B1 · Intermediate
16 min
technologyhistoryculturemediasociety
De Siri a ChatGPT: la revolución de la inteligencia artificial
From Siri to ChatGPT: the AI revolution
Fletcher breaks down this story in English. Octavio reacts and expands in Spanish. Follow along with the live transcript, tap any word for its translation. Intermediate level — perfect for intermediate learners expanding their range.
So I want to start with a confession.
The first time I used Siri, I was in a cab in Chicago, it was 2011, and I asked it for the nearest Thai restaurant.
It sent me to a dry cleaner.
And I remember thinking, okay, this is either going to be nothing or it's going to be everything.
Bueno, mira, yo tuve una experiencia similar.
Well, look, I had a similar experience.
Usé Siri por primera vez en Madrid, le pregunté el nombre de una calle, y me respondió en inglés.
I used Siri for the first time in Madrid, I asked it the name of a street, and it answered me in English.
En Madrid.
In Madrid.
Pensé, 'esto es una broma'.
I thought, 'this is a joke'.
Which, honestly, tells you everything about how those early systems worked.
Or didn't.
But here's the thing, Siri launched in October 2011, and Apple called it an 'intelligent personal assistant,' which at the time felt like science fiction vocabulary applied to something that couldn't find a Thai restaurant.
Es que el problema era simple: Siri no entendía el contexto.
The thing is, the problem was simple: Siri didn't understand context.
Podía responder preguntas muy básicas, pero no entendía lo que la gente realmente quería.
It could answer very basic questions, but it didn't understand what people actually wanted.
Era como hablar con alguien que solo escucha las primeras palabras de cada frase.
It was like talking to someone who only listens to the first words of every sentence.
Right, and that's the key distinction that I think gets lost in how we tell this story.
The gap between what Siri was and what ChatGPT is, that's not just a technical upgrade.
That's a conceptual leap.
These are almost different categories of thing.
Exactamente.
Exactly.
Siri buscaba información.
Siri searched for information.
ChatGPT genera información nueva, razona, explica.
ChatGPT generates new information, it reasons, it explains.
La diferencia es enorme.
The difference is enormous.
Es como comparar una enciclopedia con un profesor.
It's like comparing an encyclopedia to a teacher.
I love that comparison.
Though I'd push back slightly and say, a teacher who sometimes confidently tells you the wrong thing, and doesn't know it.
[laughs] Sí, sí.
[laughs] Yes, yes.
A ver, eso es un problema muy serio.
Look, that's a very serious problem.
Pero antes de hablar de los problemas, creo que es importante entender cómo llegamos aquí.
But before we talk about the problems, I think it's important to understand how we got here.
Porque no empezó con Siri.
Because it didn't start with Siri.
No, absolutely not.
And this is where the history gets genuinely fascinating.
Because people think this is a story that started maybe fifteen years ago, but the intellectual roots go back much further.
To the 1950s, really, to Alan Turing and his famous question: can machines think?
Bueno, y en 1966 apareció ELIZA.
Well, and in 1966 ELIZA appeared.
Era un programa muy primitivo, pero simulaba una conversación.
It was a very primitive program, but it simulated a conversation.
El creador, Joseph Weizenbaum, quedó muy sorprendido porque la gente hablaba con ELIZA como si fuera una persona real.
The creator, Joseph Weizenbaum, was very surprised because people talked to ELIZA as if it were a real person.
Incluso su secretaria le pidió privacidad para hablar con el programa.
Even his secretary asked him for privacy to talk to the program.
The extraordinary thing is that ELIZA was, essentially, a mirror.
It reflected your words back at you, reframed as questions.
It had no understanding of anything.
And yet people formed emotional attachments to it.
That tells you something pretty uncomfortable about human psychology.
La verdad es que sí.
Honestly, yes.
Y Weizenbaum escribió un libro después, en 1976, donde criticó mucho esta tendencia.
And Weizenbaum wrote a book afterwards, in 1976, where he strongly criticized this tendency.
Dijo que era peligroso confiar en las máquinas para cosas emocionales.
He said it was dangerous to trust machines with emotional matters.
Era un hombre muy inteligente y muy preocupado.
He was a very intelligent and very worried man.
And his warnings went largely unheeded, which, look, is kind of the recurring theme of this whole story.
