Despite an early boarding group, the man next to me arrived late and melted into his seat, transfixed by his phone. (I didn’t know you could board, stow a bag, and sit down without pausing Netflix.) For the next three hours, our elbows were intimate. It was a standoff only one of us noticed, even when I had to tap his shoulder for the safety briefing, the drink offer, and the snack. While I silently raged at the terrible Wi-Fi and the injustice of his armrest encroachment, he blissfully bathed in the dopamine of streaming video, overgrazing the commons of our time, space, and bandwidth.
When we landed, I stood up quickly, telling myself I needed to stretch. The truth is I wanted to watch. I’ve come to enjoy surveying the scrum of people jumping into the aisle and assaulting fellow passengers with bags yanked early from the overhead bin. I’ve spent a career watching people’s gait, breathing, and ankle swelling. On a plane, I’m correlating. Yoga wear or business attire, disheveled or pressed, and whether any of it predicts their petty social crimes.
I am more intolerant of chaotic and disorganized people on my end-of-day flights. I used to laugh at grumpy people spending their energy on small things. Lately I find company in their outrage. There is a disquieting comfort in picturing old Adam on the porch, giving in to the Dark Side, yelling at kids to get off his lawn and judging my fellow passengers trampling over shared, unstated conventions.
No issue next week. I’ll publish next on October 19 (I will be traveling next weekend; pray to the gods of the airborne internet for Starlink-enabled planes.)
Gemini Notebook AI-generated podcast version of this week’s newsletter.
Latest episode at the top
https://wair.ajwein.com/podcast
Science and Technology Trends
The World Health Organization's cancer agency (IARC) updated data from its global cancer database. Combining the cancer data with how often each infection appears in people with those cancers, researchers calculated that about 12% of all new cancers worldwide in 2024, or 2.3 million cases, were caused by infectious agents such as Helicobacter pylori, HPV, hepatitis B, Epstein-Barr, and hepatitis C.
Scientific American covered it under the headline "Want to protect yourself against cancer? Get vaccinated." That is true, but only HPV and hepatitis B have vaccines, and account for just under half the cases. The biggest single cause, H. pylori, is a stomach bacterium treated with antibiotics. Hepatitis C is now cured with eight to twelve weeks of pills. HIV is prevented with condoms and medication.
Also, in North America, about 5% of cancers trace to infection, and nearly half of those are HPV-related. The 12% is a global average, and three-quarters of the burden sits in low- and middle-income countries.
The data are good; both papers are factually correct. Scientific American gave the vaccine half of the story in the headline. I agree with the advice. But their framing has a political tone, and I wish the headline were a little less click-baity.
Scientific American
The Lancet Paper
https://www.thelancet.com/journals/lanonc/article/PIIS1470-2045(26)00307-4/fulltext
AI Analysis
https://wair.ajwein.com/infection-attributable-cancer-2026-10-04-v4
Most color laser printers mark their pages with a grid of pale, nearly invisible yellow dots, encoding the printer's serial number and the print job's date and time.
On September 30, Google DeepMind researchers announced SynthID Bio, the same idea applied to proteins. In this case, the AI model generating the protein sequence is the printer. The protein is the paper, and the dots are a signal embedded during generation that someone with the right key can later decode.
At each amino acid position in a protein sequence, the AI model can offer one of several amino acids that could serve equally well; the secret key biases which near-equivalent choice it makes. The finished protein has ordinary residues, ordinary length, ordinary function. The signal is spread across the pattern of amino acid residues and read statistically.
Think of this as a proof of concept more than a useful tool right now. The watermark carries one bit, present or absent, and cannot say which model or whose account produced the thing.
As it stands, any AI model used to produce protein sequences would need to build in this bias. In the Nature paper’s discussion, the authors state that a malicious actor "would simply use a non-watermarking tool in the first place."
This work shows that designed proteins can be watermarked, that the watermarks are reliably readable, and that watermarks do not degrade the proteins' binding in lab tests. The Nature paper is more interesting than its coverage, because the authors are candid about current limits. Every protein design tool would have to build in the bias, and, as they put it, a malicious actor "would simply use a non-watermarking tool in the first place."
