B2 · Upper Intermediate
16 min
technologylaw and policyeuropean affairsartificial intelligencegeopolitics
Brussels Told the Algorithm No
Antes de que la máquina decida
The European Union passed the world's first comprehensive law regulating artificial intelligence, and its reach extends well beyond Brussels. Fletcher and Octavio debate what it actually means to draw legal lines around machines, and why Europe decided to do it first.
La Unión Europea aprobó la primera ley del mundo para regular la inteligencia artificial, y sus consecuencias llegan mucho más lejos que Bruselas. Fletcher y Octavio debaten qué significa realmente poner límites a las máquinas, y por qué Europa decidió hacerlo antes que nadie.
5 essential B2-level terms from this episode, with translations and example sentences in Spanish.
| Spanish | English | Example |
|---|---|---|
| auditable | auditable | Los sistemas de IA de alto riesgo tienen que ser auditables antes de llegar al mercado. |
| biométrico | biometric | La vigilancia biométrica en espacios públicos está muy restringida por la nueva ley. |
| transparencia | transparency | La ley exige transparencia sobre los datos utilizados para entrenar los modelos de inteligencia artificial. |
| sistémico | systemic | Los modelos de riesgo sistémico pueden tener consecuencias sobre sectores enteros de la sociedad. |
| subjuntivo | subjunctive | Es necesario que los ciudadanos conozcan sus derechos: ese verbo está en subjuntivo. |
The document runs to 458 pages.
I know this because I made the mistake of downloading it, and my laptop has not quite forgiven me.
But buried in those 458 pages is the first serious attempt by any government anywhere to draw hard legal lines around artificial intelligence.
Sí, y lo que hay que entender es que Europa no llegó a esto de repente.
Yes, and what you have to understand is that Europe didn't arrive at this suddenly.
La Unión Europea lleva años pensando en cómo regular la tecnología, y el Reglamento de Inteligencia Artificial es el resultado de ese trabajo.
The European Union has spent years thinking about how to regulate technology, and the AI Act is the result of that work.
Entró en vigor en agosto de 2024 y es la primera ley de este tipo en el mundo.
It came into force in August 2024 and is the first law of its kind in the world.
First in the world.
That's not a minor footnote.
And what gets me is that it didn't come from Washington or Beijing, it came from Brussels, which tells you something real about how differently Europe thinks about the relationship between technology and rights.
La ley funciona con un sistema de categorías de riesgo.
The law works with a system of risk categories.
Hay cuatro niveles: riesgo inaceptable, riesgo alto, riesgo limitado y riesgo mínimo.
There are four levels: unacceptable risk, high risk, limited risk, and minimal risk.
Dependiendo de la categoría, las obligaciones son completamente diferentes.
Depending on the category, the obligations are completely different.
Lo que está en la categoría de riesgo inaceptable está directamente prohibido.
What falls under unacceptable risk is outright banned.
So it's not a blanket prohibition on AI, it's tiered.
The higher the risk to people, the stricter the rules.
And the things in that top tier are just flat-out illegal.
What actually lands in the unacceptable category?
Hay cosas que ya no se pueden hacer en Europa.
There are things that can no longer be done in Europe.
Los sistemas de puntuación social que usan los gobiernos para evaluar y clasificar a los ciudadanos están prohibidos.
Social scoring systems used by governments to evaluate and classify citizens are banned.
También están prohibidos los sistemas que manipulan el comportamiento de las personas de forma inconsciente, sin que ellas lo sepan ni lo consientan.
Also banned are systems that manipulate people's behavior unconsciously, without their knowledge or consent.
Social scoring.
That's the architecture China has been building for years, where a government assigns citizens a behavioral rating and that score determines what they can access, where they can travel, whether their kids get into certain schools.
The EU is saying categorically: not here, not for governments.
Correcto.
Correct.
Y también hay restricciones muy importantes sobre la vigilancia biométrica en tiempo real en espacios públicos.
And there are also very significant restrictions on real-time biometric surveillance in public spaces.
El reconocimiento facial en las calles, por ejemplo, está muy limitado.
Facial recognition in the streets, for example, is heavily restricted.
Solo se permite en situaciones muy específicas, como amenazas terroristas graves o la búsqueda de personas desaparecidas.
It's only permitted in very specific situations, such as serious terrorist threats or the search for missing persons.
Which, if you've spent any time recently in a major American airport or looked up at the cameras in central London, you realize how far ahead of the curve this puts Europe.