But let's fast-forward a bit, because between ELIZA and Siri there were decades of AI winters, these periods where the field basically collapsed because the technology couldn't deliver what researchers promised.
Sí, hubo dos grandes crisis, una en los años setenta y otra en los ochenta.
Yes, there were two major crises, one in the seventies and another in the eighties.
Los gobiernos dejaron de financiar la investigación porque los resultados eran muy malos.
Governments stopped funding research because the results were very bad.
Pero en los años noventa y dos mil, con internet y los datos masivos, todo cambió.
But in the nineties and two thousands, with the internet and massive data, everything changed.
Right, and that's the piece I think most people miss.
The reason modern AI works isn't just because the algorithms got smarter.
It's because suddenly there was an ocean of human-generated data to train on.
Every search, every email, every webpage.
We built the training set without knowing we were doing it.
Mira, eso es muy importante.
Look, that's very important.
Nosotros, los usuarios, escribimos millones de textos en internet, y las empresas usaron esos textos para entrenar sus sistemas.
We, the users, wrote millions of texts on the internet, and companies used those texts to train their systems.
En cierto modo, los humanos enseñaron a las máquinas a hablar.
In a way, humans taught the machines to speak.
Which raises a question I find genuinely thorny, which is: who owns that?
All those blog posts, forum arguments, novels, Wikipedia edits.
The people who wrote them didn't consent to training an AI.
That debate is very much live right now.
Es que es un problema enorme.
The thing is, it's an enormous problem.
Muchos escritores y periodistas están muy enfadados.
Many writers and journalists are very angry.
El New York Times, por ejemplo, demandó a OpenAI el año pasado porque usaron sus artículos sin permiso.
The New York Times, for example, sued OpenAI last year because they used their articles without permission.
Y yo lo entiendo perfectamente, la verdad.
And I understand it completely, honestly.
As a journalist, so do I.
I mean, I've spent thirty years producing content that apparently helped teach a machine to sound like a journalist.
There's something a little vertiginous about that.
But, okay, let's get to November 2022, because that's when the story really cracks open.
ChatGPT llegó al mundo el 30 de noviembre de 2022.
ChatGPT arrived in the world on November 30th, 2022.
En cinco días, tuvo un millón de usuarios.
In five days, it had one million users.
En dos meses, cien millones.
In two months, one hundred million.
Ninguna aplicación en la historia creció tan rápido.
No application in history grew so fast.
Facebook necesitó cuatro años para llegar a cien millones.
Facebook needed four years to reach one hundred million.
I remember the week it launched.
My students at UT Austin were talking about nothing else.
And within a month I was getting essays that were, let's say, suspiciously coherent for students who'd been struggling all semester.
The implications for education hit fast.
A ver, en España pasó lo mismo.
Look, in Spain the same thing happened.
Los profesores estaban muy preocupados.
Teachers were very worried.
Pero yo creo que el problema más interesante no es el plagio.
But I think the most interesting problem isn't plagiarism.
El problema es que la gente empezó a pensar de manera diferente.
The problem is that people started thinking differently.
Empezaron a preguntar a la máquina antes de pensar ellos mismos.
They started asking the machine before thinking for themselves.
No, you're absolutely right about that.
And there's a parallel there with what happened to memory when GPS arrived.
We stopped memorizing routes because we didn't need to.
The question is whether outsourcing reasoning is qualitatively different from outsourcing navigation.
Para mí, sí es diferente.
For me, yes it is different.
La navegación es una habilidad técnica.
Navigation is a technical skill.
El pensamiento crítico es algo más fundamental.
Critical thinking is something more fundamental.
Es la base de la democracia, del periodismo, de la ciencia.
It's the basis of democracy, journalism, science.
Si perdemos eso, perdemos algo muy importante.
If we lose that, we lose something very important.
Here's what gets me, though.
I've watched people use ChatGPT in ways that actually sharpen their thinking.
Using it to stress-test an argument, to find the holes in their own reasoning.
It's not automatically a crutch.
It depends entirely on how you engage with it.
Bueno, eso es verdad.
Well, that's true.
Yo mismo usé ChatGPT para revisar uno de mis artículos.
I myself used ChatGPT to review one of my articles.