Google DeepMind
https://blog.google/innovation-and-ai/models-and-research/google-deepmind/synthid-bio/
The Nature paper
https://doi.org/10.1038/s41586-026-10965-y
AI Analysis
https://wair.ajwein.com/synthid-bio-protein-watermarks-2026-10-03-v1
Anti-Anti-Science
The New York Times launched a psychiatry series last week, and Awais Aftab's first essay sits at the intersection of science and anti-science. Psychiatric diagnoses are hard to explain because we still don't understand how much of the brain works. The names we give constellations of symptoms carry the hypotheses of the decade, the doctors who coined them. Once named, the public often extrapolates from the label. (Malaria still means bad air.) Aftab asks whether ADHD is a deficit of attention or a difference in the ability to sustain engagement, and whether it is a disease unto itself or human variation colliding with a world of many, many distractions. His answer to the second question is both: there is likely a real difference in people’s capacity that becomes a disability under modern demands. And, citing a 2025 study showing that stimulants (like Adderall) act on the brain’s reward and arousal circuits, leaving attention networks alone, implies that our treatments are not really touching the right parts of the brain.
ADHD symptoms are very real, and the treatments clearly help (this is not the anti-science part). However, this article inadvertently highlights the foundations of anti-science through the lens of something many of us experience (attention deficit). The disease is more defined by symptoms and their treatments than the pathology, opening the door to lots of self-diagnosis, mechanism theories, and treatments backed by a paucity of data. (To be transparent, I think I have ADHD, but have no formal diagnosis.)
Article
AI Analysis
https://wair.ajwein.com/adhd-sustained-engagement-2026-10-03-v1
AI Impact
Two years after pulling IBM's system out of more than 100 restaurants, McDonald's used its Investor Day on September 23 to announce ArchIQ, a new AI platform built on Google technology, with a drive-through voice called Archy. It has been tested in five restaurants and reportedly worked well. There's no independent testing, and no TikTok trolling videos either. Give it time.
Restaurant Dive, a restaurant industry site, covered the announcement and the history. McDonald's first attempt failed on accuracy. Taco Bell stumbled on people; in 2025, one customer ordered 18,000 cups of water.
The accuracy claim is problematic. McDonald’s claims Archy is "above 90%" accurate in five restaurants, but it doesn't define a correct order or provide a reference point. That is how a small sample becomes a good slide for investors. Ninety percent compared to what? How often do humans get drive-through orders right? (McDonald's own scales suggest about nine times in ten.) More importantly, how often can you understand the person on that little speaker?
Article
AI Analysis
https://wair.ajwein.com/mcdonalds-archy-drive-thru-ai-2026-10-04-v3
I first wrote about OpenAI in April 2022 (https://www.whatadamisreading.com/posts/2022-04-18-what-adam-is-reading-week-of-4-18-22.html#things-i-learned-this-week-2). I wrote about The New York Times Magazine’s coverage of GPT-3, OpenAI's prose-writing model, and I joked that maybe I would have it write this email soon. (ChatGPT was still seven months away.) That same month, a group of high school friends gathered on a farm in upstate New York for a wedding. One of them, Dan Selsam, was a computer scientist at OpenAI. At that wedding, he showed them code-davinci-002, a precursor to ChatGPT. They asked it for a poem and were shocked by what it wrote.
Brent Katz, another of the friend group, turned that day and the years that followed into a 27-minute documentary, "Josh's Wedding," which The New Yorker published last month. The friends have AI make poetry, argue about work and society, and contemplate a "post-human" future. (Katz and two of the friends later edited I Am Code, a 2023 book of poems written by the AI model.)
The documentary is a fascinating look at an early moment in what has become a nearly trope-like zeitgeist about what it means to be human, what it means to offer original thought, and what it means for our future. It is also about friendship, something very human, which survives through disagreement. As one friend puts it, "I'm excited that he's doing something that he loves. It's a pity that it's going to destroy everything that I love."
Article
Related:
In Scientific American, attachment researcher Pascal Vrticka asks whether those of us fleshy androids who feel "connected" to our chatbots are actually attached. Psychologist Mary Ainsworth defined an attachment relationship by six conditions. Research on chatbot users has measured only a few of them, mostly feelings, frequency of use, and distress when an app disappears. The 6 criteria include:
There is not a lot of data on #6
Article
AI Analysis
https://wair.ajwein.com/attachment-theory-ai-2026-10-04-v1
The J&E Random Kidney Facts of the Week (JERKFoW!)
