Weekly AI Watch

Every week, what changed and what it means for you. No hurry, no alarm.

Week of 13 November 2026 | Past editions15,811 papers found · 14,944 read · 4,785 belonged · 46 stories passed every check

Observatorio de la
Inteligencia Artificial

Cada semana, qué cambió y qué significa para usted. Sin prisa y sin alarma.

Semana del 13 de noviembre de 2026 | Ediciones anteriores15,811 estudios encontrados · 14,944 leídos · 4,785 correspondían · 46 notas pasaron todas las comprobaciones

weeklyAI · Week of 13 November 2026weeklyAI · Semana del 13 de noviembre de 2026

← The good news← Las buenas noticias

other · NPJ digital medicine · la publicación, 9 jul 2026 · gratis

Un programa revisó 1.3 millones de expedientes y encontró 13 casos de una enfermedad rara de la sangre

El hallazgo es de hospitales de Polonia y sólo se probó a un tercio de las personas que el programa marcó, así que las cifras son estimaciones.

Versión breve · la versión detallada sigue, unos 7 min

Pregunte al chat de weeklyAI

Pregúnteme sobre la IA en su vida: le señalo las notas que weeklyAI ha publicado sobre el tema, esta semana y antes.

Las conversaciones se guardan mientras exista weeklyAI, para mejorar la publicación. Se responde en el idioma en que usted escribe.

El estudio, de un vistazo
Quiénes
pacientes de hospitales en Polonia
Cuántos
1,307,140 pacientes
Dónde
Polonia
Cuándo
septiembre de 2023 a noviembre de 2024
Tipo de estudio
analysis of what people did
Quién lo hizo
Saventic Health y AstraZeneca, con hospitales de Polonia
El límite que importa
Solo un tercio de los marcados se hizo la prueba de confirmación; las cifras son estimaciones.

Aciertos del programa frente a una revisión convencional anterior

programa de inteligencia artificial (13 de 119)frente arevisión convencional anterior (6.9 de cada 100)

El programa acertó en 10.92% de los enviados a la prueba, frente al 6.9% de la revisión convencional.

Si alguien vuelve una y otra vez al hospital con anemia, cansancio o un problema de la sangre sin nombre, ese ir y venir es justo el patrón que este programa aprendió a notar.
Lectura de weeklyAI

Trece personas. Ese es el número que dejó un programa de inteligencia artificial en Polonia tras leer los expedientes de 1,307,140 pacientes de 14 instituciones de salud, entre septiembre de 2023 y noviembre de 2024.

La enfermedad es la hemoglobinuria paroxística nocturna, un trastorno raro de la sangre que puede tardar años en recibir nombre: en uno de cada cuatro casos, cinco años o más. El programa buscaba señales en los campos codificados del expediente y también en el texto libre de las notas clínicas.

Marcó a 356 pacientes como de alto riesgo. Los médicos decidieron enviar a 119 a la prueba de confirmación, la citometría de flujo. De esos 119, 13 tenían la enfermedad: uno de cada nueve. La revisión recoge un estudio anterior en otro centro polaco en el que la cifra fue de casi siete de cada cien.

Lo más llamativo está en los expedientes de 10 de esos 13. Al revisar sus visitas pasadas, el programa los habría marcado en una consulta anterior en todos los casos, entre 74 y 1,337 días antes del diagnóstico. Eso no prueba que el programa acorte las esperas: es un cálculo hecho hacia atrás.

Sólo un tercio de los marcados llegó a la prueba. Hubo problemas administrativos, pacientes que no se pudieron localizar y resultados guardados en papel que el programa no podía leer. Por eso las cifras de acierto son estimaciones, no cuentas finales, y una clínica no puede prometerle este nivel de detección. Tampoco se sabe si funcionaría en un consultorio de barrio o fuera de Polonia: los datos vienen de hospitales de un solo país, con pacientes mayores y más enfermos que los de una consulta común. El funcionamiento interno del programa es privado y no se publicó, así que nadie fuera de sus creadores puede revisarlo.

Los síntomas clásicos casi no aparecieron. Sólo el 2.25% de los marcados tenía sangre en la orina, mientras que la revisión recoge registros antiguos en los que esa cifra iba del 45% al 62%. Las señales que más pesaron fueron un tipo de destrucción de glóbulos rojos llamado hemólisis Coombs negativa y la simple frecuencia de las visitas.

Si alguien vuelve una y otra vez al hospital con anemia, cansancio o un problema de la sangre sin nombre, ese ir y venir es justo el patrón que este programa aprendió a notar. Vale la pena comentarlo con el hematólogo que lo atiende.

Qué significa para usted

Si usted vuelve una y otra vez al hospital por anemia, cansancio o un problema de la sangre que nadie ha nombrado, ese ir y venir es el patrón que este programa aprendió a notar en expedientes de hospitales de Polonia. Coméntelo con el hematólogo que lo atiende. Todavía nadie ha medido si esto adelanta el diagnóstico, así que vale preguntar si en su hospital se revisan esos antecedentes.

Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0

Quién pagó: El estudio recibió apoyo parcial mediante financiación de AstraZeneca a Saventic Health; varios autores son empleados o accionistas de AstraZeneca o de Saventic Health, y el artículo no indica que los financiadores tuvieran algún papel en el diseño, el análisis o la publicación.

