What I learned from four women who reshaped the way we think with machines

I often think about where technologies come from—not just the tools themselves, but the historical context that shaped them. Modern computing feels almost impossibly dense: billions of transistors packed into a single chip, systems too complex for any one person to fully understand, and insights pulled from datasets too large to hold in the mind. It’s hard to believe the original inventors imagined any of this. But as I studied major computing milestones, I noticed something unexpected. The biggest leaps didn’t just come from invention—they came from making the information teachable.

The women who played foundational roles in computing didn’t stop at innovation—they turned breakthroughs into systems others could learn from. While many of their male peers focused on performance, scale, or theory, these women prioritized usability and education. They built tools, methods, and curricula that opened the field to people outside elite research spaces. In doing so, they didn’t just invite participation—they accelerated progress. By multiplying the number of people who could build on their work, they expanded the collective computing power of humanity itself.

By making their ideas accessible and teachable, they turned specialized breakthroughs into shared tools. That shift expanded who could think, build, and create with computers—accelerating progress through collective participation. This is a focused look at four women whose work removed barriers, scaled knowledge, and opened new eras of innovation.

Ada Lovelace (1843, First Computer Algorithm)

From numbers to symbols

Ada Lovelace was curious, mathematically disciplined, and philosophically driven. Born into British aristocracy as the daughter of poet Lord Byron, she was raised apart from him and largely in the care of others. Her mother, determined to steer her away from what she saw as poetry’s emotional volatility, emphasized logic, mathematics, and science in her education. That foundation, combined with Lovelace’s imagination, would lead to a single handwritten algorithm that helped reshape how we understand computing.

As an upper-class girl with access to elite tutors, Lovelace studied subjects rarely available to women at the time—and excelled. Yet she remained deeply imaginative, describing her approach as “poetical science.” She consciously blended analytical thinking with creative insight to explore how the world worked.

In the 1830s and 1840s, Lovelace collaborated with Charles Babbage on the Analytical Engine, a theoretical machine designed to use punch cards to carry out sequences of mathematical operations. Unlike his earlier Difference Engine, which was built to automate polynomial calculations, the Analytical Engine introduced something far more ambitious: a programmable architecture capable of handling conditional logic. At a time when automation was mostly limited to looms and number tables, this was a radical leap forward.

While translating an Italian paper on the Analytical Engine, Lovelace added extensive notes—one of them, Note G, became a foundational text in computing history.

In that note, Lovelace described how the Analytical Engine could manipulate symbols, not just numbers—envisioning it processing language, music, or any symbolic system. She also included what is now considered the first published computer algorithm: a step-by-step method for the machine to compute Bernoulli numbers. Her work didn’t just explain the machine; it demonstrated how to use it. That combination of imagination and precision is why she’s often recognized as the world’s first computer programmer.

A philosophy for programming

Lovelace saw programming as a new kind of language—one that could formalize reasoning and make it visible to others. By writing a program, she believed, people clarified their own thinking in ways they might not otherwise. That clarity could then be shared, making complex ideas more accessible. In her view, this process could benefit society by spreading knowledge, accelerating analysis, and helping more people build on each other’s insights.

“The engine has no pretensions to originate anything,” Lovelace wrote. She never expected machines to think for us. They would only follow the ideas we encoded. Yet that act of encoding mattered, because translating our reasoning into step-by-step code—and sharing it—could turn individual insight into knowledge the whole world could use and improve.

More than a century later, Alan Turing returned to this question. Could a machine’s behavior be indistinguishable from a human’s—and would that count as thinking?

He explored that idea more deeply in his 1950 paper Computing Machinery and Intelligence, where he cited Lovelace directly. Her claim that machines could only do what they were programmed to do later became known as the Lovelace Objection. Turing didn’t cite it to dismiss it. He took it seriously—as a genuine challenge that shaped his thinking and pushed him to ask whether a machine’s behavior, if indistinguishable from a human’s, could be considered intelligence.

Grace Hopper (1952, Inventor of the First Compiler)

From symbols to words

Grace Hopper was a mathematics professor at Vassar with a Ph.D. from Yale when she joined the Navy Reserve during World War II. She was assigned to work on the Harvard Mark I—one of the first electromechanical computers. Programming it meant feeding in punched paper tape or physically rewiring the machine—every task had to be built from scratch.

