knowing.
In the fall of 1904, a 26-year-old socialist writer named Upton Sinclair boarded a train to Chicago with a notebook, a dinner pail, and a mission. For seven weeks, he worked undercover in the meatpacking plants of Packingtown — living among the immigrant workers, witnessing the conditions firsthand, gathering material for what he hoped would be his great American novel about class exploitation and the plight of the working poor.
He got his novel. But he didn't get what he came for.
The Jungle was published in 1906 and became an immediate sensation — just not for the reasons Sinclair intended. Readers didn't put it down outraged about the exploitation of Lithuanian immigrants. They put it down unable to finish their breakfast. His vivid descriptions of contaminated meat, of workers falling into rendering vats1, of rats and sawdust swept into sausage, ignited a public fury that swept through Congress like a brushfire.
For decades, nearly 100 food and drug safety bills had been introduced in Congress. Every single one had died — killed by food industry money and political indifference. Within months of The Jungle's publication, President Theodore Roosevelt — who privately thought Sinclair was a crackpot — signed both the Pure Food and Drug Act and the Meat Inspection Act into law. Those two laws became the foundation of what would eventually become the FDA.
Sinclair's famous verdict on his own work: "I aimed at the public's heart, and by accident I hit it in the stomach."
He hadn't failed. He had discovered something important about how change actually works: you don't always get to choose what resonates. What matters is naming a truth so vividly, so viscerally, that people can no longer pretend they didn't know. The political will to act was never the real obstacle. The shared, undeniable, gut-level outrage wasn't there yet.
Sinclair gave it a target.
In 2015, a graduate student at MIT was building an art installation — a kind of magic mirror that would overlay an aspirational figure onto the viewer's reflection. She was using off-the-shelf facial recognition software to track the viewer's face.
The software couldn't find hers.
Joy Buolamwini had to hold a white mask in front of her face to make it work.
She could have filed it away as a technical glitch and moved on. Instead, she started asking questions. What she found became her MIT thesis, a landmark research paper, and eventually a reckoning for the entire facial recognition industry.
Her project, Gender Shades, systematically tested commercial facial recognition systems from IBM, Microsoft, and Face++. (A follow-up study included Amazon's Rekognition system.) The industry was claiming accuracy rates of around 97% — impressive numbers, until Buolamwini looked at who those numbers reflected. For light-skinned men, error rates were as low as 1%. For dark-skinned women, misclassification rates reached as high as 47%. The systems weren't accurate. They were accurate for some people — specifically, the people who looked most like the overwhelmingly male, overwhelmingly light-skinned faces those systems had been trained on.
That reframe — not "is it accurate?" but "accurate for whom?" — sounds simple. It wasn't. It cut straight through years of industry self-congratulation and asked a question the numbers couldn't dodge.
The response from companies was revealing. IBM updated their software within 24 hours of receiving her findings. Amazon pushed back aggressively, disputing her methodology. Microsoft's president Brad Smith cited her research while publicly calling for government regulation of facial recognition — an unusual move for a tech executive.
Buolamwini didn't stop there. She and fellow researcher Timnit Gebru extended the work, auditing more systems. She founded the Algorithmic Justice League. She testified before Congress. She collaborated with the ACLU, which ran members of Congress through Amazon's facial recognition system — it misidentified 28 of them as people who had been arrested. The argument was hard to ignore: if it can't reliably identify a senator, what happens when it misidentifies a Black teenager in front of a police officer?
By 2020, every U.S.-based company she had audited had stopped selling facial recognition technology to law enforcement. IBM exited the facial recognition business entirely. Amazon imposed a moratorium. Microsoft followed.2
One grad student's frustration with a broken art project had changed industry policy at three of the largest technology companies on earth. Her story was later told in Coded Bias, a documentary available on Amazon (and sometimes other streaming services) — which introduced her work to millions of people who had never heard of algorithmic bias, and probably never would have.
She coined a term for what she had discovered: the coded gaze — the way the priorities, preferences, and prejudices of the people who build AI systems get quietly embedded into the technology itself. It's a phrase that does what the best reframes always do: once you have the words for something, you can't stop seeing it everywhere.
The story of Timnit Gebru and Google is, on the surface, a familiar one: a powerful corporation tries to suppress inconvenient research, and the researcher pays the price.