No tome esto como consejo médico profesional.

Los hallazgos de otros estudios que aquí se mencionan provienen de este documento; esos estudios no se consultaron uno por uno.

Versión detalladaLos pasajes copiados del artículo, las ilustraciones y cada fuente con cuánto leímos de ella · 1460 palabras · unos 7 minLeerla →Cerrar

Un programa revisó 1.3 millones de expedientes y encontró 13 casos de una enfermedad rara que se diagnostica tarde

En 14 hospitales de Polonia, un algoritmo marcó a 356 pacientes como de alto riesgo; 119 se hicieron el estudio de confirmación y 13 salieron positivos.

La enfermedad se llama hemoglobinuria paroxística nocturna, o PNH. Es rara, adquirida y potencialmente mortal: la sangre se destruye, aparecen coágulos y la médula ósea falla. El artículo describe un estudio en el que menos del 40% de los pacientes recibe un diagnóstico dentro de los 12 meses siguientes al primer síntoma, y en 24% de los casos el diagnóstico tarda cinco años o más1. Un estudio publicado en julio de 2026 en *npj Digital Medicine* probó si un programa de inteligencia artificial puede acortar esa espera.

El programa se llama SARAH y lo desarrolló Saventic Health. Revisa expedientes electrónicos de hospital: códigos de diagnóstico, resultados de laboratorio y notas clínicas escritas en texto libre. Un paso de procesamiento de lenguaje natural saca de esas notas los síntomas que el médico escribió a mano2. Lo que se gana con eso es lo que un sistema de reglas fijas no ve: patrones sutiles en registros desordenados3. Lo que se pierde es la posibilidad de revisar el motor por dentro; la arquitectura detallada del algoritmo no puede divulgarse por restricciones de propiedad4.

Entre septiembre de 2023 y finales de noviembre de 2024, el programa revisó los expedientes de 1,307,140 pacientes en 14 organizaciones de salud de Polonia. Marcó como de alto riesgo a 356. De esos, los médicos mandaron a 119 a citometría de flujo, el estudio que confirma la enfermedad. Trece salieron positivos: una proporción de 10.92%5. Para encontrar un caso confirmado hubo que revisar los expedientes de 100,549 personas y hacer 9.15 estudios6. El artículo describe un estudio en el que esa proporción se reporta como favorable frente al 6.9% de aciertos de una revisión convencional anterior5.

Diez de los trece diagnosticados tenían historia clínica previa en el mismo hospital. Al aplicar el programa a sus visitas pasadas, apareció en todos al menos una consulta que ya habría bastado para marcarlos. La demora entre esa visita y el diagnóstico final fue de 74 a 1,337 días7. Es un cálculo hecho hacia atrás sobre el expediente, no una prueba de que el programa haya evitado esas demoras.

El hallazgo que más desconcierta es el de los síntomas. Solo 2.25% de los pacientes marcados tenía orina oscura, la señal que suele enseñarse en las aulas; el artículo describe un estudio en el que, en los registros internacionales de la enfermedad, esa proporción va de 45% a 62%8. Los marcadores que mejor separaron a los positivos de los negativos fueron la hemólisis con prueba de Coombs negativa y la frecuencia de visitas al hospital9. Lo curioso es que esos pacientes ya estaban siendo atendidos: iban y volvían, y nadie les pedía el estudio de PNH hasta que el programa los señaló10.

El estudio tiene límites que conviene decir completos. Solo 33% de los marcados se hizo el estudio de confirmación, y aunque los análisis no hallaron diferencias sistemáticas entre los que se lo hicieron y los que no, no puede descartarse que algo no medido influyera11. Las cifras de sensibilidad y especificidad que el artículo reporta son estimaciones, porque puede haber casos sin detectar entre los no marcados12. La población era de hospital, con una mediana de edad de 69.5 años, más vieja y más enferma que la de un consultorio. No se midió si el programa reduce muertes ni trombosis. Y el estudio se hizo únicamente en Polonia.

Piense en el expediente de un hospital: una carpeta que ya no es de papel, sino un sistema que guarda análisis, recetas, notas de guardia y fechas de consulta. Sobre eso trabajó el programa. La pregunta de quién ve esos datos y quién los cuida no es menor: los autores reportan que el estudio se financió en parte con dinero de AstraZeneca a Saventic Health13, y que varios firmantes son empleados o accionistas de AstraZeneca o de Saventic Health14. Eso no invalida el resultado; sí dice quién tiene interés en que el producto se vea bien.

Otros trabajos, de los que solo el resumen estaba disponible, van en la misma dirección. Un estudio probó un complemento de bajo costo para microscopios que detecta leucemia sin necesidad de escáneres de laminillas, caros y escasos en lugares con pocos recursos15, y reportó buena precisión sobre laminillas de médula ósea en varias instituciones16. Otro revisó programas que leen el electrocardiograma para detectar un corazón que bombea poco: en atención primaria aumentaron los diagnósticos nuevos17, y una revisión de ensayos controlados encontró más detecciones que con la atención habitual18. Ahí también hay un costo: la proporción de aciertos entre los señalados iba de 20% a 34%19, y no se demostró que eso cambie hospitalizaciones ni muertes20.