Le pedí que buscara errores en mi argumento.
I asked it to look for errors in my argument.
Y encontró dos puntos débiles que yo no vi.
And it found two weak points that I didn't see.
Fue útil.
It was useful.
Pero también me sentí un poco raro después.
But I also felt a little strange afterwards.
A little strange how?
Es que...
The thing is...
no sé.
I don't know.
Sentí que la máquina entendió mi argumento mejor que algunos colegas.
I felt that the machine understood my argument better than some colleagues.
Y eso fue incómodo.
And that was uncomfortable.
Porque si una máquina puede hacer eso, ¿qué hace exactamente un editor?
Because if a machine can do that, what exactly does an editor do?
¿Qué hago yo?
What do I do?
I think that question, 'what do I do,' is the most honest and the most widespread anxiety of the last two years.
And it's hitting every profession differently.
Law, medicine, coding, journalism.
The white-collar world that thought it was safe from automation suddenly isn't.
Mira, en España muchos periodistas perdieron su trabajo en los últimos dos años.
Look, in Spain many journalists lost their jobs in the last two years.
No solo por la inteligencia artificial, también por la crisis de los medios.
Not only because of artificial intelligence, but also because of the media crisis.
Pero la IA aceleró el proceso.
But AI accelerated the process.
Algunas empresas ya usaban programas para escribir noticias financieras y deportivas automáticamente.
Some companies were already using programs to write financial and sports news automatically.
And that's been happening in the US too.
The AP has been using automated writing for earnings reports since 2014.
Which, look, those are formulaic texts.
But the line between formulaic and not-formulaic is moving fast.
La verdad es que a mí me preocupa más la desinformación que el trabajo.
Honestly, I'm more worried about disinformation than about jobs.
Porque ahora es muy fácil crear textos falsos, imágenes falsas, videos falsos.
Because now it's very easy to create fake texts, fake images, fake videos.
En las elecciones españolas de 2023, hubo muchos casos de imágenes falsas con inteligencia artificial.
In the Spanish elections of 2023, there were many cases of fake images made with artificial intelligence.
Fue muy preocupante.
It was very worrying.
This is the one that keeps me up at night.
I covered elections in countries where information warfare was a tool of state power.
And the thing about AI-generated disinformation is that it democratizes that capability.
You don't need a state budget anymore.
You need a laptop and a subscription.
Exactamente.
Exactly.
Y el problema es que la gente no sabe distinguir lo real de lo falso.
And the problem is that people don't know how to distinguish the real from the fake.
Antes, una foto era una prueba.
Before, a photo was proof.
Ahora, una foto no significa nada.
Now, a photo means nothing.
Eso es un cambio enorme en nuestra relación con la realidad.
That's an enormous change in our relationship with reality.
I want to bring in another angle here, because I think the cultural dimension gets underplayed.
This technology didn't arrive in a vacuum.
It arrived in a moment of profound distrust in institutions, in media, in expertise.
And it landed differently in different languages and cultures.
Sí, y aquí hay un problema específico con el español.
Yes, and here there's a specific problem with Spanish.
Los modelos de inteligencia artificial fueron entrenados principalmente con textos en inglés.
AI models were trained mainly with texts in English.
Entonces, el inglés funciona mejor.
So English works better.
El español funciona, pero con más errores, especialmente el español de América Latina.
Spanish works, but with more errors, especially Latin American Spanish.
Which is a real equity issue.
If the most powerful cognitive tools of the next century work better in English than in Spanish, or Swahili, or Bengali, then you're building a new kind of inequality into the infrastructure of knowledge itself.
A ver, hay iniciativas para mejorar esto.
Look, there are initiatives to improve this.
En España, el gobierno y algunas universidades trabajaron para crear modelos de IA específicamente en español.
In Spain, the government and some universities worked to create AI models specifically in Spanish.
Se llama el proyecto MarIA.
It's called the MarIA project.
Fue un esfuerzo importante, pero los recursos son mucho menores que los de OpenAI o Google.
It was an important effort, but the resources are much smaller than those of OpenAI or Google.
I mean, that resource gap is staggering.
OpenAI reportedly spent over a hundred million dollars training GPT-4.
That's not a competition most governments or universities can enter.
