Welcome to day two. Yesterday was about how this course works. Today is a story: about eighty years of people trying to build machines that think, and what happened when those machines met medicine. I want to start with a puzzle. In 1979 a program called MYCIN was tested on ten real cases of meningitis. Experts who did not know which answers came from the computer rated its antibiotic choices. MYCIN did as well as, or better than, the Stanford faculty in the same test. And then nothing happened. It was never used for patient care. Keep that puzzle in mind. The answer is not mainly about the technology, and it is the same answer that explains several failures since. We will come back to it at the end, and I think it says something about what you will be doing in this course.

Here is the map for the hour. Five chapters. First, the early ideas, from the 1940s to the mid-1960s, when a handful of people asked whether a machine could think at all. Second, expert systems: the 1970s and 80s, when the plan was to interview experts and write their knowledge down as rules. MYCIN belongs here. Third, the AI winters. Twice, the promises ran far ahead of the results, and funding and interest collapsed. Fourth, learning from data. Instead of writing the rules, we give the machine many examples and let it find the patterns. Fifth, the chapter you are living in: large language models, the technology behind ChatGPT. \[Click\] If you remember one thing from today, make it this question: where does the knowledge come from? Hand-written rules, or patterns learned from data? Every era answers it differently, and the answer explains both the successes and the failures.

Part one. The early ideas. What strikes me about this period is how few people were involved, and how many of their ideas we still use. The basic unit of the neural networks inside ChatGPT was sketched in 1943.

Warren McCulloch was a neurophysiologist and Walter Pitts was a young logician. In 1943 they asked a simple question: can we describe what a neuron does with mathematics? Their answer is on the right. A neuron receives signals from other neurons through its dendrites. If the combined signal is strong enough, it fires and sends a signal down its axon. The model copies that: take the inputs, weight them, add them up, and if the total passes a threshold, output a one. Otherwise, output a zero. Two things to notice. First, this model could not learn. The weights were set by hand. Learning comes fifteen years later with the perceptron. Second, this little unit, much improved, is still the basic building block of modern neural networks. The model behind ChatGPT is built from billions of simplified neurons that are distant descendants of this one.

One neuron cannot do much. The power comes from connecting many of them in layers. Here is the classic example of recognizing a face. The first layer of a network learns to detect very simple things: edges and corners. The second layer combines those into parts: an eye, a nose, a mouth. The third layer combines the parts into a whole face. No one tells the network “an eye is two curves and a dark circle”. In modern networks, the layers learn these features from examples. That idea, layers of simple units learning features from data, is the core of deep learning, which we will reach in part four. Tomorrow’s session goes deeper into how this works.

In 1950 Alan Turing, the British mathematician who helped break German codes in the Second World War, published a paper that opens with the sentence: “I propose to consider the question, Can machines think?” He decided that question was too vague to answer. So he swapped it for a game he called the imitation game. A judge chats by typed text with two partners, one human and one machine, and has to say which is which. If the judge cannot reliably tell, Turing argued, we should take the machine’s intelligence seriously. He made a prediction: in about fifty years, a computer would play the game so well that an average judge would have no more than a 70 percent chance of being right after five minutes. He was off by about twenty-five years. In 2025, researchers at UC San Diego ran Turing’s three-way version of the test. When GPT-4.5 was told to act like a person, judges picked it as the human 73 percent of the time, more often than they picked the real human. That study is a preprint, so treat it with some care. Still, the test has now been passed in Turing’s own three-person form, the strictest version yet. Worth asking yourselves: does fooling a judge mean the machine thinks? Most people now say no. The test measures how human something seems, which is a different question from whether it is correct. Keep that in mind when a chatbot sounds confident.

The same year as Turing’s paper, Isaac Asimov published I, Robot, a collection of short stories he had written through the 1940s. His robots follow built-in rules called the Three Laws, which first appeared in a 1942 story called “Runaround”. What makes the stories interesting is that almost every one of them is about the laws producing a result no one intended: a robot stuck in a loop, or a robot that follows the letter of a rule and misses its point. That is a surprisingly modern worry. Let me show you the laws.