Pasa lo mismo con la promesa que con el trámite. Los programas que leen expedientes se presentan como imparciales porque no se cansan ni tienen antipatías. Pero se entrenan con los pacientes de ciertos hospitales, y aprenden mejor lo que se parece a ellos. En este estudio, una búsqueda posterior en los expedientes actualizados de los pacientes no marcados encontró cuatro casos de PNH, y los cuatro tenían algo en común: los síntomas les salieron de golpe y la historia clínica era muy corta. El programa no los vio porque no había mucho que ver. Eso no es un defecto del algoritmo; es el tipo de caso que un sistema así deja pasar.

Así lo lee weeklyAI. La escena que importa no es la del algoritmo marcando nombres en una pantalla, sino la de la persona señalada que nunca llega al estudio de confirmación. En este estudio, de los 356 marcados, 119 se hicieron la citometría y el resto no se la hizo durante el período del estudio. Solo 33% de los marcados se hizo el estudio de confirmación, y aunque los análisis no hallaron diferencias sistemáticas entre los que se lo hicieron y los que no, no puede descartarse que algo no medido influyera11. Una parte de los expedientes que el programa no podía leer estaba en papel, fuera del sistema. Esa es la brecha entre lo que un programa detecta y lo que un paciente recibe. Si a usted o a un familiar les avisan que están en un grupo de riesgo, pregunte desde el primer momento quién hace el estudio, dónde y cuánto cuesta, y pida el siguiente paso por escrito. Esta lectura se equivocaría si, en un servicio parecido, la mayoría de los señalados completara el estudio sin ayuda adicional.

Hay una segunda cosa que vigilar. El programa funciona, según sus desarrolladores, pero no se puede revisar por dentro. El artículo describe un estudio en el que ya se había reportado un programa parecido para PNH entrenado con una base de atención primaria del Reino Unido21, con una proporción de aciertos de 19.59% entre los clasificados como positivos22. Son dos herramientas distintas, en dos países distintos, con dos tipos de expediente. El artículo menciona la segunda como contexto, no como comparación directa. Lo que sí comparten es que ninguna de las dos se ha probado en un consultorio de barrio. Si un programa no lo marcó pero usted sigue con síntomas, dígaselo a su médico y pida que se estudie el caso por la vía normal. Pregunte si el sistema fue probado con gente como usted: misma edad, mismo idioma, misma región.

El estudio deja una pregunta abierta que los propios autores señalan: qué pasaría si el programa se aplicara en atención primaria, donde los pacientes son más jóvenes y los expedientes más delgados. No lo probaron. Tampoco probaron si detectar antes cambia el curso de la enfermedad. Eso requeriría comparar pacientes atendidos con y sin el programa y medir desenlaces: muertes, trombosis, calidad de vida. Esa investigación no existe todavía.

La pregunta que un lector puede llevarse no es si la inteligencia artificial funciona. Es si el aviso llega a manos de alguien que pueda hacer algo con él. En un hospital polaco, un programa revisó un millón trescientos mil expedientes y encontró trece personas que llevaban meses o años yendo y viniendo sin nombre para lo que tenían. Trece personas en un millón trescientos mil es poco, y a la vez es todo para esas trece. La próxima vez que en su clínica le digan que un programa revisa expedientes para buscar enfermedades, pregunte tres cosas: quién recibe el aviso, quién se asegura de que el paciente llegue al estudio, y qué pasa con los resultados que están en papel. ¿Qué haría usted si le avisan que está en un grupo de riesgo y el estudio de confirmación no está cubierto?