The Mark I could compute artillery trajectories, but shifting to a new task—like supply logistics—meant reprogramming it from scratch, which Hopper recognized as a structural bottleneck.

After the war, Hopper joined the team building UNIVAC, the first commercial electronic computer in the U.S., helping to launch computing as an industry. As computers made their way from military labs to businesses, the people using them weren’t mathematicians but everyday professionals with no technical training. Hopper is credited with writing the first computer programming manual, but she knew that kind of documentation wouldn’t be enough for business users. They needed a way to write and modify code without learning low-level programming.

Lovelace imagined machines working with symbols, not just numbers. A century later, Hopper took that further. Few people are good symbol manipulators. But what if they could use the language they already speak? Her compiler made that possible by translating commands like “subtract income tax from total pay” into machine code. Now someone in payroll could program a computer without knowing how it worked.

With a compiler, programs could be written once and used across multiple locations. Non-engineers could adjust them to new needs without starting from scratch. Business operations like payroll, accounting, and logistics could now be automated by the people doing the work—not just by specialists. This made it possible for computing to spread across departments and offices during the postwar economic boom, as companies expanded overseas and began operating as multinational corporations.

They said computers couldn’t understand English

When Hopper proposed writing programs in plain English, critics insisted it couldn’t work. Computers, they argued, were built for arithmetic—not language. But she built a fully functioning compiler anyway. Even after proving it worked, it still took years to convince others it was the future. But once it gained traction, the question became how to teach it.

Hopper had been a teacher long before she became a programmer. She joined the Vassar faculty in 1931 and was an associate professor by 1941. That background influenced how she approached programming—she believed it had to be understandable and usable by ordinary people. She wrote the first programming manuals and helped shape a new way of thinking about code that others could learn from. That foundation would prove critical once her ideas finally took hold.

As compilers gained acceptance, universities, governments, and businesses quickly adopted programming curricula and built internal teams to support them.

High-level languages like COBOL and FORTRAN became standard—used for business operations, scientific computing, and data processing. The first wave of operating systems followed, along with early classes of business software like payroll, inventory, and logistics tools. Hopper’s ideas became the foundation for how programming would be written, taught, and applied across industries.

Hopper’s contributions helped programming keep pace with computing’s rapid evolution. The integrated circuit, introduced in 1958, was a major breakthrough—it made it possible to shrink components while dramatically increasing computing power. That shift echoed what Hopper had long taught: computers needed to be small to be fast. Moore’s Law followed in 1965, forecasting exponential growth in hardware. Hopper had solved the software bottleneck. The next challenge was designing chips that could keep up.

Lynn Conway (1978–79, Co-developer of Mead–Conway VLSI Methodology)

They tried to erase her—she returned to transform chip design

For much of her life, there were two Lynn Conways—both with foundational contributions to computing. Early in her career at IBM in the 1960s, she developed a method called dynamic instruction scheduling, which allowed processors to reorder instructions during execution to avoid delays. It made modern CPUs faster and more efficient by keeping them from stalling while they waited for tasks to complete.

But Conway’s career was abruptly cut short. When she disclosed her gender transition to IBM management, she was fired—a reflection of the cultural prejudices of the time. She disappeared from the public record, leaving behind her name, credentials, and early accomplishments. What followed wasn’t just personal reinvention—it was the start of a second career that would reshape hardware design on a global scale.

By this time, Moore’s Law was well established—but it was clear that existing chip design methods couldn’t keep up. The process wasn’t standardized, and all documentation and techniques were controlled by a few major corporations. Yet just five years after being fired from IBM, and with no background in microchip design, Conway had already become a key figure at Xerox PARC—contributing to a project that would redefine how hardware was built and scaled.

Conway partnered with Caltech professor Carver Mead to develop the very-large-scale integration (VLSI) design methodology—a repeatable framework that broke chip architecture into clear, modular layers. It introduced standardized layout rules, separating logical design from physical implementation. Lambda-based abstraction replaced fixed measurements with simple design rules that worked across different manufacturing processes. Instead of redesigning from scratch for each new chip, engineers could follow general guidelines—like keeping a certain amount of space between lines—and trust the system to scale properly.

She made chip design teachable—and unstoppable

Even as the VLSI method took hold, Conway had already identified the next constraint. Chip designs were becoming too complex for manual layout, even with a standardized framework. She wrote the algorithms that powered the first generation of electronic design automation (EDA) tools—software that could arrange millions, then billions, of transistors.