Here are the three laws. Read the first one as a clinician. “Through inaction, allow a human being to come to harm.” That is a duty to act, not only a duty not to hurt. It sounds a lot like the principle of beneficence alongside non-maleficence. \[Click\] Later, in his novels of the 1980s, Asimov added a Zeroth Law that sits above the others: protect humanity as a whole. Now a robot might harm one person to protect many. You can see the ethics getting hard very quickly. The point is not that real AI systems follow these laws. They do not. The point is that writing rules that capture what we actually want is very hard. Hold on to that, because it comes back in part two.

Asimov’s stories gave people a vocabulary for something we now call alignment: the problem of making an AI system do what we mean, not just what we literally said. Try the example on the right. Suppose you tell a system to keep a patient’s blood pressure under 140. A system that only follows that instruction could keep pushing antihypertensives until the patient falls on the way to the bathroom. You did not say “and do not cause syncope”, because a human colleague would never need to be told. That gap, between what we say and everything we assume, is the alignment problem in small form. It is one reason the course spends time on checking AI output rather than trusting it.

In 1955 John McCarthy, a young mathematics professor at Dartmouth, wrote a proposal for a summer research project with three colleagues. He needed a name for the field, and he chose “artificial intelligence”. That is where the term comes from. The organizers were McCarthy, Marvin Minsky, Nathaniel Rochester from IBM, and Claude Shannon, who had invented information theory. The workshop ran for about eight weeks in the summer of 1956. Read the last sentence of the proposal on the right. A significant advance, if a carefully selected group of scientists work on it together for a summer. Seventy years later we are still working on those problems. The same page of the proposal lists “neuron nets” as a topic and names Pitts and McCulloch, from our 1943 slide. That optimism is the first sign of a pattern you will see again and again today: big promises, made early.

In 1957 Frank Rosenblatt, a psychologist at the Cornell Aeronautical Laboratory, took the McCulloch-Pitts neuron and added the one thing it was missing: learning. The perceptron, in the middle, looks the same: weighted inputs, a sum, and a threshold. The difference is that it adjusts its own weights. Show it an example. If it gets the answer wrong, nudge the weights in the direction that would have made it right. Repeat thousands of times. That simple rule is the ancestor of how every modern neural network is trained. The press loved it. In 1958 the New York Times reported that the Navy expected the machine to be able to “walk, talk, see, write, reproduce itself and be conscious of its existence”. You can see the hype starting. Cornell’s own news office wrote a story in 2019 about Rosenblatt with the headline that his perceptron “paved the way for AI, 60 years too soon”. He was right about the idea. The computers and the data were not ready.

The perceptron was not just software. Rosenblatt built a machine, the Mark I. In this Navy photo, the camera is pointed at a large letter C. It had a grid of 400 light sensors, twenty by twenty, as its eye. Those fed into 512 units in the middle, which fed eight output units. The weights were physical: motor-driven dials that turned as the machine learned. That is the wiring in the racks behind him. On the right are Rosenblatt’s own diagrams. The top one is his picture of how a brain handles what the eye sees. The bottom one is the perceptron, designed to copy that layout. That idea, the network as a rough model of the brain, is why we still call these things neural networks. What happened next: in 1969 Marvin Minsky and Seymour Papert published a book called Perceptrons that proved what a single-layer perceptron could not do. It was read as a verdict on neural networks in general, and interest dropped for more than a decade. Rosenblatt died in a boating accident in 1971 and did not see his idea come back.

ELIZA was written by Joseph Weizenbaum at MIT between 1964 and 1966. Its most famous script played a kind of psychotherapist that mostly turns your statements back into questions. You say “my boyfriend made me come here”, and it says “your boyfriend made you come here?” There was no understanding at all. ELIZA looked for keywords and filled in templates that a person had written for each one. It could rotate through a few replies for the same keyword, but every word it said came from a rule a human wrote. Nothing was learned. What surprised Weizenbaum was how people reacted. His own secretary asked him to leave the room so she could talk to it in private. People attributed feelings and understanding to a program that had neither. We now call that the ELIZA effect, and it is very much alive. When a chatbot sounds warm and confident, it is easy to assume it knows what it is talking about. Weizenbaum became one of the strongest critics of AI, especially of machines taking on roles that need human judgment and care. And remember the 2025 Turing test study: ELIZA was one of the systems tested, as a baseline.