De dónde sale cada dato de contexto, y cuánto leímos de cada documento

  1. Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Fewer than 40% of patients with PNH receive a diagnosis within 12 months of symptom onset, and 24% of all PNH diagnoses can take even 5 years or longer."
  2. Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0 - el artículo del que trata esta nota — el artículo completo — el pasaje: "This natural language processing (NLP) step enabled the extraction of important patient features related to PNH symptoms from unstructured texts."
  3. Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The algorithm’s integration of both structured (ICD-10 codes, laboratory results) and unstructured data (free-text clinical notes) reflects the complexity of real medical records, enabling detection of subtle clinical patterns that might be missed by rule-based approaches."
  4. Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Finally, the detailed architecture of the screening algorithm cannot be fully disclosed, due to proprietary restrictions, limiting independent reproducibility of our findings and external replication without collaboration with the algorithm developers."
  5. Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Screening of 1,307,140 patients identified 356 high-risk individuals; of 119 referred for flow cytometry, 13 were diagnosed (positive predictive value: 10.92%; 95% CI, 9.68%–12.30%), comparing favourably to 6.9% conventional screening hit rate."
  6. Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The Number Needed to Screen (NNS) and the Number Needed to Test (NNT) to identify one confirmed case of PNH were 100,549 (1,307,140/13) and 9.15 (119/13), respectively."
  7. Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0 - el artículo del que trata esta nota — el artículo completo — el pasaje: "It was found that every patient experienced a delay in diagnosis - each had at least one visit in the past that would allow the algorithm to flag a high risk prior to the final positive visit (the visit that ultimately resulted in their high-risk classification). This diagnostic delay ranged from 74 to 1337 days (Fig. 3b )."
  8. Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Only 2.25% presented with haemoglobinuria versus 45–62% in registry cohorts."
  9. Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Monte Carlo feature selection identified Coombs-negative haemolysis and visit frequency as strongest predictors, supporting the potential utility of AI-assisted screening for identifying atypical PNH presentations."
  10. Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0 - el artículo del que trata esta nota — el artículo completo — el pasaje: "This indicates that Coombs negative haemolysis was the most frequent factor influencing the number of patient visits, yet it did not prompt physicians to test for a PNH clone (Fig. 5b ) demonstrating that these patients were actively managed within the care system but not referred for testing until AI screening enabled their identification."
  11. Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Only 33% of flagged patients underwent confirmatory FC testing, and while statistical analyses demonstrated no systematic differences between tested and non-tested subgroups, unmeasured confounding cannot be definitively excluded."
  12. Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The performance metrics should be treated as estimates and interpreted with the caveat that undetected PNH cases may exist among non-flagged patients who did not undergo confirmatory FC, which could result in an overestimation of specificity and NPV and an underestimation of sensitivity."
  13. Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0 - el artículo del que trata esta nota — el artículo completo — el pasaje: "This study was supported in part by funding from AstraZeneca to Saventic Health."
  14. Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0 - el artículo del que trata esta nota — el artículo completo — el pasaje: "R.D. is an employee and shareholder of AstraZeneca. P.D. is an employee of AstraZeneca. K.L., M.D., G.B. are the shareholders of Saventic Health."
  15. Yan ES, Sun S, Yin Z, Kamali A, Van Cleave J, Isgor IS, et al. (2026). An AI-powered self-driving microscope for low-cost acute leukemia detection. Nature Communications. 10.1038/s41467-026-77130-x — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Current artificial intelligence systems for leukemia detection typically rely on costly whole-slide scanners, limiting accessibility in low-resource settings."
  16. Yan ES, Sun S, Yin Z, Kamali A, Van Cleave J, Isgor IS, et al. (2026). An AI-powered self-driving microscope for low-cost acute leukemia detection. Nature Communications. 10.1038/s41467-026-77130-x — solo el resumen - no se pudo obtener el texto completo — el pasaje: "ALLocate achieves an area under the receiver operating characteristic curve greater than 0.99 for region-of-interest identification, a mean average precision at 50% intersection over union of 0.90 for cell detection, and 88% accuracy for slide-level diagnosis on glass slides without requiring a whole-slide scanner."
  17. Bollmann A, Pradler V, Husser D, Kim I, Hohenstein S, Nitsche A, et al. (2026). Artificial intelligence–enabled electrocardiography for assessment of left ventricular systolic dysfunction in the era of foundation models. Frontiers in Cardiovascular Medicine. 10.3389/fcvm.2026.1899224 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "In primary care, AI-ECG increased the number of new low-ejection-fraction diagnoses and directed echocardiography preferentially to screen-positive patients."
  18. Ahmed AAO, Ahmed OAO, Ahmed RAO, Dishean AS, Gilbert C, Mohamed MMA, et al. (2026). Artificial Intelligence‐Enabled Electrocardiogram Screening for Detection of Low Ejection Fraction in Routine Care: A Systematic Review and Meta‐Analysis of Randomized Controlled Trials. Annals of Noninvasive Electrocardiology. 10.1111/anec.70248 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "AI-ECG screening significantly improved detection of new low EF diagnoses compared to usual care (RR 1.36, 95% CI 1.12 to 1.64; p = 0.002; I 2 = 0%), with a number needed to screen of 204."
  19. Ahmed AAO, Ahmed OAO, Ahmed RAO, Dishean AS, Gilbert C, Mohamed MMA, et al. (2026). Artificial Intelligence‐Enabled Electrocardiogram Screening for Detection of Low Ejection Fraction in Routine Care: A Systematic Review and Meta‐Analysis of Randomized Controlled Trials. Annals of Noninvasive Electrocardiology. 10.1111/anec.70248 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "The PPV of the AI tool ranged from approximately 20% to 34% across studies."
  20. Bollmann A, Pradler V, Husser D, Kim I, Hohenstein S, Nitsche A, et al. (2026). Artificial intelligence–enabled electrocardiography for assessment of left ventricular systolic dysfunction in the era of foundation models. Frontiers in Cardiovascular Medicine. 10.3389/fcvm.2026.1899224 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "In population cohorts, adding AI-ECG signals to PREVENT-HF improves near-term heart-failure risk discrimination and reclassification, though without demonstrated benefit on clinical outcomes such as heart-failure hospitalization or mortality."
  21. Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The authors reported population adjusted positive predictive value, defined as percentage of the cases with existing/coded diagnosis of PNH out of those classified as positive for the disease by the model, of 19.59%. Such a tool could be potentially used in primary care setting to identify patients at high risk of PNH 28 ."
  22. Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Moreover, a recent investigation utilizing the UK Health England Records database revealed a PPV of 19.59% for PNH detection 28 . Notably, while these studies validate the efficacy of AI tools, they do not provide substantial data that illustrate their application within routine clinical settings, highlighting a gap in real-world utility evidence."

Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0

Quién pagó: El estudio recibió apoyo parcial mediante financiación de AstraZeneca a Saventic Health; varios autores son empleados o accionistas de AstraZeneca o de Saventic Health, y el artículo no indica que los financiadores tuvieran algún papel en el diseño, el análisis o la publicación.