The result was hardware capable of extraordinary speed and density—but it was Conway’s earlier work at IBM that made that power usable. Dynamic instruction scheduling allowed processors to execute instructions out of order, identifying parallel tasks and accelerating performance. That breakthrough became essential to modern chips like Intel’s Core, AMD’s Ryzen, and Apple’s M-series. Without it, all that processing power would sit idle, trapped in linear execution.

Conway knew methodology alone wouldn’t be enough—she had to prove it was teachable. In 1979, with support from DARPA’s VLSI Project, she co-wrote Introduction to VLSI Systems, the first textbook aimed at students outside elite labs. University courses were launched where students learned the method and designed their own chips. DARPA helped ensure those designs were actually fabricated.

The approach reshaped the industry. Throughout the 1980s, VLSI programs spread across universities. Conway’s students founded companies like Sun Microsystems and Silicon Graphics, and helped launch the RISC microprocessor movement. Today’s fabless chip model and foundry-based manufacturing rest on the infrastructure she built.

Modern computing runs on a system that once pushed her out

Later in life, Conway came forward about her history. In 1999, she revealed her transgender identity and reclaimed the story that prejudice had forced her to leave behind. The response was immediate. She became a public figure not just in computing, but in STEM culture—mentoring young LGBTQ+ engineers and challenging systemic erasure across disciplines. In 2020, IBM issued a formal apology for firing her.

Modern computing—used in everything from defense systems to conservative media—literally runs on ideas created by a trans woman. Reactionary politics attack the very people whose contributions make their platforms possible. If Conway’s career isn’t proof that diversity drives progress, it exposes how hollow those arguments truly are. Every time someone opens a laptop, loads a webpage, checks a text, or builds a new system, they’re relying on architecture shaped by Lynn Conway.

Fei‑Fei Li (2006–09, Creator of ImageNet Dataset)

Creating the data curriculum for machines to learn

By the early 2000s, advances in software and hardware had laid the groundwork for artificial intelligence. The field had been formally established in 1956, just after Turing proposed his famous test. Early optimism led to government funding, but that dried up by the 1970s in what became known as the first “AI Winter.”

A second wave of hype came in the 1980s, driven by expert systems, and collapsed again before the decade ended. Processing power had grown—but not enough to deliver on the field’s promises.

Through the 1990s and early 2000s, researchers continued their work, but many deliberately avoided the term artificial intelligence. Instead, they used alternatives like expert systems, data mining, and machine learning.

The real shift came in the mid-2000s, when they discovered that training neural networks on GPUs could dramatically speed things up. NVIDIA’s release of CUDA in 2006 was a turning point. By 2009, Andrew Ng’s lab had achieved a 70x speed-up, completing in one day what used to take weeks. The hardware and algorithms were finally in place—but something was still missing.

Fei-Fei Li identified what was missing. Hopper and Conway had created ways to teach humans how to build computers, but Li assembled the materials needed to teach machines. She put it simply, “The world’s best computer vision algorithm can’t recognize a cat if it’s never seen one.” While others focused on fine-tuning models, Li saw that scale—and quality—of training data was the real bottleneck. So she set out to build the largest labeled image dataset AI had ever seen.

Teaching machines to see the world like we do

In 2007, as a young professor at Princeton, Fei-Fei Li launched the ImageNet project. Digital cameras and the internet had made billions of images accessible. Her goal was to create a massive dataset of labeled images, grouped into categories, so AI could see what we see.

Instead of thousands of images like earlier datasets, ImageNet contained over 14 million, organized into 22,000 categories. The scale was unprecedented. Crowdsourcing—coined just a year earlier in Wired—was still new, but Li saw an opportunity. With help from the public, her team tagged, categorized, and launched the full dataset by 2009.

After assembling the dataset, Li’s team created the ImageNet Large Scale Visual Recognition Challenge (ILSVRC) to test her theory that large-scale data was the missing piece. In 2012, a deep convolutional network called AlexNet shattered previous records, cutting the error rate by more than 10 percentage points—down to 15.3%.

What AlexNet achieved with ImageNet wasn’t just a breakthrough—it showed deep learning could outperform other training methods.

Well-labeled, large-scale datasets quickly became the standard. And because ImageNet wasn’t controlled by Big Tech, it remained freely available for non-commercial research and education. Since then, datasets have grown dramatically. Open-weight models like Stable Diffusion were trained on LAION-5B—an open dataset with 5.85 billion image-text pairs, more than 350 times the size of ImageNet.

