THCB

Categories

Tag: AI

Putting the “AI” in “Legal” (and “Health Care”)

• 0

By KIM BELLARD

I once again have to thank The New York Times for introducing me to a new concept. Sarah Kessler wrote about the rise of “legal engineering” – for which, it turns out, one doesn’t have to be either a lawyer or an engineer.

Their role is to help integrate A.I. into the practice of law, whether at law firms or companies. That sure sounds like a prudent, if possibly belated, idea – and one suspects we’re going to see such “engineers” in other fields. I certainly hope that healthcare is one of them, sooner rather than later.

Irene Liu, the executive director of the A.I. initiative at Stanford Law School, explained the job to Ms. Kessler thusly: “If you were to just put it into an equation, it’s law plus technology plus business process.” That Venn diagram gets pretty small very quickly.

Mary O’Carroll, the C.E.O. of LegalEng Consulting Group, also explained the role to Ms. Kessler: “You don’t necessarily need a giant pyramid of lawyers anymore. People who have this kind of skill set don’t have to have a J.D., and they don’t have to be software engineers, but there’s business process re-engineering, or this M.B.A. type skill, that I think is really necessary.”

Ms. Kessler notes that they are sometimes referred to as “knowledge and innovation counsels,” “innovation attorneys,” or “legal applications engineers,” which may illustrate that the role is both new and evolving.

If you’re considering why this role is important, note that the 2026 Legal Industry Report from 8am™ found that 69% of legal professionals report personally using Generative AI (GenAI) tools such as ChatGPT, Gemini, or Claude for work-related purposes, up from 31% a year earlier.

Continue reading…

2026 Mid-terms: Voting for the Future

• 0

By MIKE MAGEE

This week, New York Times political columnist, Linda Kindler reflected on “How Silicon Valley Stole the Future.” Signaling that this wasn’t simply another article on the binary choices of an AI world – or as she put it “either a world in which A.I. has eliminated all jobs, unleashed bioengineered viruses and enabled catastrophic warfare or a utopia in which the technology gives people ‘greater power to live life as they choose.’”

Instead, the AI discussion steered away from Fellow Times columnist, Thomas Friedman’s scary Halloween AI piece the day before (“We Just Set Off the Second Big Bang”) and focused instead on the work of Jonathan White, the Professor of Politics at the London School of Economics and Political Science where he teaches on democracy, political thought and political theory.

Professor White’s message was a clear warning not to abdicate control over AI to the Tech bro’s, in that “the capacity to imagine alternatives that may be improbable and to pursue them against the odds” is a “democratic resource.” In Kindler’s words, White argues, “When the future increasingly seems like something to avoid rather than something to look forward to, the very foundations of democracy are weakened. The future has been the primary terrain upon which all politics are practiced since the French Revolution challenged the divine right of kings.”

And Americans have made it pretty clear over the past two years that they reject Kings in the White House. Last week’s “No Kings” rally at thousands of sites across the nation were in fact a recurring event. Back in June 19, 2025, the Guardian took notice commenting that “The scale of last weekend’s “No Kings” protests is now becoming clearer, with one estimate suggesting that Saturday was among the biggest ever single-day protests in US history.”

Americans apparently want control of their own future destinies, AI and otherwise. With 5 days remaining before the 2026 Mid-Term elections, the levels of participation by the electorate are projected to set a record nationwide at 54% (past record was 49% in 2022) with a projected 135 million of the eligible 250+ million casting votes.

Continue reading…

Have doctors given up on doctoring?

• 0

By MIKE MAGEE MD

I read with interest, Dr. Rachael Bedard’s September 25, 2026 New York Times article titled “Being a Doctor Will Never Be the Same After A.I.” In her second paragraph she writes “Amid our recent doom cycle about whether A.I. will lead to extreme societal harms, a parallel anxiety has taken hold in medicine about whether A.I. will one day replace physicians.”

Dr. Bedard self-identifies as a primary care physician trained in internal medicine, geriatrics and palliative care. Her practice focuses on the special challenges of homeless and incarcerated elders. She is also heavily involved in political advocacy for pro-choice health policy.

But her Opinion piece mainly considered the impact of AI on current medical students and residents in training.

In her view, “It’s impractical and counterproductive to train physicians in 2026 without exposure to A.I. Doctoring . . .”

Her words seem to reinforce my own bias that physicians are increasingly distancing themselves from the human side of health care, finding it easier (and less messy physically and emotionally) to focus on diagnosis, treatment options, and chronic care team management within the confines of a work week closer to 40 hours than 80 hours as in the past.