No tome esto como consejo médico profesional.

Los hallazgos de otros estudios que aquí se mencionan provienen de este documento; esos estudios no se consultaron uno por uno.

other · NPJ digital medicine · the paper, 9 Jul 2026 · free

Software Flagged a Rare Blood Disease in 13 People. Most of Its Alarms Went Untested.

In 14 Polish hospitals, an AI tool read 1.3 million patient records and flagged 356 as high risk. Of the 119 who were then tested, 13 had the disease.

Short version · the longer version follows, about 6 min

Ask the weeklyAI chat

Ask me about AI in your life: I point you to the stories weeklyAI has published on it, this week and before.

Conversations are saved for as long as weeklyAI exists, to improve the publication. Answers come in the language you write in.

The study at a glance
Who
hospital patients whose records were screened for a rare blood disease
How many
1,307,140 patients
Where
14 hospitals in Poland
When
September 2023 to November 2024
Kind of study
analysis of what people did
Who did it
Saventic Health, Poland, with AstraZeneca funding
The limit that matters
Only 33% of flagged patients were tested, so accuracy figures are estimates.
How often the classic sign, dark urine, appeared in flagged patients compared with older registries
flagged patients in this study2.25%
older registry cohorts (low end)45%
older registry cohorts (high end)62%

Share of people with the rare blood disease who had dark urine; the registry range comes from older cohorts, not from this study.

Hit rate of AI screening versus conventional screening

AI screening in 14 Polish hospitalsagainstconventional screening in an earlier Polish study

10.92% of those tested had the disease, against 6.9% with conventional screening

The software does not treat anyone. It sorts a crowd and hands a short list to a human, and everything depends on what the human does next.
weeklyAI's reading

One in nine. That is how many of the tested patients turned out to have paroxysmal nocturnal haemoglobinuria, or PNH, a rare disease in which the body's blood cells break apart.

The numbers behind it: researchers at Saventic Health in Poland built software that reads hospital records, both the coded fields and the free-text notes, and looks for patterns that suggest PNH. They ran it across 14 healthcare organizations and 1,307,140 patients between September 2023 and November 2024. It flagged 356 people as high risk.

Doctors then decided whether to send those patients for flow cytometry, the standard blood test that confirms PNH. Of the 119 who were sent, 13 were diagnosed. That is about 1 in 9 tested. The article describes an earlier study in one Polish centre, where conventional screening found PNH in roughly 1 in 14 tested.

Look back at the records of 10 of those 13 patients. The software would have flagged every one of them at an earlier visit, between 74 and 1,337 days before the diagnosis finally came. The study calls these potentially preventable delays, estimated after the fact.

Only a third of the flagged patients were ever tested, though. The study could not find out why 71 of them were not tested. Of the rest, 114 were lost to follow-up because of administrative limits and missing contact details, 31 had another diagnosis confirmed, 14 had PNH ruled out before, 4 already had PNH confirmed, and 3 died. So the accuracy figures are estimates from one health system, not a guarantee of what a tool would find where you live.

The study ran in hospitals in Poland, on an older, sicker group than a typical clinic sees. The median age of flagged patients was 69.5 years. It does not show the software would work in a community clinic or in another country's records.

The clearest signal was not the classic one. Just 2.25% of flagged patients had blood in the urine. The article describes older registries where 45 to 62% of patients had it. The study found the closest links between Coombs-negative haemolysis and how often a person had been seen, and between haemolysis and how often a person had been seen.

One question is worth carrying to any clinic that offers a screening tool: can anyone outside the company check how it works? The detailed design of this algorithm is proprietary and not published.

If you or someone you know keeps returning with unexplained anaemia, fatigue or a blood disorder and no name for it, that repeated visiting is the pattern this software was built to notice. Raise it with the treating haematologist.

What this means for you

For now, this is a hospital tool in Poland, tested on a group older and sicker than most clinic patients, and the study did not measure whether anyone lived longer or avoided a clot. If you keep returning with unexplained anaemia, fatigue or a blood disorder that has no name, ask the haematologist whether PNH testing is warranted.

Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0

Who paid: The study was supported in part by funding from AstraZeneca to Saventic Health; several authors are employees or shareholders of AstraZeneca or Saventic Health, and the article does not state that the funders had any role in design, analysis or publication.

Do not take this as professional medical advice.

The findings of other studies mentioned here come from this document; those studies were not opened one by one.

The longer versionThe passages copied from the paper, the pictures, and every source with how much of it we read · 1158 words · about 6 minRead it →Close

An AI Flagged 356 People for a Rare Blood Disease. Thirteen Had It.

In 14 Polish hospitals, software read a million patient records and found a disease doctors had missed. The test that confirmed it reached only a third of those flagged.

The disease is paroxysmal nocturnal haemoglobinuria, PNH. It is rare, it can kill, and it is slow to be named: the article describes a study in which fewer than 40% of patients are diagnosed within a year of their first symptoms, and 24% wait five years or longer1. The classic sign in textbooks is dark urine. The article describes a study in which, among the 356 people this software flagged as high-risk, only 2.25% had it, against 45–62% in registry cohorts2.

