> ## Content Index > Fetch the complete content index at: https://letters.ovnnexus.com/llms.txt > Use this file to discover other available public pages before exploring further. # How to stay human, and the professional, in an emerging world of AI - URL: https://letters.ovnnexus.com/how-to-stay-human-and-the-professional-in-an-emerging-world-of-ai/ - Published: 2026-10-19T20:54:41.000Z - Updated: 2026-10-19T21:16:50.000Z - Author: Thad Connelly - Tags: The Practice Ledger ### Socrates had a problem with writing In Plato's *Phaedrus*, Socrates tells a story about the Egyptian god Theuth, the inventor of writing. Theuth takes his invention to King Thamus and presents it as a recipe for memory and wisdom. Thamus is not impressed. He tells Theuth that writing will do the opposite of what its inventor claims. People who trust written characters will stop exercising their own memory. They will read a great deal and learn nothing, because the words sit outside them instead of being understood from within. They will have the appearance of wisdom, Thamus says, while for the most part they know nothing at all. Socrates agreed with the king. In the dialogue he argues that a written text cannot answer back. You ask it a question and it repeats the same words forever. It cannot defend itself, it cannot tell whether its reader understands it, and it drifts into the hands of people it was never meant for. Real knowledge, in Socrates' account, is a living conversation between people who can be questioned and held to account. The irony is sitting right there. Socrates never wrote a book. Everything we know about his objection to writing survives because Plato wrote it down. The tool he distrusted carried his argument across twenty-four centuries and delivered it to us intact, which is more than his living voice managed on its own. That pattern, a serious person raising a serious objection to a new tool and the tool being adopted anyway, has repeated so often since then that we should look at a few of the repeats before we talk about the one in front of us now. ### The tool gets dismissed first When printing spread through Europe in the late fifteenth century, the men who copied books by hand saw it clearly for what it was, a threat to their livelihood, and many churchmen saw a second problem. A text that could be copied a thousand times could not be controlled the way a manuscript in a library could. Critics complained that printed books were shoddy, error-ridden, and vulgar compared with the real thing. Within two generations the printed book was the real thing, and the scribes were gone. Photography got the same reception from painters. Charles Baudelaire, writing about the Paris Salon of 1859, called photography the refuge of failed painters and warned that it would corrupt art. Working painters had plainer reasons to dislike it. A machine was taking the portrait business. Photography spent decades classified as a mechanical trade before museums began hanging it on walls. The telephone was dismissed as a novelty with no practical future. Western Union, the dominant communications company of its day, declined to buy Alexander Graham Bell's telephone patent in 1876, judging the device a toy with too many shortcomings to matter. The company changed its position quickly once customers started asking for phones, and spent years and a great deal of money building a competing network it could have owned outright at the start. Calculators were kept out of classrooms through much of the 1970s because teachers were sure, and said so in print and at school board meetings, that children allowed to use them would never learn arithmetic. The fear was reasonable on its face and it was wrong in its prediction. Calculators moved into classrooms, arithmetic instruction continued, and by the late 1980s national teaching standards were recommending their use. Those of us in clinical practice have our own versions. Digital radiography was going to be too crude to diagnose from, and the sensors were too expensive and too awkward. CBCT was going to be a toy for academics. Online scheduling and patient reviews were going to turn medicine into retail. Some of the objections in each case were legitimate and a few of the warnings were partly right. The tools were adopted anyway, because they worked, and the professionals who fared best with them were the ones who learned what the tools were actually good at and kept their own hands on the parts that mattered. AI is getting the full treatment now, the dismissal and the hype at the same time, sometimes from the same person in the same week. Before deciding what to make of either reaction, start with what the thing is. ### What AI actually is The systems in the news, and the ones creeping into our inboxes and operatories, are statistical models. They were trained on enormous amounts of material produced by people: text, images, records, code. During training the model adjusts its internal settings, billions of them, until its guesses about what comes next in a sequence match the material it was shown. Give it an input and it produces the output its training makes most probable. A few consequences follow from that. It produces fluent work, including fluent work that is wrong. Because the machinery is built to produce plausible output, an error arrives in the same confident tone as a correct answer. There is no internal flag that separates the two. It has read more than any of us and understood it in a narrower sense than any of us. It holds patterns from millions of documents. It has never treated a patient, never been responsible for an outcome, and never had to answer for a mistake in a room full of people who trusted it. It has no stake. A professional's judgment is shaped by consequences. The model has none. It will generate a referral letter, a treatment rationale, or a confident summary of a chart with the same indifference, and the responsibility for every one of those documents stays exactly where it was before the tool existed, with the licensed human whose name goes on it. Used with that understanding, it is useful. It drafts, summarizes, searches, formats, and does the first pass on work that used to eat the hours at the end of a clinical day. Used without that understanding, it produces polished errors at a speed no tired resident could match. ### Where we are now AI is already in the practice, whether we invited it or not. It filters the inbox and drafts the replies. It writes the meeting summary and the first version of the clinical note. It reads images as a second set of eyes in some operatories and answers the phone in others. Patients are using it too. They arrive having asked a model about their symptoms, sometimes better informed, sometimes frightened by a confident paragraph about the worst possibility on the list. The quality is uneven and it changes month to month. The tools are the most capable most of us have ever been handed. The results they produce range from excellent to fabricated, and telling the difference is a skill, and like every skill it is strong in people who practice it and weak in people who hand it over. That last part is the real variable of the moment. The technology is arriving on its own schedule. The discipline of the people using it is arriving on ours, and the gap between those two schedules is where most of the trouble and most of the advantage will come from. ### Six months, six years, six decades **Six months** is the easy horizon to talk about because most of it is already visible. Expect the tools to be embedded in more of the software we already use rather than arriving as separate products: the note that drafts itself from the visit recording, the prior authorization assembled before anyone asks for it, the imaging software flagging what it wants a second look at. Expect the novelty to wear off and habits to form, good and bad ones. The practices that set simple rules now, what gets checked, what never leaves the building without a human reading it, will be ahead of the ones still arguing about whether to allow it at all. That is an expectation, based on the direction the vendors and the early adopters are already moving. **Six years** is extrapolation and should be labeled as such. Training will change, because students will arrive having studied with these tools and residencies will have to decide what competence means when the first draft of everything is free. The economics of a practice will shift toward the work the models cannot do: the examination, the procedure, the conversation, the judgment call in a case that does not match the pattern. Hiring will follow that shift. So will liability, slowly, through cases none of us can predict yet. The professionals who come out of that period strongest will be the ones who treated the tool as equipment to be mastered, the way digital imaging had to be mastered, and kept their own clinical reasoning in daily use. **Six decades** is speculation, and I will leave it labeled that way because anyone who tells you they know is selling something. Sixty years ago a surgeon could not have described a modern operatory. What can be said is the long pattern from the examples above. Every tool that survived changed the work around it and left the responsibility in the same place. Writing did not end memory; it changed what memory was for. Print did not end scholarship; it changed who got to do it. If the pattern holds, the professions sixty years from now will be built around judgment exercised by people who can be questioned and held to account, because that is the part no tool in the sequence has ever taken over. Whether the people holding those jobs are as capable as the ones holding them now depends on choices being made in ordinary practices in the next few years. It depends on us. --- It is the most powerful tool we have ever invented. But if we are lazy and undisciplined, it can end us up in the world of just enough. We are more than the human in the loop. We are the creators drawing the loop.