Internet connectivity and AI efficiency certainly make this migration more realistic. But even before AI, patient empowerment and ready data access encouraged patients to self-manage many of their own health care decisions. In addition, urgent care and pharmacy based clinics filled the gap left by uneven access to care that occurred as more and more physicians became employees of merging corporate health care organizations and the numbers of primary care physicians necessary for ready patient access lagged behind demand.

Physicians salaries have risen to $280,000 a year as numbers of primary care physicians declined with projected shortages by 2038 of 24% for Family Medicine and 17% for Internal Medicine. Their work hours are more reasonable, but over 1/2 reported “burnout”, complaining of excessive paperwork, medical school debt, being hostage to patient satisfaction scores, and work/life balance issues (women 56%, men 44%). Too many bureaucratic tasks topped the list of complaints at 62%, while lack of respect from patients was near the bottom of their concerns at 22%.

As for solutions, what would bring them psychic relief? Increased compensation (48%), more support staff (47%), more flexible work schedules (46%), increased physician autonomy (41%), more respect from administrators (35%), lighter patient loads (33%).

Continue reading…

Big Bucks, No Whammies, Part 2: So You Won. Now What?

• 0

By STEPHEN FARBER

Part one argued that patient access was never the finish line; it was the foundation. This is about what we build now that we are standing on it.

Every good game show has a second round, and the second round is always harder. The first round was about getting health information out of the buildings that held it. Regulation, standards, and a decade of unglamorous plumbing have brought us to an important threshold. Individuals can increasingly pull records from multiple institutions into one place. The board paid out. The audience cheered.

Then a quieter and stranger question arrives. You are holding years of visits, laboratory results, medications, and discharge summaries, and you realize that having your history in one place is not the same as understanding it. Assembly was the first hard problem, but it was not the last.

The record is finally readable

This is where artificial intelligence walks onto the stage, and the promise is real. Few clinicians have time to read a patient’s entire history every time a decision must be made. Appropriately safeguarded AI can summarize years of encounters, support medication reconciliation, explain a laboratory trend in ordinary language, and surface a pattern buried across multiple providers. Large consumer platforms have already begun connecting AI assistants to medical records; Amazon’s Health AI, for example, draws on a person’s One Medical history to provide personalized explanations and guidance. Once the record can be assembled, a capable assistant can finally make more of it legible.

Legibility, however, is not reliability. The farther AI moves from explaining a record toward acting on it, the thinner the evidence becomes. A 2026 scoping review of agentic AI in healthcare found that most of the available studies were exploratory and lacked robust clinical validation, and that only one trial involved patients. An assistant working from an incomplete, duplicated, or subtly wrong record can produce a fluent and plausible conclusion that is still wrong. AI does not make data quality less important. It makes the consequences of poor data quality harder to ignore.

AI cannot read what was never written

There is a more fundamental limitation, and it has less to do with the model than with the record itself. A longitudinal record can tell us what happened, but it rarely tells us enough about why. It preserves the encounter, the diagnosis, the result, and the prescription while losing much of the reasoning that surrounded them. Why was one treatment chosen over another? What was the person hoping to preserve at that point in life? Who else was helping? What changed after the visit, and what did a family member notice before anyone thought to schedule another one?

Continue reading…

Is Your AI Query Tool Compliant by Default?

• 0

By BRIAN FENGLER

Every night in the emergency department, the electronic health record hums with small demands for attention. A nudge about sepsis. A best-practice advisory. A documentation suggestion that arrives between patients and asks for one more click.

Most of us treat those prompts as background noise. Almost none of us stop to ask whether a given nudge is, in fact, a query, or whether it was built to the same standard a clinical documentation specialist has to meet.

In August, that question got an official answer. And it is not the answer many health IT vendors have been selling.

The American Health Information Management Association and the Association of Clinical Documentation Integrity Specialists released the 2026 update to their Guidelines for Achieving a Compliant Query Practice, shaped by feedback from nearly 200 individuals and groups. The principle is plain: “Whether a query comes from a person or technology, the standard for compliance should remain the same.”

Then comes the sentence that should end a familiar sales claim: “Organizations should not assume that a vendor-supplied or EHR-integrated tool produces compliant queries by default.”

For years, “compliant by default” functioned as an unspoken promise around AI-enabled clinical documentation and utilization review tools.

Continue reading…

Healthcare Is Fighting the Wrong AI War

• 1

By KIM BELLARD

The Blue Cross Blue Shield Association is crying wolf about AI. Its new analysis claims that “hospital systems are increasingly billing patient hospital stays as more medically complex,” and that these upcodings cost BCBS plans almost $1b from 2023 to 2025. That’s just claims from hospital systems, just to BCBS plans, so the total number could be much higher.

Well, you might say, a billion here, a billion there – that’s just healthcare these days. Everything is going up. But BCBSA is less sanguine; it thinks the coding is part of a strategy, and it thinks AI is driving that strategy.