Facial recognition in public has expanded almost everywhere in the Western world with essentially no legal framework around it.
Ese es un punto fundamental.
That is a fundamental point.
En Europa hay una tradición diferente.
In Europe there is a different tradition.
Los derechos fundamentales, la privacidad, la dignidad humana, no son solo palabras en una constitución.
Fundamental rights, privacy, human dignity, are not just words in a constitution.
Son valores que la gente espera que sus instituciones protejan activamente.
They are values that people expect their institutions to actively protect.
Y eso explica también por qué Europa fue la primera en aprobar el RGPD, el reglamento de protección de datos personales.
And that also explains why Europe was the first to pass the GDPR, the personal data protection regulation.
The GDPR, which came into force in 2018 and effectively rewrote how every company on earth handles personal data, whether they're European or not.
Apple changed its global privacy settings because of GDPR.
That's what people call the Brussels Effect, and it's the real reason this new law matters beyond Europe's borders.
Exactamente.
Exactly.
Cuando Europa crea una regulación, las empresas que quieren operar en el mercado europeo tienen que cumplirla.
When Europe creates a regulation, companies that want to operate in the European market have to comply with it.
Y como ese mercado tiene 450 millones de consumidores, muchas veces es más eficiente aplicar las mismas normas en todo el mundo que mantener sistemas diferentes para cada región.
And since that market has 450 million consumers, it is often more efficient to apply the same rules worldwide than to maintain different systems for each region.
It's regulatory gravity.
You can't walk away from a market that size, so you end up engineering to European standards even if your headquarters is in San Francisco.
The live question is whether that same gravitational pull will work for AI regulation the way it worked for data protection.
Es una pregunta legítima.
It's a legitimate question.
Y la ley también tiene reglas específicas para los sistemas de IA de alto riesgo.
And the law also has specific rules for high-risk AI systems.
Por ejemplo, los sistemas que se usan para contratar a empleados, para decidir si alguien recibe un crédito bancario, o para evaluar el rendimiento de estudiantes.
For example, systems used to hire employees, to decide whether someone receives a bank loan, or to evaluate student performance.
Estos sistemas tienen que ser transparentes, auditables y capaces de explicar sus decisiones.
These systems must be transparent, auditable, and capable of explaining their decisions.
Auditable and explainable.
That's genuinely radical, because right now a hiring algorithm can reject your resume and owe you absolutely no explanation.
The AI Act is saying: if a system is making decisions that affect people's lives, you have a legal right to understand how it works.
Y eso no es un detalle menor.
And that is not a minor detail.
Hay estudios que demuestran que algunos algoritmos de selección de personal discriminan por género o por origen étnico, simplemente porque fueron entrenados con datos históricos que reflejaban esas mismas discriminaciones.
There are studies showing that some recruitment algorithms discriminate by gender or ethnic origin, simply because they were trained on historical data that reflected those same discriminations.
La ley exige que las empresas detecten y corrijan esos problemas antes de lanzar el sistema.
The law requires companies to detect and correct those problems before launching the system.
Which raises a question I keep circling back to.
How do you audit a system that even its own creators don't fully understand?
Some of these large models, the people who built them will openly admit they can't always explain why the model produces a specific output.
It's not evasion, it's a genuine technical limitation.
Esa es la tensión central de toda la ley, y no tiene una respuesta fácil.
That is the central tension of the entire law, and it has no easy answer.
Hay mucha gente en el sector tecnológico que dice exactamente eso: que no pueden cumplir estas reglas porque la tecnología no funciona así.
Many people in the tech sector say exactly that: that they cannot comply with these rules because the technology doesn't work that way.
Pero los defensores de la ley responden algo muy simple: si no puedes explicar cómo funciona tu sistema, quizás no debería tomar decisiones sobre personas reales.
But the law's defenders respond with something very simple: if you can't explain how your system works, perhaps it shouldn't be making decisions about real people.
That's a philosophical disagreement that runs pretty deep.
In the United States, the default assumption tends to be: let the technology develop, let the market sort it out, regulate after things go wrong.
Europe is inverting that entirely.
Establish the rules first, then build inside them.
Sí, y esa diferencia tiene raíces históricas muy concretas.
Yes, and that difference has very concrete historical roots.
Europa ha vivido lo que ocurre cuando el Estado, o las empresas, tienen demasiado poder sobre las personas sin ningún tipo de control democrático.
Europe has experienced what happens when the state, or companies, have too much power over people without any democratic oversight.
Hay una memoria colectiva de lo que significa perder los derechos fundamentales.
