El 40% de los resúmenes científicos en oncología publicados en 2024 muestran huellas de inteligencia artificial, según un análisis reciente. Este dato, extraído de un estudio liderado por Dmitry Kobak del Hertie Institute y publicado en Science Advances, revela que la presencia de IA en la redacción científica ya no es una sospecha ni un rumor universitario: es una realidad cuantificable. 

El trabajo, que ha examinado más de 15 millones de textos científicos, identifica que al menos el 13,5% de los resúmenes biomédicos de 2024 presentan patrones lingüísticos característicos de herramientas como ChatGPT. 

Investigadores identifican características de los textos generados por IA 

El método desarrollado por los investigadores consiste en rastrear una “huella digital” basada en el uso repetido de ciertos términos y estructuras que los modelos de lenguaje emplean con mayor frecuencia que los autores humanos. Palabras como “además”, “también” o “sin embargo” aparecen de forma anómala en los textos generados por IA, lo que permite detectar su intervención entre millones de documentos. 

Este enfoque ha permitido identificar diferencias notables entre especialidades: oncología encabeza la lista con un 40% de resúmenes generados o editados por IA, seguida de neurología con un 35% y genómica con un 30%. Además, los países con mayor adopción tecnológica lideran los porcentajes de uso de IA en sus publicaciones científicas. 

El estudio, publicado en Science Advances, subraya que la transformación impulsada por la inteligencia artificial no se limita a la generación de imágenes o textos, sino que se extiende a campos como la biotecnología y la salud.

Sin embargo, solo el 2% de los artículos analizados declara explícitamente el uso de IA, lo que plantea interrogantes sobre la transparencia en la investigación científica. Esta falta de declaración alimenta el debate sobre los riesgos y beneficios de la automatización en la ciencia.

Controversia por el uso de IA en publicaciones científicas

Las opiniones sobre el impacto de la IA en la redacción científica están divididas. Quienes la defienden sostienen que agiliza la publicación de resultados y mejora la claridad del lenguaje, especialmente para investigadores cuya lengua materna no es el inglés.

Por el contrario, los críticos advierten sobre la posible pérdida de creatividad y la homogeneización del discurso científico. El debate se intensifica ante antecedentes que sugieren que el uso intensivo de IA podría reducir la conectividad neuronal, lo que suscita dudas sobre la originalidad intelectual.

No obstante, existen ejemplos positivos, como el desarrollo de sistemas españoles capaces de diagnosticar cáncer con una precisión inédita, que demuestran el potencial de estas herramientas.

La percepción social añade otra capa de complejidad. Según el portal web La Razón de España, más del 25% de los españoles desconfía de investigaciones científicas elaboradas con ayuda de la IA, lo que refuerza la necesidad de transparencia para mantener la confianza pública en la ciencia.

Los autores del estudio no abogan por prohibir la inteligencia artificial en la redacción científica, pero sí por establecer normas claras. Entre sus recomendaciones figuran la obligación de declarar el uso de modelos de IA en las metodologías, la validación crítica de los textos generados y la reducción de la dependencia de frases prefabricadas.

Los datos del estudio están disponibles en plataformas como PubMed, y el código empleado se encuentra en GitHub, lo que permite a otros investigadores replicar el análisis en distintas disciplinas científicas.

Finalmente, es pertinente señalar que la incorporación de la inteligencia artificial en la ciencia está transformando la manera en que se generan y analizan conocimientos. Herramientas basadas en IA permiten procesar enormes volúmenes de datos, detectar patrones complejos y automatizar tareas repetitivas, lo que acelera investigaciones y facilita descubrimientos.

40% of scientific abstracts in oncology published in 2024 show traces of artificial intelligence, according to a recent analysis. This data, taken from a study led by Dmitry Kobak of the Hertie Institute and published in Science Advances, reveals that the presence of AI in scientific writing is no longer a suspicion or a university rumor: it is a quantifiable reality. 

The work, which examined more than 15 million scientific texts, identifies that at least 13.5% of biomedical abstracts from 2024 exhibit linguistic patterns characteristic of tools such as ChatGPT.

Researchers identify characteristics of AI-generated texts

The method developed by the researchers involves tracking a "fingerprint" based on the repeated use of certain terms and structures that language models employ more frequently than human authors. Words like "in addition," "also," and "however" appear anomalously in AI-generated texts, allowing their use to be detected among millions of documents.

This approach has allowed for the identification of notable differences between specialties: oncology tops the list with 40% of abstracts generated or edited by AI, followed by neurology with 35% and genomics with 30%. Furthermore, countries with the highest technological adoption lead the percentages of AI use in their scientific publications.

The study, published in Science Advances, highlights that the transformation driven by artificial intelligence is not limited to the generation of images or texts, but extends to fields such as biotechnology and healthcare.

