Americas — Godfathers and Giants

How North America created deep learning and the AI giants while Latin America leapt toward adoption

Yesterday — The Winters and the Persistent Ones

There exists a virtue that success manuals often forget: obstinacy in obscurity.

During the winters of artificial intelligence — those periods when funding collapsed, when unfulfilled promises discouraged investors, when the brightest researchers left the field for more predictable careers — a handful of researchers persisted. They worked on neural networks that the dominant scientific community considered a dead end. They published papers that few read. They trained students who struggled to find positions.

Three names emerged from this obscurity: Geoffrey Hinton, Yoshua Bengio, Yann LeCun.

Geoffrey Hinton, born in Great Britain, had settled in Canada. At the University of Toronto, he continued to believe in neural networks when the academic world had abandoned them. Yoshua Bengio, at the University of Montreal, had founded the Mila laboratory in 1993 — an island of deep learning research in an ocean of skepticism. Yann LeCun, a Frenchman who had emigrated to the United States, developed convolutional neural networks at Bell Labs, then at New York University, perfecting architectures that no one seemed to want to use.

They knew each other. They collaborated. They persisted.

In 2006, Hinton and his students published a paper that relaunched everything. They demonstrated that it was possible to train deep neural networks — with multiple layers — without the process collapsing. The technique of "unsupervised pre-training" opened a door that the community had believed permanently closed.

In 2012, Alex Krizhevsky, a doctoral student of Hinton's, trained AlexNet on two video game graphics cards in his parents' bedroom. On September 30, AlexNet won the ImageNet challenge with an error rate of 15.3% — ten points better than the best competing system. ImageNet itself had been created by Fei-Fei Li, born in China, a professor at Stanford — a database of fourteen million images classified into twenty thousand categories, the fuel that had made deep learning possible.

The winter was over. The summer that was opening would be the longest and most transformative in the history of AI.

In March 2019, the Association for Computing Machinery awarded the Turing Prize — often called the "Nobel of computing" — to Geoffrey Hinton, Yoshua Bengio, and Yann LeCun "for their conceptual discoveries and engineering advances that made deep neural networks an essential component of computing." They were nicknamed the "godfathers of deep learning."

Two of them worked in Canada.

Canada, almost by accident, had become the epicenter of the revolution. While Silicon Valley pursued other mirages, Montreal and Toronto had welcomed the obstinate researchers. In 2017, Canada became the first country in the world to establish a national artificial intelligence strategy — the Pan-Canadian AI Strategy — with initial funding of one hundred twenty-five million Canadian dollars to support three institutes: Mila in Montreal, the Vector Institute in Toronto, Amii in Edmonton.

In 2024, the Canadian federal budget announced two billion four hundred million dollars in investment in the AI ecosystem — including two billion for a sovereign computing strategy. The small country that had sheltered the persistent ones was now investing massively to maintain its lead.

Yoshua Bengio had become, according to academic metrics, the most cited scientist in the world — across all disciplines. Geoffrey Hinton, in October 2024, received the Nobel Prize in Physics with John Hopfield "for their foundational discoveries and inventions enabling machine learning with artificial neural networks." The godfathers of AI had become laureates of the highest distinctions that humanity bestows.

Today — The Race of the Giants

Then came the giants.

In December 2015, a nonprofit organization was founded in San Francisco. It was called OpenAI. Its founders included Sam Altman, then president of the startup accelerator Y Combinator, and Elon Musk, the billionaire of Tesla and SpaceX. The stated objective: to develop artificial intelligence beneficial to all of humanity, without the commercial constraints of large corporations.

One billion dollars were pledged by the founders and their allies — Reid Hoffman, Peter Thiel, Amazon Web Services, Infosys. The reality was more modest: only one hundred thirty million had actually been collected by 2019.

But OpenAI changed the world.

On November 30, 2022, OpenAI launched ChatGPT — a conversational system based on the GPT-3.5 model. In five days, the application reached one million users. In two months, one hundred million. It was the fastest-growing consumer application in the history of the Internet — faster than TikTok, which had taken nine months, faster than Instagram, which had taken two and a half years.