The tool, built by a Polish company called Saventic Health and run inside its SARAH platform, reads hospital records the way a careful clerk would: lab results, diagnosis codes, and the free-text notes where doctors describe what they saw. It pulls symptoms out of ordinary prose, and it looks for combinations that might otherwise go unnoticed34. Across 14 hospitals in Poland, from September 2023 to the end of November 2024, it screened 1,307,140 patients and flagged 3565. Physicians reviewed the flags and sent 119 of those patients for flow cytometry, the standard confirmatory blood test. Thirteen had PNH5.

That is a hit rate of 10.92% among those tested, with a range of 9.68% to 12.30%5. The authors compare it with a 6.9% hit rate reported in an earlier Polish study of conventional screening5. Put in people: one confirmed case for every 9.15 patients sent to the test6. In the other direction, finding one case meant screening 100,549 records6.

Ten of the thirteen diagnosed patients had old records the team could re-read. Every one of them had at least one past visit that would have been enough for the software to raise its flag. The gap between that visit and the diagnosis ran from 74 to 1,337 days7. That is a retrospective estimate of what might have been caught earlier, not a measurement of time saved.

The strongest signals the analysis found were Coombs-negative haemolysis and how often a patient came back8. In plain terms: patients whose blood was breaking down for no identified immune reason, and who kept returning to the hospital, were being managed but not tested for PNH until the software pointed at them9.

The limits are as concrete as the result. Only 33% of flagged patients were ever tested, and while the team found no systematic differences between those tested and those not, unmeasured differences cannot be ruled out10. The accuracy figures are estimates: some unflagged patients were never checked, so the true numbers could be better or worse11. The software's internal design is proprietary and not published, so no outside group can rebuild it12. The patients the software flagged were older and had more other conditions, as is typical of hospital populations, which limits how far the results travel to outpatient clinics.

Consider what a hospital record actually contains. A lab value typed into a field. A discharge summary written at the end of a long shift. A referral that never happened. The Saventic tool reads both the tidy parts and the messy ones, which is why its authors argue it can catch patterns that simpler rule-based checks would miss3. The text-reading step is what turns a doctor's sentence into a symptom the software can count4.

This is not the first attempt at PNH screening by machine. The article describes an earlier effort that used a UK primary care database and reported a positive predictive value of 19.59%1314. That number should not be set beside the Polish 10.92% as a contest: different country, different records, different definition of what counts as a hit. What the Polish authors say the earlier work lacked was evidence from routine clinical use, and that is the gap this deployment tried to fill14.

The same idea is being tried in other diseases. According to the summary of a review of AI applied to heart-rhythm readings, the only part of it available, an FDA-cleared version of such software exists to help clinicians screen for a weak heart pump from a standard ECG15. According to the summary of a pooled analysis of three randomized trials, the only part of it available, that kind of screening raised the number of new low-pump-function diagnoses compared with usual care, with 204 people needing screening for one extra diagnosis16. According to the summary of a study of a low-cost microscope add-on, the only part of it available, a plugin for ordinary microscopes reached 88% accuracy in diagnosing leukemia slides without an expensive whole-slide scanner17.

A pattern runs through these. The software does not treat anyone. It sorts a crowd and hands a short list to a human, and everything depends on what the human does next. In the Polish hospitals, two-thirds of the flagged patients were never sent for the confirming test. A follow-up analysis found that administrative barriers, including lost contact with patients and records kept only on paper, were the main reasons those patients were not tested.

Here is weeklyAI's reading. A disease that hides behind tiredness and low blood counts will keep being missed by routines built around the visible sign. Expect the same shape wherever this kind of tool is tried: a small number of true cases found among a much larger number of worried people, and the outcome decided not by the algorithm but by whether the confirming test actually gets scheduled. This reading would be wrong if hospitals already test for PNH at similar rates without software, or if the flagged patients turned out to be mostly the ones with the obvious symptom. For a reader, the practical move is simple: if you or someone in your family has unexplained tiredness, low blood counts, or repeated transfusions with no clear cause, ask the doctor directly whether a rare blood condition has been ruled out, and ask which test would settle it. Silence is not the same as an answer.

One more thing weeklyAI would flag. When a tool judges people, the people are rarely told they were judged. Ask your clinic whether any automated system reviews your records, and if a test is recommended, ask for the result in writing, so it does not sit unseen in a file.

The authors are explicit about what comes next: refining the tool, testing whether earlier detection actually changes outcomes such as survival or blood clots, and adapting it for clinics outside hospitals. None of that has been done here.

The bill, the waiting room, the folder of old results: the software read all of it and found thirteen names. What it cannot do is walk anyone to the lab. If a screening tool ever flags you or someone you love, the question worth carrying into the appointment is this: which test confirms it, and when is it booked?