Except that's not quite how it ended.
Gebru came to Google in 2018 as one of the most respected researchers in AI ethics — co-author, with Joy Buolamwini, of the Gender Shades paper that had already shaken the facial recognition industry. At Google, she co-led the Ethical AI team, built it into one of the most diverse and consequential research groups in the field, and kept asking the kinds of questions that made her employer uncomfortable.
In 2020, she co-authored a paper called "On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?" It was a careful, technical argument that the AI industry's race to build ever-larger language models — the same technology now powering ChatGPT, Google's own products, and most of the AI tools you've encountered — was creating serious problems the industry wasn't acknowledging. Environmental costs. Training data riddled with bias. Systems that could generate confident-sounding text without anything resembling understanding. Accountability structures that didn't exist.
These were not fringe concerns. They were questions any serious researcher should have been asking. Google didn't want them published.
The company pressured Gebru to retract the paper or remove the names of Google employees from it. She refused. In December 2020, she was fired — or, in Google's careful corporate phrasing, her resignation was "accepted," a claim she disputed publicly and immediately.3
What happened next is the part Google did not anticipate.
Within days, more than 2,700 Google employees signed a letter condemning the firing. More than 4,300 academics, researchers, and supporters outside the company joined them. Nine members of Congress wrote to Google demanding an explanation. The AI research conference where the paper had been accepted — and which Google sponsored — removed Google from its list of sponsors in protest. Senior researchers began resigning. The story didn't disappear. It became the story.
The paper Google tried to bury is now one of the most cited documents in AI ethics. The questions it raised — about bias, about environmental cost, about corporate accountability — are now central to how governments, researchers, and increasingly the public think about AI regulation. Its concerns have shaped AI governance discussions at regulatory bodies and standards organizations worldwide. The term "stochastic parrots" entered the cultural vocabulary as shorthand for a real and recognized problem. Sam Altman, CEO of OpenAI, used it to describe his own product after ChatGPT launched.
Gebru didn't quietly retreat. She founded DAIR — the Distributed AI Research Institute — an independent research organization explicitly free from corporate funding and corporate pressure. It exists, in part, because her firing made the case that ethical AI research cannot survive inside companies whose business models depend on ignoring ethical questions.
There's a painful irony at the center of this story that's worth sitting with: Google built an Ethical AI team, hired one of the world's leading AI ethics researchers, and then fired her for doing AI ethics research. The contradiction was so glaring that the attempted suppression did more to advance the cause of AI accountability than the research alone ever could have.
Suppression is an admission. When a corporation fires someone for telling the truth, it tells the world exactly how threatening that truth is.
Three people. A century apart, more or less. Different industries, different tactics, different outcomes. What connects them?
None of them set out to be movement figures. Sinclair wanted to write a socialist novel. Buolamwini wanted to finish her art project. Gebru wanted to publish her research. The impact came from doing the work honestly and refusing to look away from what it revealed.
All of them faced real personal cost. Sinclair was dismissed by the president he inadvertently helped. Buolamwini faced aggressive pushback from one of the most powerful companies in the world. Gebru lost her job — and with it, the institutional platform and resources she had spent years building.
And all of them reframed a problem in a way that made it impossible to ignore. Not by winning an argument, but by making something visible that people couldn't unsee. Contaminated sausage. A 47% error rate on dark-skinned women. A company firing its own ethics researcher for doing ethics research.
But here's the thing that matters most for what comes next: none of them did it alone.
Sinclair had the muckraking tradition behind him — a whole generation of journalists who had been priming the public to distrust industrial capitalism. Buolamwini and Gebru were collaborators before they were separate stories; Gender Shades was both of theirs. Gebru's firing only became a reckoning because thousands of people — researchers, employees, lawmakers, ordinary people who had never heard of stochastic parrots — decided it mattered.
The individual was the spark. The collective was the fire. Every time.
Here's what I think we should take from these three stories — and it's not what you might expect.
I'm not asking you to go undercover in a data center. I'm not asking you to risk your career publishing research your employer doesn't want you to publish. I'm not asking you to testify before Congress or found an independent research institute.