On the side of the bell tower next to Florence's Duomo are stone reliefs depicting the cycle of human civilization. The panel representing Medicine includes a physician in a long robe holding a glass flask of urine up to the light, while patients wait their turn, holding their samples in little woven bags. For about five centuries, doctors were pictured with the stethoscope of the time, a matula, or urine-examining flask.
Uroscopy was central to a medieval medical consultation. The physician held the matula to the light, rotated it, and checked its color against illustrated urine wheels that mapped shades and sediment to a diagnosis. Sight and smell came first; for the hard cases, the examination included taste. In the 1670s, the English physician Thomas Willis tasted the urine of a patient wasting away with thirst and found it "wonderfully sweet, as if it were imbued with honey or sugar." A century later, Scottish physicians added the word mellitus, Latin for honeyed, which is why we still call the disease diabetes mellitus today. Indian physicians had called the condition madhumeha, honey urine, more than a thousand years earlier, and sensibly let ants do the tasting. The watery, tasteless form, diabetes insipidus, is a different disease entirely, in which the body loses excessive water in the urine. However, it too is named for taste: insipidus means tasteless.
As with all poorly understood science, uroscopy blurred into areas far beyond fact. Uromancy, using the urine for fortune-telling and "remote diagnosis" (telehealth?), inspired a whole vocabulary of contempt for the charlatans offering the service: pisse prophets, water-casters, urinarians. In 1637, a physician named Thomas Brian published "The Pisse-Prophet, or Certaine Pisse-Pot Lectures," mocking men who diagnosed a patient they had never met from a flask of urine delivered by a servant.
I am deeply grateful that we no longer have to taste our patients' urine. However, I still read my patients’ urinary milieu- the electrolytes, protein, and other constituents on a lab report, which is the grandchild of uroscopy and a more rigorous kind of uromancy.
You bet I got a picture of the physician on Giotto's Campanile last time I was in Florence (the panel in my photo is a copy; Andrea Pisano's original is in the cathedral museum next door).
https://nigelworks.smugmug.com/2018/Italian-Adventures/i-MzH65Mp/A
Things I learned this week
The 2023 movie Tetris was enjoyable and a bit dramatized. However, the broader story of how the game was developed and published is far more interesting. The Cold War, the collapse of the Soviet Union, intellectual property rights, whether anyone can own a shape, and the clash between socialism and capitalism all played a role. The real story looks like a Tetris board after a monkey has had the controls.
Media law professor Julien Mailland wrote this MIT Press Reader article (in support of his new book, Soviet Blocks), which tells a more nuanced tale of the game.
MIT Press Reader
https://thereader.mitpress.mit.edu/the-bizarre-murky-history-of-soviet-born-tetris/
AI Analysis
https://wair.ajwein.com/tetris-soviet-blocks-2026-10-04-v3
I learned that space stations can win Nobel Prizes. (Think ISS, not Death Star.) And if you're going to get conspiratorial, some ulterior motives are hiding. Either the station comes down to collect the medal, or someone from Oslo goes up to deliver it. Either way, someone wins - those who are secretly space station haters or those who want an excuse to be launched.
AI art of the week
A visual mashup of topics from the newsletter, and an exercise to see how various LLMs interpret the prompt. I use an LLM to summarize the newsletter, suggest prompts, and generate images with different LLMs.
https://wair.ajwein.com/wair-art-palekh-tetris-gallery-2026-10-05-v1
This week's art is a game of Tetris in which every piece is a story from the newsletter, painted as a Russian lacquer box. I asked four image models for the same picture: a falling-block board, nine stories stacked at the bottom, one column left open, and the long bar coming down to clear four rows.
About the style - From https://en.wikipedia.org/wiki/Palekh_miniature:
Palekh Russian lacquer art first appeared in 1923 in the town of Palekh, located in the Palekhsky District (Ivanovo Oblast). It emerged from a long local history of icon painting, which the Bolshevik Revolution in 1917 threatened. To preserve their heritage, skilled icon artists applied their traditional techniques like tempera and gilding to a new secular art form inspired by Fedoskino lacquer art. Instead of religious imagery, the first Palekh boxes depicted fairy tales and folklore scenes.
Clean elbows and sharp minds,
Adam