TruaceTracing the truth around AITuesday, September 15, 2026
The Index

What the evidence says.What the public feels.

Ranks distinct AI gain and problem claims from the published record. Scores reward impact, independent source strength, scale, confidence, and recency.

1,419 results
Show filters and sorting

AI gains · 788

68
GainClimate· Stable· Evidence: Moderate (1 source)

A probabilistic causal machine learning framework trained on long-term monitoring data from a full-scale wastewater plant can predict N2O hot moments and convert interpretable outputs into risk-decision rules for adaptive aeration and load regulation to reduce emissions.

Researchers developed a probabilistic causal machine learning framework using long-term online monitoring data from a full-scale biological wastewater treatment plant to address intermittent nitrous oxide emission hot moments. The approach combined predictive modeling with cohort-based SHAP for nonlinear effects, LiNGAM-based causal discovery for pathway identification, and copula-based joint probability analysis for risk quantification.

Impact 30%49
Evidence 25%95
Scale 20%35
Confidence 15%87
Recency 10%94

Updated Aug 17, 2026 · TRV-2026-0802

68
GainHealth· Stable· Evidence: Moderate (1 source)

AI provides powerful computational methods to extract diagnostic, biological and epidemiological insights from high-volume microbiology datasets, with potential to enhance diagnostics, antimicrobial stewardship and infection prevention when integrated into lab workflows.

Published August 15, 2026, this narrative review in Clinical Microbiology and Infection introduces the machine learning lifecycle from a clinical microbiology perspective, covering data preparation, model development, evaluation, and deployment, drawing on applied research and AI development guidelines for healthcare.

Impact 30%49
Evidence 25%95
Scale 20%35
Confidence 15%87
Recency 10%94

Updated Aug 17, 2026 · TRV-2026-0800

68
GainHealth· Stable· Evidence: Moderate (1 source)

A multi-modal deep learning framework using Heckmatt scores from six key muscles improved speed and diagnostic performance for muscle ultrasound, predicting neuromuscular pathology with an area under the precision-recall curve of 0.87 on a test set of 320 patients.

Researchers developed a single-center multi-modal deep learning framework that fuses muscle ultrasound Heckmatt scores from six key muscles with patient BMI and age to screen for neuromuscular pathology. Tested on 320 patients, the model achieved an area under the precision-recall curve of 0.87 for distinguishing presence versus absence of disease.

Impact 30%49
Evidence 25%95
Scale 20%35
Confidence 15%87
Recency 10%94

Updated Aug 17, 2026 · TRV-2026-0799

68
GainHealth· Stable· Evidence: Moderate (1 source)

Converting numerical clinical data into 24-bit rectangular coded images and training ResNet architectures improved heart failure survival prediction, reaching reported accuracy up to 0.9617.

On August 15, 2026, a peer-reviewed paper described a method that converts numerical heart failure data into 24-bit rectangular coded images to fit deep learning input sizes, then augments the dataset through horizontal augmentation and rotation in multiples of 15b0. The resulting images were used to train ResNet18 and ResNet50 models for survival prediction.

Impact 30%49
Evidence 25%95
Scale 20%35
Confidence 15%87
Recency 10%94

Updated Aug 17, 2026 · TRV-2026-0797

AI problems · 631

67
ProblemHealth· Stable· Evidence: Moderate (1 source)

AI integration in drug discovery still limited by data quality and bias, lack of transparency and interpretability, high computational demands, and privacy and fairness risks.

Published December 12, 2025 as a peer-reviewed review, the article synthesizes how AI and bioinformatics are being applied across pharmaceutical R&D, from target identification to clinical use. It highlights advances in deep learning, graph networks, transformers, foundation models, and tools like AlphaFold, RFdiffusion, and AlphaFold3, reporting observed capabilities such as large-scale structure prediction and workflow compression from five years to 12-18 months.

Impact 30%49
Evidence 25%95
Scale 20%35
Confidence 15%87
Recency 10%89

Updated Jul 20, 2026 · TRV-2026-0414

67
ProblemPolicy· Stable· Evidence: Moderate (1 source)

Use of the AI system for traffic violation appeals impacts decisions made by legal experts and creates tensions between street-level bureaucrats, screen-level bureaucrats and street-level algorithms.

Researchers conducted action research on the development and deployment of an AI system to process traffic violation appeals at a Dutch court, using interviews, observations, documents and a user-experiment to compare decisions made by, with and without the system.

Impact 30%49
Evidence 25%95
Scale 20%35
Confidence 15%87
Recency 10%89

Updated Jul 20, 2026 · TRV-2026-0413

67
ProblemHealth· Stable· Evidence: Moderate (1 source)

ChatGPT responses contained incorrect information in more than one instance and were written at a college graduate reading level, requiring caution for patient education.

In a study published June 22, 2024, two hip preservation surgeons graded ChatGPT 3.5 answers to ten common hip arthroscopy questions drawn from patient education sites, using an A-to-D scale and readability scores FRES and FKGL.

Impact 30%49
Evidence 25%95
Scale 20%35
Confidence 15%87
Recency 10%89

Updated Jul 20, 2026 · TRV-2026-0412

67
ProblemMedia & Arts· Stable· Evidence: Moderate (1 source)

Applying conventional Western-rooted AI tools to Mijikenda musical heritage risks perpetuating digital colonialism through decontextualisation of sacred practices, infringement of data sovereignty, and cultural appropriation.

This peer-reviewed paper examines ethical implications of applying artificial intelligence to Indigenous musical heritage of the Mijikenda communities on Kenya's coast. It finds a divergence between AI's extractive logic and holistic Indigenous Knowledge Systems, with risks of decontextualising sacred practices and infringing data sovereignty.

Impact 30%49
Evidence 25%95
Scale 20%35
Confidence 15%87
Recency 10%89

Updated Jul 20, 2026 · TRV-2026-0409

Recomputed live from the record · Sep 15, 2026, 4:12 PM