Those things matter enormously, and the people who do them deserve our respect and our support. But they are not the only things that matter. And waiting until you're ready to be Upton Sinclair is no more effective than doomscrolling.
Think about those 4,300 researchers and academics who signed a letter after Timnit Gebru was fired. Most of them didn't risk much. They signed their name to a statement saying this is wrong. That's it. Some of them probably agonized over it. But together, they turned a corporate HR incident into a global reckoning that reshaped how we think about AI accountability.
Or consider the people who watched Coded Bias and then mentioned it to someone who hadn't. Who shared Buolamwini's research in a work meeting when someone said "but AI is neutral." Who asked their city council candidate what they thought about algorithmic hiring in municipal contracts. Small acts, but not nothing.
The spectrum of participation runs from signing a letter to founding an institution, and every point on that spectrum is necessary. The people at the bold end need the people at the quiet end to matter. Gebru needed her 4,300. Buolamwini needed the journalists and lawmakers who amplified her work. Sinclair needed the readers who got angry enough to write to their congressmen.
You are not powerless. You are, at minimum, one of the 4,300.
And once you're ready to think bigger than that — once you want to move from individual acts to collective structures, from signing letters to building something — that's where the next articles in this series come in. Worker cooperatives. Community land trusts. Local politics. Mutual aid. The places where individual commitment aggregates into durable power.
The fight for economic justice in the age of AI is not going to be won by exceptional individuals acting alone. But it is going to require each of us — people like the three in this article, and people like you and me — to decide that the truth is worth telling, and that action is worth taking.
The alternative is waiting for someone else to be Upton Sinclair.
Don't wait.
This is the second in a series examining AI, automation, and economic justice. The first piece, AI Abundance is a Lie, examined why technological productivity alone won't create broadly shared prosperity. Future pieces will explore specific policy solutions, successful organizing models, and the role of tech workers in shaping how AI is deployed.
The rendering vat scenes in The Jungle are among the novel's most infamous passages, but their literal accuracy has been disputed. Roosevelt's own investigators confirmed widespread unsanitary conditions in the meatpacking plants, but could not verify the specific claim that workers fell into vats and were processed into lard. Sinclair himself stood behind the book's overall portrayal of conditions, if not every specific detail. The broader picture — contaminated meat, dangerous working conditions, rampant adulteration — was confirmed by multiple independent investigations. ↩︎
The 2020 moratoriums by Amazon, IBM, and Microsoft applied specifically to law enforcement sales and were presented as temporary or conditional pauses pending federal legislation that largely never came. The picture today is more complicated: while those companies pulled back from direct police sales, federal agencies including ICE have continued to use facial recognition as part of broader surveillance systems that combine commercial data brokers, DMV records, and biometric databases. A 2022 Georgetown Law report, American Dragnet, documented how ICE had assembled what researchers called a "surveillance dragnet," accessing the driver's license photos of roughly one in three adults in the US. The ACLU has separately documented that there are almost no explicit legal limits on how agencies like ICE use facial recognition technology. Buolamwini has continued to advocate for binding legislation rather than voluntary corporate commitments — and after Trump's 2025 revocation of Biden's AI executive order, she spoke directly to the risk of relying on executive action alone: "Executive orders can be reversed — that's why we need laws to institutionalize protections." ↩︎
The exact circumstances of Gebru's departure remain contested. Google's official position, stated by AI chief Jeff Dean in an internal email that was later made public, was that Gebru had effectively resigned by setting conditions for her continued employment that Google was unwilling to meet. Gebru disputed this characterization strongly and publicly, saying she had asked to negotiate a final date and was instead cut off from her corporate email before she returned from vacation. The leaked emails and subsequent media coverage suggest the situation was more abrupt and less voluntary than Google's framing implied. For the purposes of this article, "fired" reflects Gebru's own account and the interpretation of most outside observers. ↩︎
At Davos 2026, Elon Musk sat across from BlackRock CEO Larry Fink and said: "With robotics and AI, this is really the path to abundance for all." He wasn't alone. Tech leaders and futurists have been selling a similar vision for years - a world where artificial intelligence and automation handle the dirty work while humanity reaps the benefits. More productivity, less scarcity, prosperity for everyone.
It's a compelling story. It's also a lie.
Use the share icon here if you liked it.
Thanks!