On March 14, 2023, OpenAI released GPT-4. The new model passed American bar examinations. It solved complex mathematical problems. It wrote code. The boundary between what machines could and could not do was shifting at dizzying speed.

Google triggered a "code red" in December 2022. The search giant, which had acquired DeepMind ten years earlier, suddenly found itself surpassed by a startup. Microsoft, for its part, invested ten billion dollars in OpenAI in January 2023, integrating the technology into its products — from Bing to Office.

But OpenAI was not alone.

In 2021, seven former OpenAI employees founded Anthropic. Among them, Dario Amodei, former vice president of research at OpenAI, and his sister Daniela. Their motivation: growing concern about the risks of increasingly powerful AI systems, and disagreement over the direction OpenAI was taking.

Anthropic was constituted as a "public benefit corporation" — a structure allowing it to pursue a social mission while raising capital. Its stated objective: to develop safer, more transparent AI that is better aligned with human values. Claude, its conversational assistant, was launched in March 2023.

Anthropic's approach was distinctive. The company gave the British government early access to its models for safety testing — a first in the industry. It committed to building institutional safety measures proportionate to risks before even training new models. OpenAI and DeepMind followed, adopting similar policies.

Amazon invested four billion dollars in Anthropic. Revenue went from zero to one hundred million dollars in 2023, then one billion in 2024. In 2024, several key OpenAI employees — including Jan Leike, former director of the "Superalignment" team — joined Anthropic.

Google, Meta, Elon Musk's xAI: the giants were multiplying. Each was developing its own models — Gemini, Llama, Grok. Meta chose to release Llama as open source, allowing anyone to use and modify it. The race for large language models had become the new space race.

But this race had a cost. Training a state-of-the-art model required hundreds of millions of dollars in computing, energy, and data. Only companies with considerable resources could participate. The concentration of technological power was accelerating.

In November 2023, OpenAI's board of directors briefly fired Sam Altman. A revolt by employees and Microsoft — which held forty-nine percent of OpenAI's commercial branch — brought him back to power five days later. The episode revealed tensions between the organization's original mission — to develop AI beneficial to humanity — and the commercial realities of a company valued at tens of billions of dollars.

By November 2025, ChatGPT was approaching its third anniversary with eight hundred million weekly users.

Geoffrey Hinton was no longer just a godfather. He had become a prophet of concern.

In May 2023, Hinton resigned from Google. He wanted to be able to speak freely about the risks of artificial intelligence without being constrained by his employer's interests. In the months that followed, he multiplied public appearances to warn about the potential dangers of the systems he had helped create.

"When I was doing research, I thought I was building something good for humanity," he declared. "Now, I'm not sure anymore."

Yoshua Bengio shared this concern. He advocated for global AI governance, for moratoriums on the riskiest developments, for collective reflection before technology outpaced wisdom. The godfathers who had lit the fire were now trying to control the flames.

This tension — between acceleration and prudence, between the commercial race and ethical responsibility — defined North America's AI landscape. It was creating the most powerful systems in the world. It was also producing the most worried voices.

The Other America

Meanwhile, south of the Rio Grande, another revolution was taking place.

Latin America did not have the fundamental research laboratories of Canada. It did not have the capital of Silicon Valley. But it had something else: concrete problems to solve and a population ready to adopt new technologies.

In 2024, the artificial intelligence adoption rate in Latin America jumped eighteen points to reach forty percent — exceeding the global average in enthusiasm and optimism. The Latin American AI market, valued at four billion seventy-one million dollars in 2024, was projected to reach thirty billion two hundred million by 2033.

Brazil led the charge. Ninety percent of large Brazilian companies were using AI. The country had one hundred fifty-four artificial intelligence companies — the largest number in the region. Agriculture, which represented nearly thirty percent of Brazilian GDP, was deploying AI for precision farming. In 2024, three Brazilian agtech companies exceeded one hundred million dollars in valuation: Genica (biotechnology), Agrolend (agricultural financing), and Solinftec (agricultural AI).