Where each piece of context comes from, and how much of it we read

  1. Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0 - the article this story is about — the whole article — the passage: "Fewer than 40% of patients with PNH receive a diagnosis within 12 months of symptom onset, and 24% of all PNH diagnoses can take even 5 years or longer."
  2. Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0 - the article this story is about — the whole article — the passage: "Only 2.25% presented with haemoglobinuria versus 45–62% in registry cohorts."
  3. Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0 - the article this story is about — the whole article — the passage: "The algorithm’s integration of both structured (ICD-10 codes, laboratory results) and unstructured data (free-text clinical notes) reflects the complexity of real medical records, enabling detection of subtle clinical patterns that might be missed by rule-based approaches."
  4. Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0 - the article this story is about — the whole article — the passage: "This natural language processing (NLP) step enabled the extraction of important patient features related to PNH symptoms from unstructured texts."
  5. Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0 - the article this story is about — the whole article — the passage: "Screening of 1,307,140 patients identified 356 high-risk individuals; of 119 referred for flow cytometry, 13 were diagnosed (positive predictive value: 10.92%; 95% CI, 9.68%–12.30%), comparing favourably to 6.9% conventional screening hit rate."
  6. Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0 - the article this story is about — the whole article — the passage: "The Number Needed to Screen (NNS) and the Number Needed to Test (NNT) to identify one confirmed case of PNH were 100,549 (1,307,140/13) and 9.15 (119/13), respectively."
  7. Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0 - the article this story is about — the whole article — the passage: "It was found that every patient experienced a delay in diagnosis - each had at least one visit in the past that would allow the algorithm to flag a high risk prior to the final positive visit (the visit that ultimately resulted in their high-risk classification). This diagnostic delay ranged from 74 to 1337 days (Fig. 3b )."
  8. Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0 - the article this story is about — the whole article — the passage: "Monte Carlo feature selection identified Coombs-negative haemolysis and visit frequency as strongest predictors, supporting the potential utility of AI-assisted screening for identifying atypical PNH presentations."
  9. Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0 - the article this story is about — the whole article — the passage: "This indicates that Coombs negative haemolysis was the most frequent factor influencing the number of patient visits, yet it did not prompt physicians to test for a PNH clone (Fig. 5b ) demonstrating that these patients were actively managed within the care system but not referred for testing until AI screening enabled their identification."
  10. Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0 - the article this story is about — the whole article — the passage: "Only 33% of flagged patients underwent confirmatory FC testing, and while statistical analyses demonstrated no systematic differences between tested and non-tested subgroups, unmeasured confounding cannot be definitively excluded."
  11. Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0 - the article this story is about — the whole article — the passage: "The performance metrics should be treated as estimates and interpreted with the caveat that undetected PNH cases may exist among non-flagged patients who did not undergo confirmatory FC, which could result in an overestimation of specificity and NPV and an underestimation of sensitivity."
  12. Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0 - the article this story is about — the whole article — the passage: "Finally, the detailed architecture of the screening algorithm cannot be fully disclosed, due to proprietary restrictions, limiting independent reproducibility of our findings and external replication without collaboration with the algorithm developers."
  13. Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0 - the article this story is about — the whole article — the passage: "The authors reported population adjusted positive predictive value, defined as percentage of the cases with existing/coded diagnosis of PNH out of those classified as positive for the disease by the model, of 19.59%. Such a tool could be potentially used in primary care setting to identify patients at high risk of PNH 28 ."
  14. Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0 - the article this story is about — the whole article — the passage: "Moreover, a recent investigation utilizing the UK Health England Records database revealed a PPV of 19.59% for PNH detection 28 . Notably, while these studies validate the efficacy of AI tools, they do not provide substantial data that illustrate their application within routine clinical settings, highlighting a gap in real-world utility evidence."
  15. Bollmann A, Pradler V, Husser D, Kim I, Hohenstein S, Nitsche A, et al. (2026). Artificial intelligence–enabled electrocardiography for assessment of left ventricular systolic dysfunction in the era of foundation models. Frontiers in Cardiovascular Medicine. 10.3389/fcvm.2026.1899224 — only the abstract - the full text could not be fetched — the passage: "FDA-cleared software is available for left ventricular ejection fraction ≤40% screening from 12-lead ECGs as clinician decision support."
  16. Ahmed AAO, Ahmed OAO, Ahmed RAO, Dishean AS, Gilbert C, Mohamed MMA, et al. (2026). Artificial Intelligence‐Enabled Electrocardiogram Screening for Detection of Low Ejection Fraction in Routine Care: A Systematic Review and Meta‐Analysis of Randomized Controlled Trials. Annals of Noninvasive Electrocardiology. 10.1111/anec.70248 — only the abstract - the full text could not be fetched — the passage: "AI-ECG screening significantly improved detection of new low EF diagnoses compared to usual care (RR 1.36, 95% CI 1.12 to 1.64; p = 0.002; I 2 = 0%), with a number needed to screen of 204."
  17. Yan ES, Sun S, Yin Z, Kamali A, Van Cleave J, Isgor IS, et al. (2026). An AI-powered self-driving microscope for low-cost acute leukemia detection. Nature Communications. 10.1038/s41467-026-77130-x — only the abstract - the full text could not be fetched — the passage: "ALLocate achieves an area under the receiver operating characteristic curve greater than 0.99 for region-of-interest identification, a mean average precision at 50% intersection over union of 0.90 for cell detection, and 88% accuracy for slide-level diagnosis on glass slides without requiring a whole-slide scanner."

Dewor, R., Dabrowski, M. J., Więcek, Ł. et al. (2026). AI-driven diagnostic algorithm enhances early detection of paroxysmal nocturnal hemoglobinuria in real-world settings. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02980-0

Who paid: The study was supported in part by funding from AstraZeneca to Saventic Health; several authors are employees or shareholders of AstraZeneca or Saventic Health, and the article does not state that the funders had any role in design, analysis or publication.

