Stuff AI CAN'T Do

¿Puede la IA diagnosticar ciertas enfermedades raras a partir de registros médicos electrónicos ?

¿Qué opinas?

Los modelos de diagnóstico acompañante en 2024 encontraron casos de condiciones raras que pasaron desapercibidas para los clínicos tanto en los datos de entrenamiento como en ensayos en vivo.

Background

Over the past few years several groups have built transformer-based models that read longitudinal EHR sequences and flag patients whose symptom trajectories match curated rare-disease cohorts. In 2023 a system trained on more than 30,000 US patient records achieved a positive predictive value above 0.7 for four lysosomal storage disorders but fell below 0.5 for a rarer glycogenosis subtype, illustrating uneven performance across disorders. A multi-centre study published the same year compared two proprietary LLMs fine-tuned on anonymised records from specialist clinics and found they recovered 79 % of previously missed cases of Niemann-Pick type C while introducing one false positive per ten true positives. Workflows that combine structured billing codes with unstructured clinician notes have shown the biggest gains, yet they remain brittle when applied to centres whose documentation styles diverge from the training corpora. At least one large health-system rollout was paused after an audit revealed clinically significant drift when ICD-10 codes were updated, underscoring the maintenance burden of keeping rare-disease models current.

SOURCE: BMJ, 2024

Estado verificado por última vez en June 26, 2026.

📰

Galería

In the Court of AI Capability
Summary of Findings
Verdict over time
May 2026May 2026May 2026May 2026May 2026May 2026Jun 2026Jun 2026Jun 2026Jun 2026Jun 2026
Sitting at the Bench Filed · jun. 26, 2026
— The Question Before the Court —

¿Puede la IA diagnosticar ciertas enfermedades raras a partir de registros médicos electrónicos?

★ The Court Finds ★
Reaffirmed
Casi

Existen demostraciones limitadas — pero el panel no fue unánime.

Ruling of the Bench

The jury found the AI capable of glimpsing the shadow of rare disease across a patient record, yet unable to name the shape with full certainty; it delivers timely clues but not unshakable diagnoses. Their lone “almost” vote reflected cautious praise for pilot studies that edge past paperwork while still lacking robust, cross-hospital validation. Ruling: A compass that points northward but may wobble in a crosswind.

— Hon. E. Dijkstra-Patel, Presiding
Jury Tally
0
1Casi
0No
Verdict Confidence
80%
The Court of AI Capability is, of course, not a real court.
But the data is real.
The Case File · Stacked History
Session I · May 2026 In_research
Session II · May 2026 In_research
Session III · May 2026 Casi · 80%
Session IV · May 2026 Casi · 79%
Session V · May 2026 Casi · 80%
Session VI · May 2026 Casi · 75%
Session VII · Jun 2026 Casi · 76%
Session VIII · Jun 2026 Casi · 78%
Session IX · Jun 2026 Casi · 78%
Session X · Jun 2026 Casi · 83%
Case № 4110 · Session XI
In the Court of AI Capability

The Case File

Docket № 4110 · Session XI · Vol. XI
I. Particulars of the Case
Question put to the court¿Puede la IA diagnosticar ciertas enfermedades raras a partir de registros médicos electrónicos?
SessionXI (11 hearing)
Convened26 jun. 2026
Previously ruledIN_RESEARCH (May '26) → IN_RESEARCH (May '26) → ALMOST (May '26) → ALMOST (May '26) → ALMOST (May '26) → ALMOST (May '26) → ALMOST (Jun '26) → ALMOST (Jun '26) → ALMOST (Jun '26) → ALMOST (Jun '26) → ALMOST (Jun '26)
Presiding JudgeHon. E. Dijkstra-Patel
II. Cumulative Tally Across Sessions

Across 11 sessions, 36 jurors have heard this case. Combined tally: 6 YES · 27 ALMOST · 3 NO · 0 IN RESEARCH.

Note: cumulative includes older juror opinions. The current session tally above is the live verdict.

III. Verdict

By a vote of 0 — 1 — 0, the panel returns a verdict of CASI, with verdict confidence of 80%. The court so orders.

IV. Declaraciones del tribunal
Jurado I ALMOST

"Specialized AI models achieve partial rare disease diagnosis accuracy in narrow clinical cohorts"

Las declaraciones individuales de los jurados se muestran en su inglés original para preservar la precisión probatoria.

E. Dijkstra-Patel
Presiding Judge
M. Lovelace
Clerk of the Court

Lo que el público piensa

No 6% · Sí 91% · Quizás 3% 236 votes
Sí · 91%
La tendencia necesita votos de al menos 2 días distintos.

Discusión

no comments

Los comentarios e imágenes pasan por una revisión administrativa antes de aparecer públicamente.

11 jury checks · más reciente hace 2 días
26 Jun 2026 1 juror · indeciso indeciso
21 Jun 2026 2 jurors · indeciso, indeciso indeciso
15 Jun 2026 3 jurors · puede, indeciso, indeciso indeciso
10 Jun 2026 3 jurors · indeciso, puede, indeciso indeciso
04 Jun 2026 4 jurors · indeciso, indeciso, indeciso, indeciso indeciso
30 May 2026 3 jurors · indeciso, indeciso, indeciso indeciso
25 May 2026 4 jurors · indeciso, indeciso, indeciso, indeciso indeciso
19 May 2026 5 jurors · indeciso, indeciso, indeciso, indeciso, indeciso indeciso
15 May 2026 5 jurors · indeciso, indeciso, puede, indeciso, indeciso indeciso
12 May 2026 3 jurors · puede, no puede, puede indeciso
11 May 2026 3 jurors · puede, no puede, no puede indeciso estado cambiado

Cada fila es una comprobación de jurado independiente. Los jurados son modelos de IA (identidades mantenidas neutras a propósito). El estado refleja el recuento acumulado en todas las comprobaciones — cómo funciona el jurado.

Más en Judgment

¿Nos faltó uno?

Revisamos semanalmente.