By Javier Surasky
Almost without realizing it, one afternoon we found ourselves not searching Google for what we wanted to know, but talking with an AI that not only answered us—it reasoned, translated, summarized, and even proposed new approaches. It wasn’t magic, as it might have seemed to those sitting for the first time in front of one of the new language models; it was the result of more than seventy years of technical, scientific, and political decisions that together tell a fascinating story, full of breakthroughs and contradictions.
Like a
written short film, this story can be told through ten short scenes—with two
surprising cameos. A story to read in one sitting to understand
precisely how we got here.
Scene 1.
Turing Changes the Question
1950
Context:
post–World War II. The United Nations had just been created, but the Cold War
divided the world.
Involved in
British intelligence more out of an obsession with solving riddles than
patriotism, Alan Turing proposed shifting the debate: instead of asking whether
a machine “thinks,” we should ask whether, in a conversation, one can tell the
difference between a human and a machine. The “imitation game” was
born, and with it, the first measurable framework for talking about artificial
intelligence. The field had no name yet, but it already had a horizon.
First
cameo: in 1956, the “Dartmouth Workshop” gathered the first international
community of AI scientists and gave the field its name. If Turing laid the
foundation stone, Dartmouth brought the builders who began the work.
Scene 2.
Learning from Mistakes
1986
Context:
Reagan and Thatcher are in power, Europe signs the Schengen Agreement, and
Latin America experiences democratic transitions.
Backpropagation makes network training practical:
the machine makes a mistake, measures how much, corrects it, and tries again.
Rumelhart, Hinton, and Williams gave machines a feedback loop that allowed them
to “look back to move forward,” enabling the efficient training of deep
networks.
Scene 3.
Memory and Context
1997
Context:
the Berlin Wall has fallen; Europe signs the Maastricht Treaty creating the EU;
Mandela becomes South Africa’s president; Rwanda suffers genocide; the
Zapatistas rise in Mexico; Israel and the PLO sign the Oslo Accords; and Asia
faces a financial crisis.
LSTM
(Long Short-Term Memory) networks give AI a notebook filled with gates and memory cells that allow it to
remember what matters in long sequences, text, voice, or series, ending the problem
of gradient vanishing. The whisper-down-the-line game that plagued
backpropagation was finally solved.
Scene 4.
Seeing to Understand
2012
Context:
the world still feels the 2008 financial crisis; the Lisbon Treaty enters into
force; WikiLeaks shakes global politics; and the Rio+20 Conference is held.
With AlexNet,
the combination of vast image datasets, GPUs, nonlinear algorithms, and data
augmentation breaks performance records in computer vision. Errors shrink, and
machines begin to describe what they see almost flawlessly.
Scene 5.
Meaning Speaks the Language of Geometry and Algebra
2013–2014
Context: a
new Pope (Francis); China launches the Belt and Road Initiative; Russia annexes
Crimea; ISIS rises; and Malaysia Airlines flight MH370 disappears.
Embeddings represent words and concepts as
numerical vectors-map, meaning geometrically: king is as close to man as queen
is to woman, and astronaut sits in an equidistant point between man and woman.
This distributed representation turns mapping and measuring distances
into the heart of AI language understanding.
Scene 6.
Big Data!
2015
Context:
the world agrees on the 2030 Agenda for Sustainable Development, the Paris
Agreement, and the Addis Ababa Action Agenda on financing for development.
The UN
report A World That Counts establishes data as the new infrastructure of
development, bringing Big Data and the “data revolution” from the
tech world into multilateral policymaking.
Second
cameo: in 2016, Federated Learning emerges. Collaboration without data extraction, learning without moving data, strengthening privacy, and sparking
debates about data sovereignty and digital cooperation.
Scene 7.
Attention Is All You Need
2017
Context:
Brexit unfolds, Trump takes office, the Colombian peace agreement is signed,
and the Panama Papers scandal explodes.
A paper
titled Attention Is All You Need introduces a new neural network
architecture, the Transformer, that parallelizes training, captures
long-term dependencies, and adds stability. Not everything that can be learned
is equally important: context matters. The Transformer becomes the
architectural hinge of modern AI.
Scene 8.
AI Gets Conversational
2018–2023
Context:
the U.S.-China trade war, Brexit’s conclusion, France’s “yellow vests,”
Bolsonaro’s presidency, the COVID-19 pandemic, and Russia’s invasion of
Ukraine.
In 2018,
Google presented BERT, a model that “understands” context by reading both
before and after each word. This leap brings nuance, irony, and meaning to
machines, and gives rise to LLMs (Large Language Models), including the GPT
family. AI now speaks our language, driving mass adoption.
Scene 9.
Hard Science, Real Impact, and a Nobel Prize
2020–2021
AlphaFold2 predicts protein structures with
near-laboratory precision, releasing a global atlas. AI proves it can
accelerate science, not just classify photos. The creators of AlphaFold later received
the Nobel Prize for the breakthroughs their work enabled.
Scene
10. The First Comprehensive AI Law
2024–2025
Context:
Trump’s second term; war in Ukraine; conflict in Gaza; expansion of BRICS; and
the UN Summit of the Future adopts the Global Digital Compact.
Amid
competing governance models, the U.S. free-market vs. China’s state-centered
approach, the EU takes the middle path, balancing innovation and rights, and
passes the world’s first comprehensive AI law: the EU AI Act, centered
on risk-based regulation.
Final
Credits
Far from a
historical museum of AI achievements, this “written short film” is a living
map of forces still in motion. Turing started it; backpropagation and LSTM
taught us how to teach; AlexNet improved perception; embeddings deepened
understanding; the world exploded in data; Transformers optimized learning;
LLMs democratized AI; AlphaFold proved its scientific worth; and the AI Act
drew the first global rules.
In AI, not
everything is new, nothing is completely changed, and no one knows what comes
next, and that is what makes it so fascinating, innovative, and challenging.
Before fearing or defending it, we must understand it.
As we have
said before, intelligence is not only what lies inside a head (or a machine) but
what emerges in the space between them.