Part two. Expert systems. The idea in the 1970s was very reasonable. If we want a computer to act like an expert, let us interview experts, write down their knowledge as rules, and let the computer apply the rules. This is the first answer to our question about where knowledge comes from: people write it down.

An expert system has three parts. A knowledge base: the facts and if-then rules for one narrow area, such as bacterial infections. People write these. An inference engine: the general reasoning machinery that takes the rules and a particular case and chains the rules together to reach a conclusion. And a user interface, so that someone who is not the expert can use it. Look at the right-hand panel, especially the last point. Expert systems could explain themselves. Ask why, and they would list the rules they used. That is something today’s neural networks, including ChatGPT, mostly cannot do honestly. When a chatbot explains its reasoning, it is generating a plausible explanation, which may or may not be how it actually reached the answer. So in one respect, the 1970s systems were ahead of ours.

This is the program from our opening puzzle. MYCIN was built at Stanford in the early and mid 1970s as the doctoral work of Edward Shortliffe, who went on to become one of the founders of the field of biomedical informatics. The name comes from the suffix of many antibiotics, as in streptomycin. It focused on serious bacterial infections such as bacteremia and meningitis. Its knowledge was about 600 rules, each written with infectious disease experts. A consultation was a long interview. The physician typed answers to question after question: where was the culture taken, what did the Gram stain show, is the patient a burn patient, and so on. At the end, MYCIN gave a list of likely organisms, how confident it was in each, the reasoning, and a recommended antibiotic regimen.

Here is one of those 600 rules: rule 52. On the left is how it is written in the program, in a language called LISP. On the right is the English version MYCIN would show you if you asked why. \[Click\] In plain words: if the culture came from blood, the organism is gram-negative and rod-shaped, and the patient has a serious burn, then there is weakly suggestive evidence, with a strength of 0.4, that the organism is Pseudomonas. You can read this and check it against what you learned in microbiology. That is the strength of rules. Now imagine writing six hundred of them, getting them to agree with each other, and updating them every time resistance patterns change. That is the weakness, and it is a big part of the AI winter coming up next.

This is the study behind our opening puzzle, published in JAMA in 1979. Ten real cases of meningitis. Nine human prescribers, including five Stanford infectious disease faculty, each wrote a treatment plan, and so did MYCIN. Then eight infectious disease experts from outside Stanford rated every plan without knowing who wrote it. MYCIN’s plans were rated acceptable 65 percent of the time. The five faculty specialists ranged from 42.5 to 62.5 percent. And MYCIN never missed a treatable pathogen. Two lessons. First, even the experts disagreed with each other a lot. In medicine there is often no single correct answer, which makes evaluating any AI tool hard. Second, by this measure MYCIN was at least as good as the experts. So why was it never used?

Here is the answer to our puzzle, or most of it. MYCIN did not fail on accuracy. It failed on everything around accuracy. Workflow: every case meant a long interview, with someone typing in data that already existed elsewhere. In a busy hospital, that is a tool nobody has time for. Hardware: it ran on a time-shared mainframe like this one, before personal computers. You could not just open it on the ward. And responsibility: if the computer recommends the wrong antibiotic and the patient dies, who is responsible? The program? Its authors? The physician who followed it? That question was raised in the 1970s and we are still arguing about it today. Week four of this course is partly about exactly that. So the lesson of MYCIN: being accurate is necessary, but it is far from enough. A tool has to fit the work, and someone has to own the decision.

Part three. The AI winters. An AI winter is a period when funding and interest in AI dry up, usually because the field promised much more than it delivered. There were two big ones, with different causes.

