A brain-bleed detector that sat silent in roughly 2,000 hospitals since 2018 just started writing the report itself, and I want to know who is checking its work. Two hundred miles away, an NHS ward began flagging which patients would get infected before any of them felt ill. Follow the thread and watch foundation models step out of the chat window and into the room.
The silent reader learned to write
Aidoc’s brain-bleed algorithm has watched scans without saying a word across roughly 2,000 hospitals since 2018. Its new CARE foundation model just started drafting the entire radiology report, and a $150M Series E says the market already believes it. Read what it drafts now →
Kent’s wards started predicting the future
Reading one scan is one thing; reading a whole ward before anyone spikes a fever is another. East Kent became the first NHS trust to trial Sanome’s MEMORI, which scores infection risk before symptoms surface, and its founder says 85 percent of clinicians now open it every day. See who trusts it →
The reasoning got up and walked
That kind of judgment used to live behind glass. Since April, every Spot on Boston Dynamics’ Orbit tier runs Google’s Gemini Robotics-ER 1.6, reading industrial gauges it was never trained on at 93 percent accuracy and then explaining, out loud, how it got there. Watch it reason →
Now you can own a piece of it
A machine that walks and reasons is also a machine with a balance sheet. China just cleared Unitree for a $619M Shanghai listing, putting a profitable, pure-play humanoid maker on the public market before a single US rival has managed it. Read the timing →
Second Opinion
Here is the part that keeps me up. A radiologist editing a machine’s draft does not read a scan the way one starting from a blank page does; the draft anchors you, and automation bias quietly rewards agreement. The liability still lands on the human signature at the bottom, so the question that matters is not how good the draft is, it is who audits the drafts, and how often.
One number
93 percent. That is how often a Spot robot correctly reads a gauge nobody trained it on, which is this whole issue in a single figure: judgment that used to sit on a screen now stands in a boiler room and reports back. See how it gets there →
One thing to use
AI medical scribes, explained. Before you trust any of this in a clinic, it helps to understand the version already sitting in exam rooms, drafting the visit note while your doctor talks. I wrote the physician’s guide to what it does well and where it quietly fails. Read the guide →
From a scan to a ward to a boiler room to a stock ticker, the through-line this week is simple: AI stopped waiting in a chat window and started doing the work. Hit reply and tell me which story deserves the full treatment next; I read everything. Forwarded this? Claim your own copy.
Dr. Joseph Joshua
P.S. Next issue: the bill for all this trust comes due. Stolen models, quiet backlash, and the AI your team is already using without telling anyone.
AI and business tech news, verified by a physician who reads the filings. One email a week, no noise.
