Signals in the Machine

History

Long before artificial intelligence became a product, it was a question: can thought be described, measured, and built?

Signals in the Machine is a cinematic documentary series about the people, ideas, breakthroughs, disappointments, and recurring ambitions behind AI. Season one travels from mechanical calculators and symbolic logic through wartime computing, the Dartmouth workshop, AI winters, expert systems, machine learning, deep learning, transformers, and generative AI. The series avoids both hype and panic to ask a more durable question: what does the history of intelligent machines reveal about the humans building them?

The Question Before the Machine

Long before computers, philosophers, clockmakers, and mathematicians wondered whether reasoning could be made mechanical. We begin in a seventeenth-century workshop with Blaise Pascal’s arithmetic machine, then follow an intellectual line through Thomas Hobbes, Gottfried Wilhelm Leibniz, Ada Lovelace, and George Boole. Their machines were not artificial intelligence, but their wager still powers AI today: that thought can be represented as symbols and transformed by rules. The modern history of AI begins with a question older than electricity—what is a thought made of?

Produced by Rogue Media Network. Narration was created with an AI voice selected by the producer. Acoustic music: “Poetic Lyrical Beautiful Cinematic Piano Cello Music” by Denis Pavlov, used under the Pixabay Content License. This series is independently produced and is not affiliated with the people, companies, or institutions discussed.

See Privacy Policy at https://art19.com/privacy and California Privacy Notice at https://art19.com/privacy#do-not-sell-my-info.

The Future Has a Past

The history of AI is often presented as a march toward the inevitable. It is better understood as a series of choices: what to measure, what to fund, whose data to collect, and who bears the cost when systems fail. In the season finale, we revisit the recurring cycle of bold claims, remarkable demonstrations, practical limits, and reinvention. Then we look beyond algorithms to the infrastructure and institutions beneath them—chips, data centres, workers, electricity, companies, and governments. History cannot predict AI’s future, but it can remind us that the future is still open to argument.

See Privacy Policy at https://art19.com/privacy and California Privacy Notice at https://art19.com/privacy#do-not-sell-my-info.

Attention

To understand a sentence, a reader constantly decides which words matter to one another. The transformer turned a version of that relationship into a scalable mathematical mechanism called attention. Beginning with the 2017 paper “Attention Is All You Need,” this episode explains why transformers changed language modelling, why parallel processing mattered, and how scale became a research strategy. More data, compute, and parameters produced increasingly fluent systems—and an increasingly important warning: language that sounds meaningful can still be wrong. Fluency is an achievement, not proof of understanding.

See Privacy Policy at https://art19.com/privacy and California Privacy Notice at https://art19.com/privacy#do-not-sell-my-info.

The Conversation Arrives

For most people, AI was once hidden inside recommendation engines, laboratories, and specialist tools. Then a blank chat box invited anyone to ask a question. We trace the path from language models to instruction-following and reinforcement learning from human feedback, explaining why helpfulness and fluency are engineered behaviours rather than evidence of personhood. The conversational interface opened remarkable uses in writing, translation, accessibility, coding, and tutoring. It also made hallucinations, bias, copyright disputes, and labour questions impossible to ignore. When a machine becomes persuasive, literacy becomes essential.

See Privacy Policy at https://art19.com/privacy and California Privacy Notice at https://art19.com/privacy#do-not-sell-my-info.

The Image That Changed Everything

In 2012, a system called AlexNet dramatically improved image-recognition results in the ImageNet competition. The breakthrough did not come from one magical idea. It came from millions of labelled images, specialised graphics processors, better training methods, and decades of neural-network research finally aligning. We explain how convolutional networks learn edges, shapes, and objects, then follow deep learning into speech, translation, and games such as AlphaGo. These systems were extraordinary pattern learners—not general minds—but they changed what researchers, investors, and the public believed machines could do.

See Privacy Policy at https://art19.com/privacy and California Privacy Notice at https://art19.com/privacy#do-not-sell-my-info.

When Machines Began to Learn

A handwritten number can take thousands of forms, yet people recognize it instantly. Teaching a computer to do the same helped change the direction of AI. This episode explains training data, parameters, error correction, and backpropagation without the equations, then follows neural networks through their revival in the 1980s and the data-rich world created by digitisation and the web. Deep Blue’s victory over Garry Kasparov became a cultural milestone, while biased datasets exposed a harder truth: machines learn patterns from human records, including our blind spots.

See Privacy Policy at https://art19.com/privacy and California Privacy Notice at https://art19.com/privacy#do-not-sell-my-info.

The Expert in a Box

In the 1970s and 1980s, AI narrowed its ambition and became useful. Expert systems such as DENDRAL and MYCIN encoded specialist knowledge as chains of “if/then” rules, helping computers reason inside carefully defined domains. Businesses rushed to capture expertise and make it repeatable. Then the rule books grew, experts disagreed, conditions changed, and maintenance became overwhelming. We explore the commercial boom, the knowledge-engineering bottleneck, and the collapse that followed—along with the durable insight expert systems left behind: reliable AI depends on constraints, evaluation, and real domain knowledge.

See Privacy Policy at https://art19.com/privacy and California Privacy Notice at https://art19.com/privacy#do-not-sell-my-info.

The First Winter

The first AI boom met a problem more stubborn than chess: reality. Machine translation struggled with context and culture. Perceptrons ran into mathematical and computational limits. Robots that looked impressive inside carefully arranged laboratories proved brittle elsewhere. As public promises outran dependable results, funders lost patience and the field entered its first “AI winter.” Yet research never stopped. This episode separates the myth of AI’s death from the quieter truth—and finds an evergreen lesson the industry keeps relearning: a dazzling demonstration and a trustworthy system are not the same thing.

See Privacy Policy at https://art19.com/privacy and California Privacy Notice at https://art19.com/privacy#do-not-sell-my-info.

The Summer AI Was Named

In 1956, a small summer workshop at Dartmouth gave artificial intelligence its name. John McCarthy, Marvin Minsky, Claude Shannon, Nathaniel Rochester, and their colleagues believed that the essential features of intelligence might be described precisely enough for a machine to simulate them. Early programs such as Logic Theorist made that confidence feel reasonable. But beneath the breakthroughs were tiny problem worlds, hand-built assumptions, and severe limits. This episode follows the birth of two rival dreams—intelligence as symbolic rules and intelligence as learned patterns—and asks why both remain alive today.

See Privacy Policy at https://art19.com/privacy and California Privacy Notice at https://art19.com/privacy#do-not-sell-my-info.

The Wartime Mind

World War II did not invent artificial intelligence, but it created the machines, institutions, and urgency that made it imaginable. At Bletchley Park, thousands of people and machines worked together to break encrypted communications. We trace Alan Turing’s role without the lone-genius mythology, examine the Bombe and Colossus, and follow Turing into his 1950 proposal for the imitation game. The war revealed that machines could perform immense chains of logical work. It also left a question no behavioural test can settle: when a machine sounds intelligent, what exactly have we learned?

See Privacy Policy at https://art19.com/privacy and California Privacy Notice at https://art19.com/privacy#do-not-sell-my-info.