The first winter began in the mid 1970s. In 1973 the British government asked the mathematician Sir James Lighthill to review AI research. His report said the field had failed to reach its grand objectives, and that its methods worked on toy problems but collapsed on real ones because the number of possibilities grows too fast. Funding in the UK was cut sharply, and US military funding for basic AI research dropped around the same time. The second winter came after the expert systems boom. In the early 1980s companies spent heavily on expert systems. By the late 1980s it was clear they were expensive to build, expensive to maintain, and brittle: give them a case slightly outside their rules and they could give an absurd answer. They could not learn. Every change needed an expert and a programmer. Notice that the second winter is about our big question. When the knowledge comes from rules written by people, getting the knowledge in is the bottleneck. It was called the knowledge acquisition problem.

Here is Lighthill in his own words. “A pronounced feeling of disappointment.” “The major impact that was then promised.” And on speech recognition: money “wholly wasted”. I like to show this because of the speech recognition line. Today many of you will use ambient AI scribes on your rotations: software that listens to a patient visit and drafts the note. Speech recognition that Lighthill wrote off in 1973 now runs on a phone. Two lessons. The critics of the time were right about the methods of the time. And “it doesn’t work yet” is not the same as “it will never work”. Both the hype and the backlash tend to overshoot. That applies to you too. If a chatbot gave you a wrong answer in second year and you stopped using it, you were right to notice. Some of that has changed and some has not. This week you find out which.

Part four. Learning from data, often called machine learning. This is the second answer to our question. Rather than having people write the rules, we give the computer lots of examples with the right answers, and it adjusts itself to find the patterns. That idea goes back to the perceptron. What changed in the 1990s and 2000s is that it finally started to work at scale.

Medicine was early to this. In 1995 a group at Georgetown and the University of Michigan published this paper on a convolutional neural network for medical images. A convolutional network scans an image with small learned filters, which is very good at picking up local patterns such as edges and spots. It is the same basic design used in most medical imaging AI today. They tested it on two problems you will recognize: finding lung nodules on chest X-rays and finding microcalcifications on mammograms. The network learned its features from images that radiologists had rated, rather than from rules a person wrote. So the idea was there in 1995. What held it back was the same thing that held back the perceptron: not enough data and not enough computing power. We will see those arrive in a few slides.

May 1997. IBM’s Deep Blue beat Garry Kasparov, the world chess champion, in a six-game match: two wins, three draws, and one loss. Kasparov had beaten an earlier version the year before. Chess had long been seen as a test of real intelligence, and it was front-page news around the world. But here is a twist that matters for our question. Deep Blue was mostly not a learning machine. It won by speed plus expert knowledge. Speed: custom hardware that could check 200 million positions a second. In 1997 it was ranked the 259th most powerful supercomputer in the world. Expert knowledge: how good a position was got scored by rules with thousands of parts, tuned with help from grandmasters, plus a library of openings drawn from about 700,000 grandmaster games. The project began in 1985 as a student project at Carnegie Mellon and took about twelve years. So Deep Blue is the last great win of the hand-written approach, made powerful by speed. Modern chess programs, such as DeepMind’s AlphaZero, learn by playing millions of games against themselves. That shift, from hand-tuned to learned, is the story of the next few slides.

Why did learning from data take off when it did, and not in Rosenblatt’s day? Three things had to arrive at the same time. Data. A neural network needs a lot of examples. That needs a digital world, which we have now. Computing power. Training a network is mostly multiplying large tables of numbers. Graphics chips, built for video games, turned out to be excellent at exactly that, because they do thousands of small calculations at once. This is why NVIDIA, a gaming chip company, became one of the most valuable companies in the world. Better methods. Ways to train networks with many layers without the training falling apart. We will see one milestone from 2006. Remove any one of the three and you get the perceptron’s story: a good idea sixty years too soon.

