More than half think the technology they are building could replace human beings
En los meses transcurridos desde que ChatGPT irrumpió en la vida cotidiana, la inteligencia artificial ha dejado de ser una abstracción para convertirse en una fuerza que moldea el trabajo, la opinión pública y, según quienes la construyen, quizás el destino de la especie. Una encuesta de NewScientist a casi 3.000 investigadores revela no un consenso tranquilizador, sino un mapa de esperanzas y temores en tensión: la mayoría de los expertos reconoce riesgos existenciales mientras insiste en que aún son manejables, una paradoja que dice tanto sobre la naturaleza humana como sobre la tecnología misma.
- Más de la mitad de los principales investigadores en IA del mundo estiman que existe un 5% de probabilidad de que su propia creación conduzca a la extinción humana, una cifra pequeña con implicaciones enormes.
- Las capacidades se multiplican a velocidad desconcertante: sistemas que resuelven olimpiadas matemáticas, detectan plástico en los océanos y sustituyen tareas de oficina, todo en el lapso de meses.
- El horizonte que dibuja la encuesta es concreto y perturbador: IA sobrehumana para 2047, automatización total del trabajo para 2116, y ya hoy más del 70% de los expertos advierte sobre manipulación de opinión pública y diseño de armas.
- La profesión entera se debate entre la urgencia de crear marcos de seguridad y la inercia de una carrera tecnológica que avanza más rápido que la capacidad colectiva de regularla.
Cuando Sam Altman lanzó ChatGPT a finales de 2022, la inteligencia artificial dejó de ser una preocupación teórica lejana para convertirse en algo que cualquiera podía tocar. En los meses siguientes, los sistemas aprendieron a acompañar a personas mayores, detectar patrones en enormes conjuntos de datos, identificar plástico en los océanos y gestionar el trabajo repetitivo de las oficinas. Google DeepMind construyó AlphaGeometry, capaz de resolver problemas de geometría al nivel de un campeón olímpico. Las capacidades no dejaron de multiplicarse.
Ahora, una encuesta de NewScientist a casi 3.000 investigadores en IA —personas que han publicado trabajo serio sobre el tema en los últimos meses— ha añadido una nueva capa a la conversación. Lo que emergió no fue consenso, sino un paisaje de ansiedades y esperanzas en competencia. El 58% de los encuestados cree que existe un 5% de probabilidad de que la IA conduzca a la extinción de la humanidad. El porcentaje de riesgo es pequeño, pero lo que representa es mayor: más de la mitad de los investigadores líderes del mundo piensa que la tecnología que están construyendo podría, bajo algún escenario, reemplazar a los seres humanos por completo.
La encuesta también sondeó expectativas a largo plazo. La mitad de los investigadores predice que existirá una IA sobrehumana —sistemas que superen a los humanos en la mayoría de los dominios— para 2047, y que el trabajo estará completamente automatizado para 2116. Pero quizás lo más llamativo es lo que preocupa a los investigadores antes de que se alcancen esos hitos lejanos: más del 70% advierte que la tecnología ya podría manipular la opinión pública a escala y ser utilizada para diseñar armas. No son daños hipotéticos, sino capacidades que podrían emerger de sistemas que se están construyendo ahora mismo. La encuesta retrata a una profesión que lidia con haber creado algo lo suficientemente poderoso como para remodelar la sociedad, consciente de que esa remodelación ya ha comenzado.
When Sam Altman released ChatGPT to the world in late 2022, artificial intelligence stopped being a distant theoretical concern and became something everyone could touch. In the months that followed, the technology evolved with startling speed. AI systems learned to mimic human performance through neural networks, to keep elderly people company, to spot patterns in massive datasets, to detect plastic choking the oceans, to handle the repetitive work that fills office days. Google DeepMind built AlphaGeometry, a system that solves geometry problems at the level of a mathematical olympiad champion. The capabilities kept multiplying.
Yet as the technology advanced, warnings came too. Researchers at the Center for AI Safety began flagging potential dangers. Now, a survey by NewScientist has added another layer to the conversation—one that reveals how fractured expert opinion actually is. Nearly 3,000 AI researchers, people who have published serious work on the subject in recent months, answered questions about what artificial intelligence might do to the world. What emerged was not consensus but a landscape of competing anxieties and hopes.
When asked about the existential stakes, 58 percent of respondents said they believed there was a 5 percent chance that AI could drive humanity to extinction. The number itself is small—five percent—but what it represents is larger. More than half of the world's leading AI researchers think the technology they are building could, under some scenario, replace human beings entirely. Some of those same researchers were quick to add that alarm is premature, that such an outcome remains impossible given where the technology stands today. But the fact that they felt compelled to make that caveat suggests the question itself no longer feels absurd.
The survey probed deeper into what researchers expect from AI in the coming decades. Half believe the technology will eventually complete tasks entirely on its own, without needing a human to step in and correct course. Half predict that superhuman artificial intelligence—systems that outperform humans across most domains—will exist by 2047. Another half think that by 2116, work will be completely automated, that human labor will have become optional. These timelines are not certainties in the researchers' minds, but they are plausible enough to name.
Perhaps most striking is what researchers worry AI could do before it ever reaches those distant milestones. More than 70 percent said the technology could manipulate public opinion at scale and could be used to design weapons. These are not hypothetical harms. They are capabilities that could emerge from systems we are building right now, in the next few years, not centuries from now. The survey captures a profession grappling with the fact that it has created something powerful enough to reshape society, and that the reshaping has already begun.
Notable Quotes
Some researchers noted that while extinction risk exists as a possibility, current conditions make such an outcome impossible at present— Survey respondents
The Hearth Conversation Another angle on the story
When you read that 58 percent of experts think there's a 5 percent extinction risk, what does that number actually mean to you?
It means more than half the people building this technology think it's possible—genuinely possible—that it could end us. But they're also saying it's unlikely right now. It's the difference between "this is a real risk" and "this is an imminent threat."
So why does the distinction matter?
Because it shapes how we act. If you think something is impossible, you don't prepare for it. If you think it's a small probability, you start building safeguards. The survey suggests researchers are in that second camp.
What about the timeline predictions—superhuman AI by 2047, full automation by 2116? Do those feel real to the people saying them?
They're saying it's plausible enough to think about seriously. Whether they believe it will actually happen is different from whether they think it could. The survey doesn't quite separate those two things.
And the weapons, the manipulation of public opinion—those feel more immediate?
Much more. Those aren't 2047 problems. Those are things that could happen in the next five to ten years, with systems we're already building. That's what makes 70 percent of researchers flagging those risks so significant.