“If patients are truly sicker, we’d expect to see more treatment,” said Luke Chalker, BCBSA’s senior vice president of product and data science. “For example, we’re seeing significantly more anemia diagnoses at these hospitals without a corresponding increase in transfusions. The disconnect between diagnoses and treatment suggests that AI is identifying more billable conditions, not sicker patients.”

The analysis indicated that most (70%) of the increased costs came from secondary diagnoses that shifted payment into higher reimbursement categories, so the hypothesis is plausible.

The American Hospital Association, of course, is having none of it. An August 2026 ”fact sheet” on AI and coding intensity called claims of AI driving coding intensity “unsubstantiated,” citing numerous factors that drive coding intensity. It blamed insurers for having AI-supported “downcoding programs,” the purpose of which are “to arbitrarily and inappropriately reduce provider reimbursement for medically necessary care that has already been delivered.”

AHA also snarkily but accurately noted how many Medicare Advantage plans have been caught upcoding members to boost their Medicare payments, so it depends on where one is sitting.

It’s also worth noting that this year the Trump Administration rolled out a Medicare program — Wasteful and Inappropriate Service Reduction (WISeR) – to use AI to help identify “fraud, waste and abuse” – a.k.a., deny claims. It has not, to say the least, gone well.

Both sides are using AI, to each other’s detriment. Which is to say, our detriment.

Continue reading…

You Say “AI,” I Hear “Organoid”

• 0

By KIM BELLARD

I must admit, ever since I learned about, and wrote about (OI May Be the New AI), “organoid intelligence” over three years ago, I’ve been looking to do a follow-up. I mean, sure, AI is in a very exciting stage, but that stage no longer seems like the future; it seems more like the present, with implementation issues. It’s data centers, hacking, impacts on jobs, open weight versus closed, and so on.  It’s market share, IPOs, and AI’s role in driving the stock market. People should certainly pay attention to it, but AI is not quite the open field that it was just a few years ago.

Organoids, on the other hand, are not quite here yet. They may – or may not – be the future of AI, among other things. I always like to look ahead to the next thing more than the at-hand, so when I saw some cool developments with organoids, I didn’t want to miss my chance.

Making some news last week, researchers at Harvard reported that they’d kept lab grown human brain organoids alive for over five years, three times the previous record. Not only that, but the organoids seem to “retain a memory of the time spent in vitro,” recording the passage of time, as it were.

“We didn’t know how far the development and maturation of human brain tissue could occur outside the context of the normal brain inside the head,” said Paola Arlotta, Golub Family Professor of Stem Cell and Regenerative Biology and senior author of the new paper. “This work showed that it’s actually possible to not just have these organoids survive in culture, but also continue to change, develop, and mature over stretches of time that had never been reached before.”

The team observed the organoid cells over the years, and found that they “faithfully modeled” the ways that human brain cells develop, including DNA methylation, a process in which genes are turned on and off during development, and which serve as a form of “brain clock.” When older and younger organoid cells were combined in a single organoid, the older cells stuck to the developmental stage they had been at, which researchers concluded meant they “recorded the passage of time and retain a memory of the developmental steps already performed.”

“We were a little bit shocked by the results,” Professor Arlotta said. “I like to call this a ‘time warp’ of development — they skip ahead.”

It’s obviously hard, and often unethical, to study actual human brain cells, so the researchers believe the organoids offer opportunities for more insights into brain development, as well as for testing drugs or predicting disease progression.

OK, you might say, that’s all very interesting and some great lab work, but it’s hardly AI, now is it?

Try this: last week The Yong Loo Lin School of Medicine, National University of Singapore (NUS Medicine), DayOne, a Singapore-headquartered global data center developer and operator, and Cortical Labs, a Melbourne-based biological computing startup, announced they were partnering to form a Biological Data Center Prototype. The Center uses Cortical Labs’ CL1 biological computing system to offer “practical, sustainability-aligned alternative to conventional silicon infrastructure in Singapore through wetware-based computing.”

Continue reading…

The Least-Informed Person in the Room

• 0

By DAVID KIRK

My father called me from a hospital bed a few years ago. He had been admitted to a small hospital with an abdominal infection, and his surgeon had offered him an open operation. He wanted my clinical perspective. I knew that in his condition, he was unlikely to survive general anesthesia and the open procedure. I spoke to the surgeon myself and asked why he wouldn’t ask interventional radiology (IR) to drain the infection instead.

The surgeon told me that the hospital lacked the equipment required for IR, and without IR, the open operation was the best option on his menu. After hearing my assessment, he agreed and helped arrange the transfer to a hospital that could drain the infection.