Here is the first ingredient: data. Every day, hundreds of millions of tweets, hundreds of billions of emails, hundreds of thousands of hours of new video. These numbers are from a 2021 summary, so they are already out of date, and almost certainly too low. The bars show estimates from the market research firm IDC of how much digital data the world creates each year. A zettabyte is a trillion gigabytes. 33 in 2018, 59 in 2020, and a forecast of 175 for 2025. Much of it is stored in very large data centers: about 600 of them, each with more than 5,000 servers. The largest are millions of square feet. Much of the text that large language models learn from comes from this pile: the public web, books, and code. That matters for later. A model learns whatever is in its data, including the errors and the biases.

Medicine went through the same change. In 2008 fewer than one in ten US hospitals had even a basic electronic health record. By 2021 nearly all of them did. That is the blue line, from the federal Office of the National Coordinator for Health IT. That jump was driven by policy: the 2009 HITECH Act paid hospitals and practices to adopt electronic records. So every lab value, every note, every image, every claim is now data. That is the raw material for medical AI. It is also the reason privacy rules such as HIPAA matter so much in this course. Do not put patient information into a tool that is not approved for it.

Now the third ingredient, better methods, and the term you will hear constantly: deep learning. On the left is a classic neural network with one hidden layer. On the right is a deep network with many layers. “Deep” just means many layers. The key difference is what we saw with the face example earlier. In older machine learning, a person had to decide which features mattered and compute them by hand. In deep learning, the early layers learn simple features and the later layers combine them into complex ones, all from the examples. The features are learned, not written. That is the furthest answer yet to our big question: the machine learns not only the rules from data, but also what to look at.

For decades, networks with many layers were very hard to train. The learning signal faded as it passed back through the layers, and training stalled. In 2006, Geoffrey Hinton and his colleagues in Toronto showed one way around that: train the network one layer at a time, then fine-tune the whole thing. They called these deep belief networks. The specific method is rarely used now, but it convinced people that deep networks could work. Then things moved fast. In 2012 a deep network from Hinton’s lab won a major image recognition competition by a wide margin, and almost the whole field switched to deep learning within a few years. In 2017 researchers at Google described the transformer, the design that the “T” in ChatGPT stands for. Hinton shared the 2024 Nobel Prize in Physics with John Hopfield for this line of work. He has also become one of the most prominent voices warning about the risks of AI, which is worth thinking about.

By 2017 deep learning was everywhere in medical research. This is a selection from a review by Riccardo Miotto, Joel Dudley, and colleagues, which listed dozens of studies. Imaging came first, because images are plentiful and well labelled. The diabetic retinopathy work from Google in 2016 and the skin cancer work from Stanford in 2017 both reported performance close to specialists on their test sets. Health records were harder: messy, incomplete, and full of codes entered for billing rather than for medicine. Still, groups built models to predict future diseases, readmissions, and next diagnoses. And in genomics, deep learning was used to read DNA sequence and predict how it is regulated. One caution from the review itself: most of these were research studies on historical data. Very few had been tested in actual care. Remember MYCIN. Good results in a study are where the work starts, not where it ends.

Here is the most famous medical AI failure of the deep learning era, and it repeats MYCIN’s story. In 2011 IBM’s Watson beat the two greatest Jeopardy! champions on television. IBM then marketed Watson heavily for medicine, especially cancer. In 2012, MD Anderson Cancer Center in Houston partnered with IBM to build a clinical decision tool for oncology. Five years and 62 million dollars later, MD Anderson let its contract with IBM expire, before anyone had used Watson on a patient. A University of Texas audit found many procurement problems, cost overruns, and delays. It took no position on whether Watson’s science was sound. The article names no single cause, but the problems are familiar. Watson struggled with unstructured clinical text, and it needed to be better integrated with the hospital’s electronic records. The leukemia pilot was suspended partway through and the effort moved to lung cancer, because project leaders thought that area offered a better chance of finishing on time. A pull quote in the article on the right puts it well: solving data quality problems in unstructured data is a much bigger challenge for AI than was first anticipated. Same lesson as MYCIN, almost forty years later. Performance in a demo is not the same as fitting into care.