Do not take this as professional medical advice.

The findings of other studies mentioned here come from this document; those studies were not opened one by one.

The free weekly drip by email

One email a week: the free story of every section, with its picture, and the headlines of the rest.

No payment, nothing to cancel. Unsubscribe from any issue.

El goteo semanal gratis por correo

Un correo por semana: la nota gratuita de cada sección, con su imagen, y los titulares de las demás.

Sin pago y sin nada que cancelar. Puede darse de baja desde cualquier envío.

From the house

About

Every week the new research on AI is read here, and this page tells you what was found, how far it can be trusted, and what you can do with it. This page is free, and every line on it says what was found.

The studies are read by a machine built and ruled by a professor of communication at Sacramento State and Universidad Nacional Autónoma de México (UNAM). Who makes this →

Diego Bonilla, Ph.D.

Full Professor of Communication Studies, Sacramento State, California, United States
Graduate instructor, Universidad Nacional Autónoma de México (UNAM), Mexico

hypergraphia.com

I am a professor of communication at Sacramento State, where I created the course Information Management and Privacy and have taught it many times, and at UNAM, in Mexico City, I teach the graduate course Electronic and Generative Literature with AI. In both I teach AI. I have taught since 1989, at first mathematics and physics to children who found them hard, and since 1997 one subject that has changed its name many times: media literacy, computer literacy, digital literacy, information literacy, ICT literacy, and now AI literacy.

This page exists because most people cannot go back to a university, and these systems are changing their work, their children's schooling, their parents' care and their vote all the same. What a university teaches about them should reach the people who need it, every week, in plain language, without enrolling anywhere. That is what I do here, and it is what I do on Thursday mornings in a seminar for the Renaissance Society at Sacramento State, built on this publication, where anyone can send a question and hear it answered.

In 2003 my doctorate at Syracuse measured what a machine can infer about a person from the traces they leave, and what I found became the privacy course I still teach. In 2007 and 2009 I sat on the committee at ETS in Princeton that set the national standard for digital literacy, and from 2014 to 2016 I was the California State University's ICT Literacy Ambassador. Six names, one job: helping people not be at the mercy of the newest machine.

I have also worked on the other side of the counter. In 2000 I began making films that a machine assembles differently for every viewer; the first of them took first prize at the Moscow International Film Festival in 2004, two decades before the industry shipped a generative model. A generative video poem I made with a poet at UNAM is in the field's definitive anthology, and in 2025 my AI short film The Human Futures Market took Best AI Short Film at the Red Rocks AI Film Festival. My recent published research is on AI and elections, and on how these models represent the cultures they were not trained on. I do not read the papers myself, and this page will never tell you I do. I built the machine that reads them and wrote the rules it reads by, and I analyze what it gives me the way I have taught my students to analyze: asking how it knows, and who benefits from what it says.

The vendors began teaching AI literacy two years ago, because before that there was nothing to sell. I am not against them. I am simply not one of them, and no one who sells these systems pays for this page. Your subscription does, and that is what keeps it honest.

The record, dated
  • 1989 Teaching begins, Mexico City
  • 1995 Hypergraphia, an early Spanish-language web and hypermedia publisher
  • 1996 Mercadotecnia e imagen en Internet, with Jesús del Toro, a book on marketing on the Internet, among the first in Spanish, Editorial Iberoamérica
  • 1999 Autorretrato, first generative poem, published by UNAM's Instituto de Investigaciones Filológicas
  • 2003 Ph.D., Syracuse University, Doctoral Prize, for the dissertation The Medium Is the Measure of Itself
  • 2004 A Space of Time: First Prize, Internet/Multimedia, XXVI Moscow International Film Festival
  • 2007, 2009 ETS iSkills committee, Princeton
  • 2008 Making Sense of Tracking Data, a book on how the traces a person leaves online allow inferences about them, including how they think, VDM Verlag
  • 2011 Keynote on generative work, UNAM
  • 2014–2016 ICT Literacy Ambassador, California State University · California Open Educational Resources Council
  • 2019 Big Data, with Rodolfo Mata: Electronic Literature Collection, Volume 4
  • 2020 Tablada Hipertextual, with Rodolfo Mata, UNAM: the computer reads a poet's words and builds a different reading from them, the ground the language models now work on
  • 2021 Uku Pacha, a generative VR film: world premiere, Vancouver International Film Festival
  • 2023 Keynote on the representation of Latin American cultures in AI, Ibero-American Meeting of Audiovisual Narratives, Trujillo, Peru · Graduate teaching at UNAM begins
  • 2023–2026 Directed the faculty learning community Advancing AI Literacy, Sacramento State · Principles for Cognitive and Pedagogical Sovereignty
  • 2024 A weekly AI publication run on real pipelines, every week since
  • 2025 Best AI Short Film, Red Rocks AI Film Festival
  • 2026 Keynote, with Nancy Huante-Tzintzun, Latin American and Caribbean Digital Humanities symposium, UT San Antonio · AI Watch, a weekly seminar for the Renaissance Society at Sacramento State

The research record →

TertulIA: weeklyAI.watch/tertulia/en/Pictures are illustrations generated by weeklyAI.watch, never photographs