Mexico was home to more than seven hundred thousand information technology professionals. The cities of Guadalajara and Mexico City were becoming hubs for agtech innovation. Aiflow was using drone imagery and predictive analytics to detect crop diseases before they caused losses.

Argentina, despite its economic turbulence, remained a major global agricultural player. The startup Kilimo had reduced water consumption by twenty percent while maintaining crop yields — saving seventy-two billion liters of water.

The Latin American Artificial Intelligence Index (ILIA), published in September 2024, evaluated nineteen countries. Chile came first with 73.07 points, followed by Brazil (69.30) and Uruguay (64.98). Argentina, Colombia, and Mexico were classified as "adopters."

But Latin America faced structural challenges. Only four out of ten rural Latin Americans had access to basic Internet — excluding millions of people from the digital economy. Total investment in startups in the region reached eight billion two hundred million dollars in 2024 — compared to one hundred ninety billion globally. The gap remained immense.

The region had more than eleven thousand four hundred funded startups, having raised three billion six hundred million dollars in 2024. Financial technologies attracted sixty-one percent of all venture capital investments. Between thirty and forty Latin American unicorns now existed — Mercado Libre, Nubank, Rappi among the best known.

Latin America was not inventing AI. It was adopting it, adapting it, applying it to its realities — agriculture, finance, logistics. It was showing that one could benefit from the revolution without being its epicenter.

Beyond — The Two Faces

The Americas present two faces of the deep learning revolution.

The first face is that of creation. From Hinton's laboratory in Toronto to OpenAI's headquarters in San Francisco, North America invented modern deep learning, created large language models, launched the race for conversational systems. It produced the godfathers — and the giants that transformed their discoveries into products used by billions of people.

The second face is that of adoption. From Brazil to Mexico, from Argentina to Chile, Latin America has shown that AI can solve concrete problems — irrigate fields more efficiently, grant credit to those without financial history, optimize complex supply chains. It has proven that the revolution did not need to wait for fundamental research laboratories.

These two faces are not opposed. They complement each other.

North America teaches us obstinacy. Hinton, Bengio, and LeCun persisted during the AI winters when no one believed. Their patience was rewarded with a Turing Prize, a Nobel Prize, and the satisfaction of having transformed the world. Fundamental innovation sometimes comes from those who refuse to give up.

It also teaches us ambivalence. The godfathers who created deep learning are now among the voices most concerned about its risks. Hinton left Google to sound the alarm freely. Bengio advocates for global governance. This capacity to create and criticize one's own creation, to accelerate and brake, defines a maturity that other regions have not yet developed.

Latin America teaches us adaptation. It did not invent large language models, but it applies them to precision agriculture, financial services, logistics. It shows that AI is not reserved for rich countries — it can serve wherever problems exist and entrepreneurs are ready to solve them.

It also teaches us limits. Four out of ten rural residents without Internet access. Investments that remain a fraction of those in the North. Adoption without creation creates dependence. Latin America uses tools designed elsewhere, according to priorities defined elsewhere, with biases encoded elsewhere. Technological sovereignty remains a distant horizon.

The question that runs through the Americas is that of responsibility.

The North American giants — OpenAI, Anthropic, Google, Meta — are developing systems that transform the entire world. Their decisions affect billions of people who never voted for them, who do not know them, who have no recourse if they make mistakes.

Dario Amodei speaks of a "race to the top" — a competition where companies vie for safety rather than raw power. Geoffrey Hinton calls for moratoriums. Yoshua Bengio calls for global governance. But the race continues, and billions of dollars flow to those who promise to go faster.

Latin America, for its part, adopts without really choosing. It benefits from the tools but does not define the rules. It solves local problems with global technologies. It participates in the revolution without shaping it.

The two Americas are linked by asymmetry. The North creates and worries. The South adopts and hopes. The question of who shapes AI — and according to what values — remains open.

The godfathers have become troubled sages. The giants have become empires. And somewhere between the Toronto laboratory and the fields of Brazil, artificial intelligence is transforming an entire continent.