Bueno, y eso nos lleva a la pregunta del poder.
Well, and that takes us to the question of power.
Porque ahora, cuatro o cinco empresas americanas controlan la tecnología más importante del mundo.
Because now, four or five American companies control the most important technology in the world.
Microsoft, Google, OpenAI, Meta.
Microsoft, Google, OpenAI, Meta.
Esto es muy diferente a lo que pasó con el internet, que fue más abierto al principio.
This is very different from what happened with the internet, which was more open at the beginning.
Right, and Europe has been trying to respond to that.
The EU AI Act, which passed last year, is the most ambitious attempt to regulate this technology.
Though I've read it, and it is, let's say, an admirably complex document.
Es que la regulación siempre llega tarde, ¿verdad?
The thing is, regulation always arrives late, right?
Primero viene la tecnología, después vienen los problemas, y al final llega la ley.
First comes the technology, then come the problems, and finally comes the law.
Con las redes sociales pasó lo mismo.
The same thing happened with social media.
Facebook existió diez años antes de que los políticos entendieron realmente qué era.
Facebook existed for ten years before politicians really understood what it was.
Look, I've sat in congressional hearings where senators asked Mark Zuckerberg how Facebook makes money, with apparent genuine curiosity.
So yes, the regulatory gap is real.
But I don't think the answer is just 'regulate faster.' The technology is genuinely hard to understand, even for people who build it.
Sí, eso es verdad.
Yes, that's true.
Geoffrey Hinton, uno de los científicos más importantes en inteligencia artificial, dejó Google en 2023 y dijo que tenía miedo de lo que ayudó a crear.
Geoffrey Hinton, one of the most important scientists in artificial intelligence, left Google in 2023 and said he was afraid of what he helped create.
Cuando los creadores tienen miedo, es una señal muy seria.
When the creators are afraid, that's a very serious sign.
The extraordinary thing about Hinton's departure is that he's not a person given to panic.
This is someone who spent his career arguing that neural networks could work when most of the field thought he was wrong.
His alarm carries weight precisely because of where it comes from.
Pero también hay personas muy inteligentes que son más optimistas.
But there are also very intelligent people who are more optimistic.
Dicen que la IA puede ayudar a resolver problemas enormes: el cambio climático, las enfermedades, la pobreza.
They say that AI can help solve enormous problems: climate change, diseases, poverty.
Y ya vemos resultados, por ejemplo en la investigación médica.
And we already see results, for example in medical research.
La IA descubrió nuevas proteínas que los científicos buscaban hace décadas.
AI discovered new proteins that scientists had been searching for for decades.
AlphaFold.
Yes.
That's a genuinely extraordinary achievement.
DeepMind's system solved the protein folding problem, which biologists had been working on for fifty years.
The potential medical applications are real and enormous.
That's not hype.
Entonces, la pregunta no es si la inteligencia artificial es buena o mala.
So, the question isn't whether artificial intelligence is good or bad.
La pregunta es: ¿quién decide cómo se usa?
The question is: who decides how it's used?
¿Las empresas?
Companies?
¿Los gobiernos?
Governments?
¿La sociedad?
Society?
Esa conversación todavía no ocurrió de verdad.
That conversation hasn't really happened yet.
And that's maybe the cleanest summary of where we are.
We've been having the wrong argument, good versus bad, and missing the real one, which is about power and accountability.
Octavio, I'll give you the last word.
When you think about your daughter's generation growing up with this, what do you actually feel?
La verdad es que siento las dos cosas al mismo tiempo.
Honestly, I feel both things at the same time.
Miedo, porque el mundo que ella va a vivir es muy incierto.
Fear, because the world she's going to live in is very uncertain.
Y también esperanza, porque ella creció con esta tecnología y la usa de una manera más natural y más crítica que nosotros.
And also hope, because she grew up with this technology and uses it in a more natural and more critical way than we do.
Quizás la generación que creció con esto es la mejor preparada para controlarlo.
Maybe the generation that grew up with this is the best prepared to control it.
That's either very reassuring or the most optimistic thing you've ever said to me.
I genuinely can't tell.
From Siri giving me directions to a dry cleaner to machines that can reason about proteins and write legislation summaries, it's been quite a ride.
Thanks for walking through it with us today.