ChatGPT Can Replace Dr. Google, But It Shouldn't Be Your Physician

 AI, LLM, ChatGPT, GPT-4… the acronyms of the Artificial Intelligence world flooded the realm of medicine these past two weeks. Yes, everyone everywhere is talking about AI, but when doctors — who normally pay little attention to anything of interest to the lay public — begin to get excited, I know something is afoot. That something seems to be a sense that AI is coming for our jobs.

The staid New England Journal of Medicine, a reliable late-adopter, is starting its own AI journal, for decency’s sake. It’s all-AI, all the time!

Now, contrarian observers might be thinking that when the NEJM gets into something, it’s time to sell. However, I think it’s almost an inevitability that AI will increasingly worm its way into the practice of medicine. In some ways, the movement will be a boon for patients. In others, it might prove to be a disappointment, or perhaps far worse. In any case, the conversation in health care is getting caught up in the fervor of the incredible efficiency which AI might bring to medicine, and, on the flip side, the obvious concerns about data security and automation. I think there’s more to discuss, though. Yes, before too long your doctor might be a computer, and perhaps even a full-on robot; it’s worth a good think, though, if that’s really the direction we want to head.

The Good

I’m not a wet blanket about the potential benefits of AI in medicine. I’m impressed with what I’ve seen the past couple weeks, observing what GPT-4 pops out in response to queries both obscure (“Does the genotype of Hereditary Alpha Tryptasemia affect the phenotype?” and commonplace (“Has any treatment for post-viral cough been proven effective by randomized controlled trial?”). It generally crushes me with the quality of its answers, to be sure, and I fancy myself a rather competent physician.

A recent NEJM article reviewed many of the ways medicine is likely to adopt AI. In citing the example of its ability to reason through a medical board test question, we see how well Chat-GPT can perform:

Now, I think most physicians would, like me, read this question and think, “Oh, this is that post-strep kidney problem;” but most of us who are not nephrologists might not remember that C3 levels would be abnormal in this setting. Chat-GPT acquits itself quite well in working through the problem logically — far better than I did.

Physicians are asked to remember and apply an absurd amount of knowledge. The internet has greatly reduced the need for a remarkable memory. Even the current iteration of GPT-4, despite not being specifically trained on medical cases, marks a huge upgrade in this regard. Information is not only rapidly gathered, but curated, as well. Ask about C3 levels in post-streptococcal conditions and you won’t just get a long list of studies, many of them esoteric, which may or may not answer the question at hand; you can request practical information, condensed and delivered in an instant.

This quality is also a great win for patients — and their doctors, therefore, too. Many of us have spent a good bit of time hand-holding the anxious patient who googled “fever for 4 days with skin rash” and came to the office convinced of their child’s leukemia diagnosis. Once patients relinquish “Dr. Google” for “Dr. ChatGPT",” they will get much, much better free medical advice.

So, yes, this technology is already poised to do a great deal of good in medical practice. It is virtually plug-and-play ready to serve as a backstop for clinical decision-making; to fill in the holes of any physician’s knowledge base in interpreting a lab test or imaging report; to write a chart note that is actually comprehensible while billing the right ICD-10 codes, simultaneously freeing up a doctor to have an extra five minutes with each patient; to coach us if we might have missed something important in a patient’s presentation. All good things!

Some of this goodness will come at a cost to our profession, but a savings to our system. Do you have a child in medical school competing viciously for a prized spot in a Dermatology or Radiology residency program, or a less-prized spot as a Pathologist? For God’s sake, tell them to match in a specialty that will still exist in five years! Those medical fields that are almost entirely pattern-recognition are at a grave threat of extinction.

The rest of us? Well, we tell ourselves that the “art of medicine” is still king. I’m not sure that argument will hold up over the coming years (maybe months!) — I suspect the next iteration or two of ChatGPT is likely to out-compete me as a clinician in almost every way, even when it comes to nuanced decision making.

That said, I still don’t think your next doctor should be a robot. Allow me to explain.

Garbage In, Garbage Out

Some things in medicine really can be competently managed by an algorithm. Choosing meds for high blood pressure, for instance, can be done as well or better by computer than physician, even if some nuance might be lost along the way (“Does she really take her meds?” “Is it time to double the lisinopril dose, or recommend a counselor to help with the stress over his step-son?”). Sometimes, though, it takes experience and judgment to wade through the conflicting recommendations of expert bodies, or decide which studies are reliable enough to update clinical recommendations before they are hashed out in formal guidelines years later.

The Covid-19 pandemic provides glimpses into this. The CDC made some asinine recommendations; would GPT-4 be able to sort out that in no universe could risk:benefit favor a healthy adult getting a second Pfizer shot if they had developed myocarditis from the first, or a twice-vaccinated male college student who had already gotten two mild cases of Covid receiving a booster shot? We can’t know since the training period for GPT-4 ended in September, 2021, and it will not wade in on such questions, but I worry. Ditto the use of Paxlovid in moderate-risk patients; most agencies push it, most savvy observers dispute it, and our only high-quality study suggests no benefit. I would be impressed at an AI tool that could figure out the side of the savvy observer over the less-savvy conventional wisdom.

An example not constrained by GPT-4’s training window is a favorite pet peeve of mine in the management of diabetes. An old and mediocre class of drugs, the sulfonylureas, have become outdated due to their propensity to cause dangerous low sugars and weight gain (and also to not work all that well over time). From many conversations with endocrinologists, I feel confident saying that there is almost no situation in which they should be continued if a patient is moved onto insulin. However, some poorly-constructed official guidelines still exist which encourage this practice. Could GPT-4 see through the noise? Not quite:

This is part of what makes medicine so hard. Sometimes money talks louder than evidence. Other times, we fall in love with convention, or are too easily wooed by the latest shiny new device. Genuine experts often disagree. Sometimes, however, thoughtful review of the evidence reveals that one side is clearly right! I acknowledge that anything I can figure out with my gorilla-brain can eventually be figured out, and then some, by a well-trained AI. However, we’re not there yet.

Reading the Room

Whatever constitutes “the art of medicine,” some of it is tuning into your patients’ unspoken cues. Most medical students are terrible at this; many of us become competent over time in this regard, however, with the help of repeated mistakes and the occasional gifted mentor.

A classic example is the “toxic” child. This is not what it means on social media, but rather a truly sick pediatric patient, one needing emergent care. Most infants and toddlers, crying from their headache and muscle pains with a 104℉ fever, will look pretty rough. Almost all of them can go home with their parents after some counseling. Every now and then, though, a child is just not acting right. The same 14 month old who was wailing with horror when you tried to examine them at their well-child visit a couple weeks ago, now calmly sitting in their mother’s lap with slightly glassy eyes and tolerating your stethoscope with nary a peep? Time to get worried!

Will a very capable computer figure out how the same behavior that could pass for “normal” with some patients in some settings is actually a red flag for serious illness in other contexts? Maybe. This strikes me as something that falls in the “not really playing to AI’s strengths” category — at least for now.

The Human Touch

To drop into the obvious for a moment: it’s nice to have another human being care for you. Now, when that human being is exhausted, running late, trying to compress an entire visit, chart note, and sick letter into fifteen minutes, and has 2500 other patients, you might not be getting a lot of that “care.” That’s a serious systems problem afflicting how we practice medicine in this country. Still — I think that even a disappointing encounter with a human is better than a leisurely conversation with a robot, in terms of its healing potential.

I know some people are fine interacting with non-sentient beings; some, it seems, even prefer a computer being to an actual boyfriend or girlfriend. However, a robotic doctor? That’s unlikely to be a therapeutic milieu. Since many studies have intimated that a doctor’s belief in the quality of the treatment they dispense — even when a placebo — will affect outcomes, it’s fair to extrapolate that having a human treating your ailments might be inherently healing. As community dwellers, we humans are also used to making agreements with other people, including agreements about managing our own health, and most of us have an innate desire to keep our end of the deal as part of our societal contracts. I don’t feel this way towards customer service chat-bots.

Worst of all is to envision a robot informing you that your CT showed a cancerous tumor. Ugh. There is no velvet-voiced, sweet-as-ET droid I could ever imagine as acceptable for breaking bad news.

Trust and Security

The practice of medicine is nothing without trust. Every doctor has felt the sting of a dubious patient — it is painful, not to be trusted doing the thing that we devote our lives to trying to do well. We get used to being trusted; as one of my first family medicine preceptors laughingly told me when I commented how much his patients loved him, “Oh, I don’t take that personally — the ones that didn’t like me all left a long time ago.” I don’t like making mistakes in medicine, but they’re inevitable; I don’t care for being fired by a patient whose expectations I didn’t meet, but it happens; but for a patient not to trust that I am on their side, doing my imperfect best to serve their interests — that would be crippling.

So: how do we trust an AI robot to be working in our best interests? Can we have a long backstory with it, watch how it behaves at church or in the community, see whether it respects my elderly parents at the supermarket? Nope. Do we know that it will have the same (programmed) values today that it displayed yesterday? Negative. Does even its programmer know how it will behave as time passes and inputs accrue? Apparently not. What if its programmer actually wishes to weaken your health, sell your secrets, or influence you to spend money on certain medical products which enrich the programmer’s company or nation? Gulp.

For all my pea-brained limitations, my patients know exactly what to expect when I walk into the exam room. If I come in reeking of bourbon, or complaining about all the spiders crawling on my skin, they know to walk out and try to find me some help. It might not be so obvious with a computer.

Of course, on a personal level, such concerns are worrisome; on a societal level, they are terrifying.

To properly harness all the data that could train AI systems to be fully-actualized physicians, we will be told that all our medical information must be up in the cloud where it can be accessed. We will be assured that it will remain private. We will be promised that the fruits of knowledge, the cost savings, the suffering and death prevented, will be far more valuable than the risks of automating and integrating our entire health care systems.

There will be merit to these arguments. They’ll probably win the day, because Science. Government and Industry will agree that the tremendous upside of turning health care over to AI is worth the risks.

It just makes me nervous. It’s scary enough to imagine the bulls-eyes that AI-controlled energy, transportation, and financial systems will have painted over them for our adversaries; but whole health care systems, too? Hospitals that are reliant on AI-augmented nurses, doctors, and surgeons to function will also be attractive targets. Malware attacks on Electronic Health Records systems are bad enough now, but once those knowledgeable humans who care for patients are replaced to the maximum extent possible by highly efficient computerized systems, taking down those systems could make a whole country cry uncle in precious little time.

I generally try to avoid futurism, because the most reliable aspect of the future is that it probably won’t look like you think it will. (Sports betting teaches me this lesson on a regular basis.) However, much as I suspect high-end concierge practices soon enough will begin to proclaim their use of EHR systems that integrate AI technology as a selling point, within a matter of years a few terrible stories will emerge that will start to push them in the opposite direction. “My practice is unplugged” or “Paper charts only, like your grandfather’s doctor kept” will be the new head-turners.

I’m not sure this is a bad thing. Yes, the internet has been a godsend for medicine; it has democratized medical knowledge away from the medical center libraries and academic journals and put it in everyone’s hands, at a faster pace than ever before. However, there is no device capable of reasoning through complex medical problems less prone to hacking, malware, and corruption than the brain of a physician. No large scale attacks can be made on this form of intelligence. Yes, the occasional foolish doctor falls for some nonsense on Twitter, or decides to tailor their oncologic recommendations towards the companies who pay their lecture fees, but these are not system-crashers, and are easily enough discerned and discarded. Ironically enough, perhaps the only way to protect our hard-won, computer-fueled medical knowledge is to store it primarily in our brains.

The heat of the excitement as one AI medical breakthrough after another is announced will have a shadow side in medicine. Once a computer can out-doctor an experienced physician, why keep pumping out physicians? It’s horribly inefficient, as well as expensive, to train doctors from scratch. If patients — the consumers of health care — don’t demand old-fashioned human doctors, with old-fashioned training, we risk losing a profession and an art which we might someday need to survive.

Everything I know was taught to me by tens of thousands of patient encounters, with a liberal sprinkling of didactic leaning and apprenticeship along the way. Much of that learning is stored, sometimes even in a format that can be rapidly retrieved! Always, though, it is available to help another soul, and always with the charge to “do no harm.”

It’s hard to replace that with a computer. Let’s not try.

https://doctorbuzz.substack.com/p/chatgpt-can-replace-dr-google-but

The Smearing of Clarence Thomas

 The left’s assault on the Supreme Court is continuing, and the latest front is the news that Justice Clarence Thomas has a rich friend who has hosted the Justice on his private plane, his yacht, and his vacation resort. That’s it. That’s the story. Yet this non-bombshell has triggered breathless claims that the Court must be investigated, and that Justice Thomas must resign or be impeached. Those demands give away the real political game here.

***

ProPublica, a left-leaning website, kicked off the fun with a report Thursday that Justice Thomas has a longtime friendship with Harlan Crow, a wealthy Texas real-estate developer. The intrepid reporters roamed far and wide to discover that the Justice has sometimes traveled on Mr. Crow’s “Bombardier Global 5000 jet” and that each summer the Justice and his wife spend a vacation week at Mr. Crow’s place in the Adirondacks.

The piece is loaded with words and phrases intended to convey that this is all somehow disreputable: “superyacht”; “luxury trips”; “exclusive California all-male retreat”; “sprawling ranch”; “private chefs”; “elegant accommodation”; “opulent lodge”; “lavishing the justice with gifts.” And more.

Adjectival overkill is the method of bad polemicists who don’t have much to report. The ProPublica writers suggest that Justice Thomas may have violated ethics rules, and they quote a couple of cherry-picked ethicists to express their dismay.

But it seems clear that the Court’s rules at the time all of this happened did not require that gifts of personal hospitality be disclosed. This includes the private plane trips. ProPublica fails to make clear to readers that the U.S. Judicial Conference recently changed its rules to require more disclosure. The new rules took effect last month.

Justice Thomas would have been obliged to disclose gifts that posed a conflict of interest involving cases that would be heard by the High Court. But there is no evidence that Mr. Crow has had any such business before the Court, and Mr. Crow says he has “never asked about a pending or lower court case.”

The most ProPublica can come up with is that “Crow has deep connections in conservative politics.” Oh dear. One hilarious section reports that a painting at Mr. Crow’s New York resort includes Mr. Crow, Justice Thomas and three friends smoking cigars. One of the friends is Leonard Leo, “the Federalist Society leader regarded as an architect of the Supreme Court’s recent turn to the right,” ProPublica says.

This conspiracy is so secret that it’s hiding in plain sight. Can anyone imagine such a story ever being written about a liberal Justice on the Court?

Justice Thomas said in a Friday statement that “early in my tenure at the court, I sought guidance from my colleagues and others in the judiciary, and was advised that this sort of personal hospitality from close personal friends, who did not have business before the Court, was not reportable.” He added that “these guidelines are now being changed,” as the Judicial Conference announced new guidance. “And, it is, of course, my intent to follow this guidance in the future,” the statement said.

***

None of this will stop the political stampede to portray the Crow-Thomas friendship as a scandal to tarnish the Justice and the current Court. This campaign has been underway with particular ferocity since the Court’s majority shifted during the Trump years toward Justices who have an originalist view of interpreting the law and Constitution.

Justice Thomas is the leading target because his long-held views on the law have found new adherents. As the longest-tenured Justice now on the Court, he has the power to assign writing duties for opinions when he is in the majority but Chief Justice Roberts isn’t.

“Is Supreme Court Justice Clarence Thomas corrupt? I don’t know,” tweeted Democratic Rep. Ted Lieu. “But his secretive actions absolutely have the appearance of corruption. And he apparently violated the law. For the good of the country, he should resign.” You gotta love the “I don’t know” but he should resign anyway formulation.

The liberal press—pardon the redundancy—has climbed onto its ethical high horse and is demanding “reform” at the Court. “All of this needs robust investigation,” demanded Sen. Sheldon Whitehouse. Senate Judiciary Chairman Dick Durbin dutifully saluted upon Mr. Whitehouse’s order and said his committee “will act” to impose a new “enforceable code of conduct” for the Court.

This ethics talk is really about setting up an apparatus that politicians can then use against the Justices if there is any transgression, however minor or inadvertent. The claims of corruption are intended to smear the conservative Justices and tarnish the Court to tee up case recusals, impeachment or a Court-packing scheme if Democrats get enough Senate votes to break the filibuster.

It’s all ugly politics, but the left is furious it lost control of the Court, and it wants it back by whatever means possible.

https://www.wsj.com/articles/clarence-thomas-supreme-court-propublica-harlan-crow-1c4c2f41

California's Education Priorities Focused On Equity Quotas

 by Christian Milord via The Epoch Times,

A March 26 front page article in the Orange County Register bemoaned the fact that too few black and Latino males become teachers in the California public schools in proportion to Caucasian men and women. It didn’t mention Asian male teachers, but painted minority teachers as “people of color” as if Caucasians don’t come in many shades of skin tone. The article claimed that many minority males are hesitant to enter teaching as they think teaching in the K-12 public schools is primarily a female profession.

The column went on to mention that most public school teachers in California and the Southland are Caucasian men and women. Superintendent of Public Instruction Tony Thurmond has discussed this issue before, while state and university initiatives have been drawn up to remedy this alleged “inequity.” The state aims to recruit, train, and retain greater numbers of minority group teachers. California State University–Fullerton is one of several universities taking on this proposal.

On the surface, there is nothing wrong with encouraging high school and university students to pursue education careers as well as STEM professions. However, if you dig deeper, you find that these initiatives have little to do with equality of opportunity, but rather are an attempt to check off boxes in order to attain government-driven “equity” quotas. Sometimes initial good intentions can end up with biased outcomes.

Moreover, the article assumed that minority teachers want to instruct students who look like them, and students want teachers who look like them. This takes for granted that these minorities have shared life experiences. What kind of objective research was carried out to arrive at this conclusion? Don’t most parents and students hope for teachers who are qualified and possess the temperament to educate their children regardless of ethnicity, gender, or race?

This reality came to the forefront during the pandemic when parents demanded more input into the curriculum content being taught. Most parents objected to the focus on critical race theory; diversity, equity, and inclusion; and transgender ideologies to the detriment of civics and other rigorous core courses. They also opposed the promotion of Marxist theories in the classroom.

A participant holds up a sign during a rally against critical race theory being taught in schools, at the Loudoun County Government center in Leesburg, Va., on June 12, 2021. (Andrew Caballero-Reynolds/AFP via Getty Images)

Some parents questioned the over-emphasis on district, site, and state tests that align with the factory model of education. Despite this emphasis, math and reading test scores are in the cellar in this state. Indeed, many families have left the public schools and opted for charter, home, or private schools due to the indoctrination churned out in many public schools.

As an educator and mentor, I’ve witnessed the gradual watering down of California’s public education system. The results have been troubling. Far too many upper-course university students have the writing skills of students at the junior high or early high school level. A fixation on electronic devices is partially to blame for this decline. Students aren’t reading enough to curate knowledge in order to develop writing skills in a variety of genres.

Students would benefit if the state and university schools of education placed greater focus on intellectual diversity and an emphasis on character and skills. Martin Luther King, Jr., wisely noted that folks ought to be assessed on the content of their character rather than the color of their skin. In a nation where equality of opportunity is systemic, anyone can select their own career and be successful without bureaucratic tinkering in the local schools. Experience, guidance, research, and trial and error can assist individuals in their quest to forge their own occupational destinies.

Utilizing the equity of outcomes approach punishes those who earn rewards through hard work and study habits. Moreover, it’s a disservice to minorities who want to compete on merit and don’t wish to be handed special favors based on arbitrary standards.

If education officials would pivot from an obsession with color, gender, and race to an emphasis on character, merit, and rigor, the public schools could be transformed into wholesome environments for all students. People wouldn’t care what teachers look like as long as they possess a passion to educate students as individuals, regardless of their backgrounds.

Hopefully, teachers would recognize that all human beings are “people of color” so that this meaningless phrase could be relegated to the ash heap. They might also become aware that everyone is equal in the eyes of the Divine Creator.

https://www.zerohedge.com/political/californias-education-priorities-are-focused-equity-quotas