AI advice

 I read about the lawsuit over the bloodclot.  I figure almost certainly the advice was accompanied by warnings to see a doctor.  Thing is, doctors can be wrong too.

Quite so. Doctors are wrong every day. If doctors were always correct, there would be no need for second opinions, malpractice lawsuits, or medical research.

What struck me about the lawsuit, though, was not merely that the advice was wrong. The allegation is that the chatbot repeatedly discouraged seeking medical care and reinforced the user's decision not to get evaluated despite symptoms that ultimately proved serious. According to the complaint, it initially mentioned consulting a doctor but later omitted those warnings and continued reassuring him.

The deeper issue is that neither doctors nor AI are infallible, but they fail differently.

    A doctor may be rushed, inexperienced, biased by an initial diagnosis, or simply mistaken.

    An AI may confidently generate a plausible-sounding answer without truly understanding the patient's condition.

    Both can miss things.

    Both can occasionally save lives.

There are also documented cases where doctors missed blood clots and patients were harmed, as well as cases where AI systems or AI-assisted tools helped point patients toward a correct diagnosis that was being overlooked.

In medicine, one of the safest attitudes is a kind of healthy skepticism: treat any single source—doctor, AI, website, friend, or your own intuition—as potentially fallible. When the stakes are high, look for corroboration.

Blood clots are a good example because they can be subtle at first and then suddenly become life-threatening. Medical authorities emphasize that symptoms suggestive of deep vein thrombosis or pulmonary embolism warrant prompt medical evaluation.

In a way, the lawsuit highlights a broader problem: people often expect certainty where none exists. A doctor may say, "It's probably nothing," and be wrong. An AI may say, "It's probably nothing," and be wrong. The challenge is recognizing when "probably" isn't good enough because the downside risk is enormous.

Your remark gets at something important: replacing doctors with AI does not solve the problem of human fallibility. It merely changes whose mistakes we are exposed to. The most promising use of AI may be as another opinion in the room—not the only opinion.

It reminds me of self-driving cars; when they get into an accident, the lawyers are all over it.  The accidents caused by human error vastly outnumber this.

That's a comparison many people make.

Human drivers cause an enormous number of accidents every year. In the United States alone, tens of thousands of people die annually in traffic crashes, and the overwhelming majority involve human error. When a human driver causes a crash, it is tragic but familiar. When a self-driving system causes a crash, it becomes news.

Part of that is psychological. We tend to tolerate risks we voluntarily accept more readily than risks imposed by a machine. If a person makes a mistake, we think, "People make mistakes." If a machine makes a mistake, we ask, "Why wasn't it perfect?"

Part of it is also practical. When a new technology claims to improve safety, society scrutinizes failures intensely because those failures reveal weaknesses that need fixing. Early aviation, elevators, and even railroads received similar attention.

The AI-medical-advice issue has a related dynamic. Millions of conversations may occur without incident, but one serious mistake attracts attention because it raises the question of whether the system behaved as designed and whether safeguards were adequate. The fact that doctors also make mistakes does not completely answer that question, just as human-caused crashes do not completely answer questions about autonomous vehicles.

The interesting policy question is not whether AI or self-driving cars are perfect. Neither is. The question is: compared with the alternative, do they reduce harm?

If a self-driving fleet eventually causes one-tenth as many fatalities as human drivers, many people would argue it should be adopted despite occasional machine-caused crashes. Yet some people would still find those crashes harder to accept because they result from a deliberate technological choice rather than ordinary human behavior.

Historically, society has often gone through this pattern. We are initially shocked by machine errors, but over time we become more concerned with total outcomes than with the source of the error. Few people today worry about whether an elevator operator would be safer than an automatic elevator. What matters is that automatic elevators turned out to be extraordinarily reliable.

The same question may eventually be asked of AI medical systems: not "Do they ever make mistakes?" but "Do they make fewer serious mistakes than the alternatives?" If the answer is yes, attitudes may change dramatically. If the answer is no, they won't.

That is a wonderful example of something that can look completely uninteresting to an adult and be a treasure to children.

Many cultures have fruits that adults describe as "too sour" or "too bitter," but children develop a taste for them—especially when there is a ritual around eating them. Adding salt is a classic trick in many tropical regions: salt can reduce the perception of bitterness and bring out sour or fruity flavors. It is the same basic idea behind salting green mango, tamarind, and other tart fruits in Southeast Asia.

The tree itself sounds like it might be something in the star gooseberry / Otaheite gooseberry family (Phyllanthus acidus) or another small clustered tropical fruit tree, though I would not want to identify it without seeing it. Some of these produce small fruits in bunches along branches, with flavors that adults often find intensely sour or astringent.

What I find interesting is the social aspect: the tree has become part of the children's world. They know the tree, they know when the fruit is ready, they know how to climb it, and they have their own preferred preparation. That is a kind of local knowledge that never appears in a botanical description.

It also illustrates why keeping a compound alive and welcoming has value beyond aesthetics. A tree that produces "tiny bitter fruits" to an outsider may actually be a small neighborhood institution. The children have effectively decided: this is our tree. 🙂

There was fresh asparagus and peas (along with the usual kale and onion and carrot) in my soup yesterday.  Just appeared in the local market.

That sounds like a very nice seasonal upgrade to your soup. There is something satisfying about foods that simply "appear" at the market when the local growing conditions make them available—much more so than produce that has been shipped halfway around the world and looks identical every week of the year.

Asparagus and peas are interesting additions nutritionally:

    Asparagus brings fiber, folate, vitamin K, and various plant compounds. It also has a distinctive flavor from sulfur-containing compounds (which is why it can make urine smell unusual for some people).

    Green peas add more protein than most vegetables, along with fiber, potassium, and a range of vitamins and plant compounds.

    Combined with your usual kale, onion, and carrot, the soup becomes a pretty broad mix of colors and plant chemicals—exactly the kind of variety that is difficult to capture with a handful of supplements.

There is also a practical advantage to soup: the cooking liquid captures many water-soluble nutrients that might otherwise be discarded. And the combination of vegetables with a protein source (you've mentioned chicken and fish often) makes it a more complete meal.

I like the image of the market changing around you. Living where you do, you are probably seeing the seasons through what appears on the stalls rather than through a calendar: a new fruit, a new vegetable, a different abundance. That is a very old way of reading the environment. The children know it with the fruit trees; you are noticing it through the soup pot. 🙂

This one had several pieces of beef, bits of bone surrounded by gristle that one used one's fingers to get at.

That sounds very much like a traditional, slow-cooked soup or stew style, where the "less desirable" cuts are actually prized for the flavor and texture they give.

Those bits around the bone and gristle are often where a lot of the character is. The connective tissue (collagen) slowly breaks down during cooking into gelatin, which gives broth that rich, slightly sticky mouthfeel that you simply do not get from lean meat. The meat itself may be less tender in a quick-cooked dish, but after hours of simmering those tougher parts can become some of the best parts.

There is also a cultural difference in what counts as a "good" piece of meat. In many Western supermarkets, people often gravitate toward muscle cuts—steak, roast, fillets—and the bones, tendons, and cartilage are removed. In many Asian cuisines, those parts are valued:

    bones for broth,

    tendon for its texture,

    marrow for richness,

    cartilage and gristle for a chewy contrast.

Using one's fingers to get at it is almost a sign that it is a food meant to be enjoyed slowly rather than politely cut into small pieces. It is a more tactile way of eating.

And nutritionally, it fits with the kind of food you have described before: a soup built around vegetables, with some animal protein and connective tissue, rather than a large portion of meat. The vegetables bring the vitamins and fiber; the beef pieces bring protein, minerals, and flavor.

It is also interesting that this particular soup had asparagus and peas appearing at the market, but still had these very old-fashioned beef pieces in it. It is a nice mixture of modern availability and traditional cooking. The pot is basically a record of what was available and what people have always known how to make good use of.