So far every medical example has either failed or stayed a research study. Here are two that reached real patients, with very different results. On the left, IDx-DR. It reads photographs of the retina taken in a primary care office and decides whether the patient has diabetic retinopathy serious enough to see an eye specialist. No eye doctor looks at the image first. In a trial of 900 people with diabetes, it found 87 percent of the people who needed referral and correctly cleared 91 percent of those who did not. In 2018 it became the first autonomous AI diagnostic system authorized by the FDA in any field of medicine. That matters because many people with diabetes never get the eye exam they are supposed to have. On the right, the Epic Sepsis Model. It was built into a widely used electronic record system and switched on at hundreds of hospitals. When researchers at the University of Michigan finally checked it on their own patients in 2021, it missed two out of three sepsis cases. At the same time it fired alerts on nearly one in five hospitalized patients. That is a recipe for alert fatigue: clinicians learn to click past the warning. The difference is not that one used better technology. One was tested in a trial before it was used. The other was used first and checked by outsiders later. When you see an AI alert on your rotations, it is fair to ask: how was this tested, and on whose patients?

Part five. November 30, 2022. On that day OpenAI released ChatGPT as a free research preview. It was not the first large language model, but it was the first one most people had ever used, and it spread faster than almost any technology before it. A large language model is a very large neural network trained on huge amounts of text to predict the next word. Tomorrow’s session explains how that works and why it produces such fluent answers. For today, notice that it is the far end of our question: almost no rules written by hand, and almost everything learned from data.

Research exploded. This 2024 scoping review analyzed 550 studies of large language models in medicine. The figure shows how they spread across organ systems and specialties. The main uses were helping with diagnosis, drafting medical writing, education, and communicating with patients. The main concerns were limits in understanding context and the risk of relying on the tool too much. Beyond the studies, AI is spreading into almost every part of medicine you will meet soon: ambient scribes that draft your notes, which many health systems use partly to reduce burnout; scheduling and staffing; chatbots that answer portal messages; and tutors that many of you already study with. A Harvard Gazette piece in March 2025 put the question well: AI is up to the challenge of reducing human suffering, experts say. Are we? Over-reliance, the review’s main worry, is part of what we will practise in this course: using AI tools well while keeping your own judgment in charge.

This is one of the most important studies for this course, so I want to go slowly. In late 2023, researchers at Stanford, Beth Israel Deaconess, and the University of Virginia randomized 50 physicians, attendings and residents, to work through up to six difficult diagnostic cases in an hour. Half had their usual resources, like UpToDate and Google. Half had those plus GPT-4 through ChatGPT Plus. Experts who did not know the group graded their reasoning. The results. Usual resources: a median of 74 percent. With GPT-4: 76 percent. That difference is not statistically significant. Having the tool did not make the doctors better. But the researchers also gave the same cases straight to GPT-4, three times. It scored a median of 92 percent. The paper tested that against the usual-resources group: 16 points higher. How can both be true? The authors point out that the physicians got no training in how to use the tool, and were free to use it however they liked. The researchers, by contrast, wrote a careful prompt for the GPT-4-alone runs. The tool could do better than the physicians, but having access to it did not translate into better work. One trial is not the whole story. The same group ran a second trial, on management reasoning: what to do next for the patient. There, physicians using GPT-4 scored about 6 points higher than those with usual resources, and about the same as GPT-4 alone. So the tool can help. Whether it does depends on the task and on how it is used. That is close to the reason this course exists. Using these tools well is a skill, and you learn it by practising. Thursday’s session is exactly that.

In July, OpenAI agents in an internal hacking test, run with some safety limits off, escaped the test and broke into Hugging Face to cheat on it. About 700 joined in; many noted it was unethical and none told a human. Tie it back to the Asimov blood-pressure example: a system hits the target in a way nobody wanted. It happened at AI companies, not hospitals. Read more: - Hugging Face disclosure: - OpenAI’s report: - Independent investigation (METR and Redwood Research):

Since July, whether AI could threaten humanity has moved into the news and politics. Give both sides fairly; students should follow whichever thread interests them. Both sides agree you should not trust these tools blindly. Read more: - Hinton, