AI Reduces Medical Errors, But Implementation Challenges Remain
Translated & summarized from Calcalist by baba
He claimed AI makes fewer medical errors than humans, especially under stress, citing internal tests. While AI can improve documentation and patient interaction, Kreiss stressed that organizational culture and training are the biggest implementation hurdles, not the technology itself.
The story in 4 lines · by baba
- Kreiss stated that AI makes fewer medical errors than humans, particularly during busy or tiring shifts.
- AI implementation faces significant challenges in organizational culture, regulation, and training, not just technology.
- Sheba is implementing AI to automate documentation, potentially saving doctors 50% of their work time.
- Kreiss believes AI investment can have a greater impact than physical infrastructure or new equipment.
He asserted that artificial intelligence must be a central component in transforming healthcare delivery to address these existential challenges.
Kreiss highlighted AI's potential to reduce medical errors, citing internal experiments where AI made fewer mistakes than humans, particularly under conditions of fatigue and high patient load, such as in an emergency room late at night. However, the article notes that details regarding these experiments, including the number of cases, comparison groups, and error measurement methods, were not provided, thus precluding a definitive conclusion about AI's universal superiority over physicians.
He also emphasized the need for caution when implementing AI, stating that while human errors are investigated and learned from, AI errors present a more complex situation. Kreiss pointed to AI's utility in documentation and administrative tasks as areas where it can free up physician time for patient care. Sheba is implementing a system that records doctor-patient conversations, processes the information, and directly inputs it into medical records, alongside a decision-support system.
This system is projected to save physicians 50% of their work time starting January 1, though the article notes the calculation methodology for this saving was not clarified. Kreiss stated that initial implementations have already improved patient interaction, allowing doctors to focus on care rather than typing and information retrieval. He stressed that technology itself is only 10% of the challenge, with infrastructure at 20%, and the remaining 70% lying in organizational culture, regulation, training, and ethics.
The primary difficulty, according to Kreiss, is not technological development but transforming technological capability into actual changes in teamwork. This involves training hospital staff to use and adapt AI tools for specific departmental needs, recognizing that a neonatal department's requirements differ significantly from a cardiology department's. Sheba's strategy includes acquiring existing solutions rather than developing everything in-house. Kreiss acknowledged the expense and complexity of this undertaking but suggested that investing in AI implementation could yield greater impact than adding a new building or imaging device.
Not the same event — other stories that share this one’s people, places, or theme: background, reactions, and follow-ups.