There is a collective memory of what it means to lose fundamental rights.
Eso no es retórica política;
That is not political rhetoric;
es historia reciente.
it is recent history.
That's a point that gets consistently lost in the American coverage of European tech regulation, which usually frames it as cautious bureaucrats putting the brakes on innovation.
The historical context is completely real.
You're talking about a continent that was running surveillance states within living memory.
Exacto.
Exactly.
Ahora bien, los críticos también tienen argumentos serios que merecen respuesta.
Now, the critics also have serious arguments that deserve a response.
Europa no tiene un equivalente de Google, de Meta, de OpenAI.
Europe doesn't have an equivalent of Google, Meta, or OpenAI.
El sector tecnológico europeo existe, claro, pero no domina a nivel global.
The European tech sector exists, of course, but it doesn't dominate globally.
Y algunos analistas argumentan que una regulación tan exigente hace más difícil que las empresas europeas crezcan y compitan.
And some analysts argue that such demanding regulation makes it harder for European companies to grow and compete.
The innovation argument, and it carries real weight.
If you're running a startup in Berlin or Barcelona working with AI, and you're facing compliance costs and legal uncertainty that your competitor in Seattle simply doesn't have, that's a structural disadvantage.
I don't think you can just wave that away.
Es un punto válido, pero también hay que ser honesto sobre quién se queja más.
It's a valid point, but we also have to be honest about who is complaining the most.
La mayoría de las voces que critican la ley con más fuerza son las grandes empresas tecnológicas americanas, no las startups europeas.
The loudest voices criticizing the law are largely major American tech companies, not European startups.
Y para esas grandes empresas, el coste de cumplir la regulación es perfectamente manejable.
And for those big companies, the cost of complying with the regulation is perfectly manageable.
Lo que de verdad les molesta es la transparencia que exige la ley.
What really bothers them is the transparency the law demands.
That's fair.
And then there's the geopolitical layer underneath all of this, which I find genuinely fascinating.
Because you effectively have three competing models in play right now: Europe, regulating heavily to protect individual rights;
the United States, moving cautiously and leaving most of it to industry;
and China, deploying AI as an instrument of state power.
Y esa competencia es completamente real.
And that competition is completely real.
China invierte cantidades enormes en inteligencia artificial con el respaldo directo del Estado, sin las restricciones que tiene Europa.
China is investing enormous amounts in artificial intelligence with direct state backing, without the restrictions Europe has.
Hay analistas que dicen que Europa se está disparando en el pie.
There are analysts who say Europe is shooting itself in the foot.
Pero otros responden que ser el referente mundial en IA ética es también una forma de influencia y de poder.
But others respond that being the world reference for ethical AI is also a form of influence and power.
Soft power through regulatory standard-setting.
Which is precisely what happened with GDPR.
Europe wrote the rules, and then countries that had no legal obligation to adopt them started doing so anyway because the framework acquired international legitimacy.
Canada, Brazil, Japan, all moving in that direction.
La ley también tiene un régimen específico para los modelos de IA de propósito general, los que llaman en inglés GPAI.
The law also has a specific regime for general-purpose AI models, called GPAI in English.
Estos son los modelos grandes, como los que funcionan detrás de ChatGPT o Gemini.
These are the large models, like those behind ChatGPT or Gemini.
Tienen obligaciones de transparencia importantes, y entre otras cosas tienen que publicar información sobre los materiales con los que fueron entrenados.
They have significant transparency obligations, and among other things they must publish information about the materials used to train them.
Which is actually consequential, because one of the biggest unresolved legal battles right now is whether training these models on copyrighted material, books, journalism, creative work, constitutes infringement.
Requiring companies to disclose their training data drags that fight into daylight whether they want it there or not.
Y los modelos que se consideran de riesgo sistémico, porque son tan grandes y tan capaces que podrían tener efectos sobre sectores enteros de la sociedad, tienen obligaciones todavía mayores.
And models considered to pose systemic risk, because they are so large and capable that they could have effects on entire sectors of society, have even greater obligations.
Tienen que hacer evaluaciones de riesgos, cooperar con las autoridades europeas y someterse a auditorías independientes.
They must conduct risk assessments, cooperate with European authorities, and submit to independent audits.
The penalties are also worth being concrete about, because they're serious.
For the most severe violations, deploying a system that's flatly banned, the fine can reach 35 million euros or 7 percent of global annual turnover, whichever is the larger number.
For a company the size of Meta or Google, 7 percent of global turnover is not a rounding error.