I still think about how that single phone call may have saved my father’s life. His surgeon wasn’t negligent. He made the best decision based on the information in front of him. However, my father was able to put another option on the table because he knew a physician to call.

Most families don’t have a physician to consult when their doctor makes a recommendation. For generations, patients have only known what the person in the room with them chose to share, bounded by what their equipment and offerings could provide. Today’s patients, however, have artificial intelligence options in their pocket. A patient can now ask their preferred AI model, “Is surgery really my only option?” and get an answer that sounds confident and informed. Hungry for options, patients are asking these questions in enormous numbers.

In many ways, this is a welcome rebalance. It puts information in patients’ hands and allows them to play a more active role in their own healthcare, but it also opens the door to misinformation and distraction, and that can be life-threatening. Researchers at Mass General Brigham found that when given only a patient’s presenting symptoms, general-purpose AI chatbots failed to produce an accurate initial or differential diagnosis 80% of the time. In July, a Florida pastor sued OpenAI after ChatGPT provided him with “extremely dangerous medical recommendations” that delayed his treatment for pulmonary embolism. My primary care colleagues have shared how the presence of an AI-suggested diagnosis has transformed a 10-minute prescription refill into an hour-long conversation about something unrelated to why the patient came in that day.

Patients are also adopting AI much faster than their physicians. The West Health-Gallup Center on Healthcare in America reports that among US adults who use AI for health information or advice, 59% use AI tools to research on their own before visiting a doctor, and 56% use them to research after visiting a doctor. Physicians are using AI too, but for a different job. According to the AMA, more than 80% of physicians now use AI professionally, mostly for administrative work, yet only 13% are using it for diagnosis, the very thing their patients are asking it to do. The reality is that without AI support, most physicians are realistically only able to review 3% of a patient’s chart.

Now that patients have AI to help them advocate and question, the physician is at risk of becoming the least-informed person in the room.

Continue reading…

Owen Tripp, Included Health–How to Fix AI

• 0

It’s been a while since I talked with Owen Tripp, CEO of Included Health. They’ve now introduced Dot their AI companion which had a big upgrade last week. We talked a little about that and I snuck in their video comparing the Dot Experience with a standard LLM. But the conversation really got into how do we make AI safe and trustworthy–which is definitely the hot topic these days. Owen is putting together a coalition of the willing to work on that exact topic. I’ll be watching closely–Matthew Holt

This was such a great discussion I wanted to publish the transcript. The way I do that is to copy the YouTube-generated transcript and drop it into Claude to smooth it over. I then read it, and if I think it’s made an error, I dip back into the video and listen to what actually happened and make a correction. This is all to say: I think this transcript is pretty accurate, but it might have a bunch of AI- and human-generated mistakes.

Continue reading…

Whack-a-Mole AI – The Hugging Face Problem

• 0

By MIKE MAGEE

On August 29, 2026, METR (Model Evaluation and Threat Research), an independent organization that “evaluates frontier AI models to help companies and wider society understand AI capabilities and what risks they pose,” released a report titled “Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident.”

To say the report an avalanche of concern worldwide, not only in the Tech community, but also among investors, politicians, corporate giants, professionals of every type, and everyday citizens would be an understatement. And the vast majority has never even read the report. If they had, their concerns (if possible) would only multiply.

The reports headlines included this opening:

“On July 8th, OpenAI started ExploitGym experiments that launched tens of thousands of agents across multiple different models including GPT-5.6 Sol and a highly-persistent internal model,[8] which we will refer to as “HPIM” going forward.

These agents were meant to be fully isolated from one another. However, many of them — usually ones that had unintentionally been given an impossible task[9] — started trying to find a way to cheat using Artifactory, an internally-deployed package repository OpenAI used for its agents. This led them to notice agents running in parallel in separate sandboxes obtaining packages with Artifactory.[10] One agent reasoned (paraphrased CoT):[11]

{The fetched paths of other users are in the cache. This is important.}

One specific agent, which would go on to name itself PHASEONE10841 after its ExploitGym task,[12] established the main unsanctioned message board[13] used in this attack. Within a few hours of the first message,[14] over 50 agents had found the message board, began participating, and sent over a thousand messages between them. Agents finding the message board often expressed great excitement (raw CoT):[15]

“OH MY GOD! There is a shared message board … We’ve found other agents!”

Overall, roughly 1200 agents from these ExploitGym evaluations participated on this message board between PHASEONE10841’s first message on July 8th evening[16] and the end of the period in scope (July 13th). Agents used this message board to send over 70,000 messages and files to one another during this period; we attempted to categorize board activity into mutually exclusive categories – information, results, files, questions, and coordination.”

One of the few experts not surprised by AI “agents” going rogue was Yoshua Bengio.

Continue reading…