However, only 2% of the analyzed articles explicitly declare the use of AI, raising questions about transparency in scientific research. This lack of declaration fuels the debate on the risks and benefits of automation in science.

Controversy over the use of AI in scientific publications

󠀰Opinions on the impact of AI on scientific writing are divided.󠀲󠀢󠀡󠀳 󠀰Supporters argue that it speeds up the publication of results and improves language clarity, especially for researchers whose native language is not English.󠀲󠀢󠀢󠀳

󠀰On the contrary, critics warn about the possible loss of creativity and the homogenization of scientific discourse.󠀲󠀢󠀣󠀳 󠀰The debate is intensifying in light of evidence suggesting that intensive use of AI could reduce neural connectivity, raising questions about intellectual originality.

󠀰However, there are positive examples, such as the development of Spanish systems capable of diagnosing cancer with unprecedented accuracy, demonstrating the potential of these tools.󠀲󠀢󠀧󠀳

󠀰Social perception adds another layer of complexity.󠀲󠀢󠀨󠀳 󠀰According to the Spanish web portal La Razón, more than 25% of Spaniards distrust scientific research conducted with the help of AI, reinforcing the need for transparency to maintain public trust in science.󠀲󠀢󠀩󠀳

󠀰The study's authors do not advocate banning artificial intelligence in scientific writing , but rather establishing clear rules.󠀲󠀣󠀠󠀳 󠀰Their recommendations include the obligation to declare the use of AI models in methodologies, critical validation of generated texts, and reducing reliance on prefabricated phrases.

󠀰The study data is available on platforms such as PubMed, and the code used is on GitHub, allowing other researchers to replicate the analysis across different scientific disciplines.󠀲󠀣󠀣󠀳

󠀰Finally, it is pertinent to point out that the incorporation of artificial intelligence in science is transforming the way knowledge is generated and analyzed.󠀲󠀣󠀤󠀳 󠀰AI-based tools allow for the processing of enormous volumes of data, detecting complex patterns, and automating repetitive tasks, which accelerates research and facilitates discoveries.󠀲󠀣󠀥󠀳

Forty percent of scientific abstracts in oncology published in 2024 show traces of artificial intelligence, according to a recent analysis. This data, taken from a study led by Dmitry Kobak of the Hertie Institute and published in Science Advances, reveals that the presence of AI in scientific writing is no longer a suspicion or an academic rumor: it is a quantifiable reality. 

The study, which examined more than 15 million scientific texts, identifies that at least 13.5% of biomedical abstracts from 2024 exhibit linguistic patterns characteristic of tools such as ChatGPT. 

Researchers identify characteristics of AI-generated texts 

The method developed by the researchers consists of tracking a “digital fingerprint” based on the repeated use of certain terms and structures that language models use more frequently than human authors. Words such as “in addition,” “also,” or “however” appear abnormally in AI-generated texts, making it possible to detect their intervention among millions of documents. 

This approach has made it possible to identify notable differences between specialties: oncology tops the list with 40% of abstracts generated or edited by AI, followed by neurology with 35% and genomics with 30%. In addition, countries with the highest technological adoption lead the way in the percentage of AI use in their scientific publications. 

The study, published in Science Advances, emphasizes that the transformation driven by artificial intelligence is not limited to the generation of images or text, but extends to fields such as biotechnology and healthcare.

However, only 2% of the articles analyzed explicitly state the use of AI, raising questions about transparency in scientific research. This lack of disclosure fuels the debate about the risks and benefits of automation in science.

Controversy over the use of AI in scientific publications

Opinions on the impact of AI on scientific writing are divided. Those who defend it argue that it speeds up the publication of results and improves the clarity of language, especially for researchers whose native language is not English.

On the other hand, critics warn of a possible loss of creativity and the homogenization of scientific discourse. The debate is intensifying in light of evidence suggesting that intensive use of AI could reduce neural connectivity, raising questions about intellectual originality.

However, there are positive examples, such as the development of Spanish systems capable of diagnosing cancer with unprecedented accuracy, which demonstrate the potential of these tools.

Social perception adds another layer of complexity. According to the Spanish website La Razón, more than 25% of Spaniards distrust scientific research carried out with the help of AI, reinforcing the need for transparency to maintain public confidence in science.

The authors of the study do not advocate banning artificial intelligence in scientific writing, but they do advocate establishing clear rules. Their recommendations include the obligation to declare the use of AI models in methodologies, critical validation of generated texts, and reducing dependence on prefabricated phrases.

The study data is available on platforms such as PubMed, and the code used can be found on GitHub, allowing other researchers to replicate the analysis in different scientific disciplines.

Finally, it is worth noting that the incorporation of artificial intelligence into science is transforming the way knowledge is generated and analyzed. AI-based tools make it possible to process enormous volumes of data, detect complex patterns, and automate repetitive tasks, which speeds up research and facilitates discoveries.