Li’s background in both computer science and cognitive neuroscience helped shape her vision. But she didn’t stop at infrastructure—she became one of the field’s most vocal advocates for transparency, inclusion, and ethical design. From the beginning, her approach was human-centered. And like many of the researchers responsible for real breakthroughs in AI, her warnings about bias and harm should not be ignored.

Beyond computation is the question of meaning

Ada Lovelace and Alan Turing were both correct

The trajectory of computing has mirrored human development. First came symbolic language and memory, followed by physical bodies (hardware) and minds (software and operating systems). The latest generation shows signs of abstraction and reasoning. But the next developmental step—both for humans and machines—is conscience: values, ethics, and meaning.

Artificial intelligence isn’t just a new capability. It marks the shift from machines that calculate to machines that generate and act. If we are recreating ourselves through technology, then the next phase must be moral.

The Lovelace Objection remains unresolved. Ada Lovelace argued that machines can only do what they’re explicitly told—that they can’t originate ideas or act creatively. Alan Turing disagreed, suggesting that symbolic systems might one day produce novel results indistinguishable from human thinking. That debate still stands, but it’s more nuanced now.

AI can generate code, writing, images, and music. But it has no internal experience—no emotion, no values, no intent. Critics say what looks like originality is just pattern recognition drawn from training data. Others argue that combining learned material in new ways is a valid kind of origination.

These systems can look at more data than a human ever could, identify gaps in our knowledge, generate plausible new hypotheses, and offer innovative solutions to unsolved problems.

Whether that counts as origination depends on how we define it. But the outputs are getting more powerful—faster, larger, and more persuasive than anything a person could produce alone. We build tools to extend ourselves. So the real question isn’t whether machines can create—it’s whether what they create has meaning.

I believe both Lovelace and Turing were right. Machines are producing results indistinguishable from human work—but only because we made that possible. Humans designed the systems, selected the data, wrote the algorithms, and built the hardware to run it all.

The quality of the output still depends on the quality of the human directing it. Fei-Fei Li is right to warn us. If we don’t embed values now, we risk building tools that reflect the worst in us—bias, greed, cruelty, and exploitation. The flood of low-quality AI output isn’t the fault of the tools. It’s the result of who’s using them—and the absence of ethics, regulation, and responsibility.

What Ada Lovelace couldn’t have imagined

When Ada Lovelace described machines manipulating symbols rather than just numbers, she was imagining a future beyond her time. But even she couldn’t have foreseen the scale of symbolic compression and abstraction possible today. The latent space behind generative AI is symbolic at its core. It encodes relationships, patterns, and potential drawn from massive datasets—allowing machines to generate outputs that resemble creativity. It’s symbolic reasoning in compressed, multidimensional form—an extension of Lovelace’s insight, operating at scales she couldn’t have conceived.

That’s what makes the Lovelace Objection so powerful—and so hard to resolve. Machines are no longer just tools. They generate knowledge, culture, and decisions that shape the world. But they don’t understand what they’re doing, and they can’t care. That’s the bottleneck now. We’ve scaled hardware, software, and data. We’ve taught machines how to see, speak, write, and design. The next leap—arguably the most important—is teaching them how to do it ethically.

Progress isn’t just invention. It’s generosity.

Each of these women removed a major bottleneck in computing’s evolution. Lovelace introduced symbolic computation and algorithms. Hopper gave us the tools to program. Conway gave us the foundation to scale hardware. Li gave us the data to teach the systems we built.

None of them kept their breakthroughs locked behind proprietary walls. They opened them up. That’s what unleashed exponential progress. Their work wasn’t just innovation—it was translation. They turned complexity into something more people could understand, shape, and build on. They made computing accessible, explainable, and teachable.

Fei-Fei Li stands out because she didn’t just teach people—she gave AI its first real curriculum. Her work gave machines the structure and data needed to learn. And she’s become one of the strongest voices, reminding us that what they learn next still depends on us.

We don’t know if machines will ever reach consciousness or moral agency—or what that would even mean. But we’re close enough to ask the question seriously. And like every phase before it, we’ll reach another bottleneck. Somewhere, someone will see past it—and show us how to move forward.

I can’t help but wonder if that person will be another ingenious woman.