Sí, y eso es una decisión consciente.
Yes, and that is a deliberate decision.
Las multas del RGPD también eran muy altas en teoría, pero la aplicación fue irregular y lenta en muchos casos.
GDPR fines were also very high in theory, but enforcement was uneven and slow in many cases.
Con la ley de IA, la Unión Europea ha creado una Oficina de Inteligencia Artificial específica para supervisar el cumplimiento.
With the AI Act, the European Union has created a dedicated AI Office to oversee compliance.
La pregunta real es si tendrá los recursos y la voluntad política para aplicarla con firmeza.
The real question is whether it will have the resources and the political will to enforce it firmly.
The enforcement gap is the thing that haunts every ambitious regulation I've ever covered.
Writing a good law is one challenge.
Building institutions that actually understand the technology well enough to police it is a completely different challenge.
Regulators are almost always running two steps behind the engineers.
Es un problema real, no lo voy a negar.
It's a real problem, I won't deny it.
Pero también hay que ver lo que la ley les da a los ciudadanos de forma concreta.
But you also have to look at what the law gives citizens in concrete terms.
Si un sistema de IA toma una decisión importante sobre ti, tienes derecho a recibir una explicación.
If an AI system makes an important decision about you, you have the right to receive an explanation.
Y tienes derecho a pedir que un ser humano revise esa decisión.
And you have the right to request that a human being review that decision.
Eso no existe con esta claridad en ningún otro lugar del mundo.
That doesn't exist with this clarity anywhere else in the world.
The right to a human review.
That lands differently the more you think about it, especially as these systems move into medical diagnostics, parole decisions, benefits eligibility.
The principle that a consequential decision about your life requires a human being in the loop is not a small thing to put into law.
No, no lo es.
No, it isn't.
Y creo que esa es la pregunta de fondo de toda esta ley: ¿quién tiene el control?
And I think that is the underlying question of this entire law: who is in control?
¿Las máquinas y las empresas que las construyen, o las personas y las instituciones democráticas que las representan?
The machines and the companies that build them, or the people and the democratic institutions that represent them?
Europa ha dado una respuesta muy clara.
Europe has given a very clear answer.
No es la única respuesta posible, pero es coherente y está fundamentada.
It is not the only possible answer, but it is coherent and grounded.
You used a construction just now that I want to ask you about, because it's one I hear from you constantly and I still can't produce it reliably myself.
You said "es importante que un ser humano revise." That verb, "revise," it doesn't sound like a regular present tense.
What's happening there?
Buena observación.
Good observation.
Ese "revise" está en subjuntivo.
That "revise" is in the subjunctive.
Cuando usas expresiones como "es importante que", "quiero que", "es necesario que", el verbo que viene después tiene que estar en subjuntivo.
When you use expressions like "es importante que", "quiero que", "es necesario que", the verb that follows must be in the subjunctive.
Así que "revisar" se convierte en "revise" en la tercera persona.
So "revisar" becomes "revise" in the third person.
Es diferente del indicativo, que sería "revisa." Escucha la diferencia: "el humano revisa" frente a "es importante que el humano revise."
It's different from the indicative, which would be "revisa." Listen to the difference: "el humano revisa" versus "es importante que el humano revise."
So the phrase "es importante que" is essentially a trigger.
Once you say it, the next verb has to change form.
The indicative "revisa" becomes the subjunctive "revise." And I assume this works the same way with other expressions, "es posible que," "es necesario que," that whole family.
Exactamente, es eso.
Exactly, that's it.
"Es posible que las empresas cumplan la ley." "Es necesario que los reguladores entiendan la tecnología." El subjuntivo aparece siempre que hay una valoración, una duda, un deseo, algo que no es simplemente un hecho declarado.
"Es posible que las empresas cumplan la ley." "Es necesario que los reguladores entiendan la tecnología." The subjunctive appears whenever there is an evaluation, a doubt, a wish, something that is not simply a stated fact.
Es una de esas cosas que con el tiempo deja de ser una decisión consciente y se convierte en un instinto.
It's one of those things that over time stops being a conscious decision and becomes an instinct.
Aunque reconozco que ese proceso lleva bastante tiempo.
Though I'll admit that process takes quite a while.
An instinct.
Right.
Well, I've been at this long enough that I should probably have that instinct by now, and I absolutely do not.
But I'll tell you what, the next time I try to say "es importante que" at a family dinner in Madrid and the verb that follows is spectacularly wrong, at least now I'll know precisely what rule I'm violating.