Dr. Gal Goshen | Healthcare AI

A short record of the education work, 2025 to 2026, based on my LinkedIn posts and the Ministry's own pages. I come from a family of educators, and I have come to think the fastest way to prepare a country for AI is to start with the people who teach. Three things I have done with the Israeli Ministry of Education so far: HEY HAY AI, February 2025. I joined as a mentor in a national initiative, run by the Ministry together with industry, to train Israeli teachers in AI. Webinar for technology teams, 11 February 2025. An inspiration lecture titled "On health, innovation and artificial intelligence", recorded and published on the Ministry's AI in education pages . Science in the Classroom, from early 2026. I lecture in this program, connecting high-school students to medtech and AI and to what a career at that edge looks like. My view, after a few years of teaching AI to clinicians as well: the barrier is not the technology, it is the human capacity to lead it...

Podcast, December 2025. The conversation is in Hebrew; this summary is in English and Hebrew. I joined Odeya Rotem and Tal Shabtai on Tel Aviv 360, the podcast of the Gray Faculty of Medical and Health Sciences at Tel Aviv University, to talk about algorithms that are already entering treatment rooms, medical decisions that start to rest on AI, and the sensitive point where technology, ethics and human responsibility meet. At the time I was Chief AI Officer at ARC Innovation, Sheba Medical Center. Some of the questions we worked through: What will the hospital of 2030 look like, and which decisions stay with the physician? Why the future hospital has to be built with AI in its physical and operational fabric, and not added on top of the old one. How to move from the "garage" model, where we fix the body only when it breaks, to predictive, personalized prevention. How AI can take over the administrative load that drives burnout, so that clinicians can return to the pati...

Adapted from a LinkedIn post, spring 2026. The carousel with the key numbers is on LinkedIn. In the largest real-world study of AI scribes so far (five health systems, two years, 8,581 clinicians, published in JAMA in April 2026), adopters cut documentation time by an average of 16 minutes per eight scheduled patient hours. Only about a third used the tool in at least half of their visits. They cut 27 minutes, and Mass General Brigham , which led the study, reports a documentation-time reduction three times larger for this group. Same tool, same hospitals, different people. Three reports landed in the same week: McKinsey's AI transformation manifesto, BCG's outlook on healthcare AI, and the EU Joint Research Centre report on cardiovascular AI. They point the same way. Organizations that redesigned how work is done are seeing returns; most health systems have little beyond scattered pilots; and without targeted effort, AI risks widening the gap between hospitals. In my exper...

From a LinkedIn post, September 2026, after the closing keynote at the 27th Annual Meeting of the Israel Spine Society. I gave the closing keynote at the Israel Spine Society's 27th Annual Meeting. Three days covered the whole field; AI was one session in it, and mine was the last slot. Robotic systems place pedicle screws with roughly 97–99% accuracy in a meta-analysis , yet only a minority of screws in North America go in that way. The system navigates; the surgeon places the screw. The gap was never capability. What stayed with me was the room. Deskilling came up before I raised it, and so did the question of who owns the call when the model and the surgeon disagree. Neither is a technology question. The part that will change most people's week is not in the operating room at all: notes that write themselves out of the conversation you already had, explanations patients actually read, and the current literature available in the room when the question comes up. Thanks to...

From a LinkedIn post, late October 2026. Hebrew original below the English summary. I spoke on a panel about ethics and innovation in deploying AI tools in medical care, at a study day run by the Legal Bureau of the Ministry of Health at Shamir Medical Center. The audience was legal counsel, not engineers, and that changed the questions. My argument was that a tool's accuracy is no longer the hard part. The hard part is the day after the pilot. "Human in the loop" is necessary but not sufficient. Someone who approves a hundred outputs without finding an error stops checking, and stays in the loop as a rubber stamp. A physician who does not understand where a tool is likely to be wrong cannot be the human in the loop. That is why literacy and training come before the tool: for the physicians, for the users, and for the management that approves the purchase. You cannot govern a tool you do not understand. Dr. Oren Asman raised "shadow use", and Dr. Yossi Wal...

From a LinkedIn post, September 2026, on joining the Dina Recanati School of Medicine at Reichman University as Chief AI Officer. Hebrew follows the English. The medical students who start this year will care for people, sick and well, until around 2070. You can't teach the medicine of 2070. It doesn't exist yet. What you can do is stop training them only for the hospital of the 2020s. The hospital is where you arrive when it's already too late. Much of these students' future medicine will happen before that: in the signals the body sends all day, in models that detect disease long before symptoms appear, in a conversation at home instead of a trip to the emergency room. And much of the work won't be about treating the sick. It will be about keeping healthy people healthy. The physician's first question may no longer be simply, "What's wrong?" It may become: "What is likely to happen to you, and how do we keep you healthy?" And in ...