We are made for co-operation, like feet, like hands. Inside: an X-ray AI that 30 newborn units taught together, better forecasts weeks ahead, and two more good things. ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏
|
This week in AI
Good news, built by many hands.
|
|
An X-ray check for newborns that 30 hospitals trained together. Better forecasts weeks ahead. Art described aloud for blind visitors. Three minutes, start to finish.
|
|
|
|
Stoic anchor
|
“For we are made for co-operation, like feet, like hands, like eyelids, like the rows of the upper and lower teeth.”
Marcus Aurelius · Meditations 2.1 · translated by George Long
Marcus meant that people are built to work together, the way two hands are. This week’s best news came from exactly that: newborn units in 20 countries and territories teaching one AI model in turn, and a museum that shaped a new tour with blind visitors.
|
|
|
The big one · Health
An adult X-ray AI, retrained for newborns.
A breathing tube leaves little margin for error, especially in a newborn. Several AI tools that check a tube’s position on an X-ray are FDA-cleared for adults, but none for infants. Researchers at Harvard Medical School took an adult model and retrained it one hospital at a time, mimicking a setup where patient data never leaves each hospital. They reported it in Nature Biomedical Engineering on Oct. 13.
|
0.67
|
Average error, cm
locating the tube tip and the point where the windpipe splits, vs. 4.30 cm for the original adult model
|
|
30
|
Newborn ICUs
in 20 countries and territories, 50 X-rays each for retraining
|
|
29
|
Of 30 hospitals
where its error was lower than a model tuned on that hospital’s own X-rays
|
Children make up less than 1 percent of public medical imaging datasets. The team thinks the same approach could carry other adult AI tools over to kids.
The catch: the study used X-rays taken in the past, and the authors say trials in real clinical workflows are essential before anyone relies on it.
Source: Nature Biomedical Engineering
|
|
Also good this week
|
Weather
Better forecasts three to four weeks out.
Forecasters call the stretch two to six weeks ahead the “predictability desert.” Researchers at MIT, Microsoft and elsewhere built a machine-learning fix that learns from a forecast model’s past misses. In Nature on Oct. 14, it doubled the modest skill score of the European weather center’s AI forecasts, and it came out on top in that center’s 2025 real-time contest against 34 teams. Kenya’s weather office has already reached out about using it for the monsoon season.
Source: Nature
|
|
Access
Art described aloud for blind visitors.
On Oct. 14, Milan’s Pinacoteca di Brera launched a pilot with AI glasses from Meta and EssilorLuxottica. A blind or low-vision visitor stands in front of one of eight selected artworks, the glasses recognize it, and an audio description plays. Blind and visually impaired people helped test and shape the tour. The glasses can be reserved Tuesday to Friday.
Source: Meta Newsroom
|
|
Public services
AI helped patch over a thousand weak spots.
On Oct. 12, the UK government said two pilots that put AI to work on cyber defense have already helped identify and fix over a thousand potential vulnerabilities across government systems. It is now extending the program to protect the public services people rely on every day.
Source: GOV.UK
|
|
|
Unexpected connection
Count the crickets, read the temperature.
In 1897, A. E. Dolbear noticed that at night the crickets in a field chirp in sync, and that warmth sets their pace: 80 chirps a minute at 60°F, 120 at 70°F. His rule: take the chirps per minute, subtract 40, divide by 4, then add 50 to get degrees Fahrenheit. The thermometer was out in the field all along.
Source: The American Naturalist (1897)
|
|
Try this week
One prompt. Not clever.
|
Here's something I've been trying to handle alone: [describe it]. Help me list three people who could help, what each one is good at, and one small, specific thing I could ask each of them this week.
|
Marcus wrote that line in a note to himself about the difficult people he expected to meet that day. Claude can help turn the idea into a few real asks.
|
|
Onward
Good news in AI, with a Stoic anchor.
|
Your turn
What's one thing AI did for you this week that actually helped? Hit reply and tell me.
|
|
You're getting this because you signed up at ryanjeffreysmith.com.
Unsubscribe · View in browser
|
|
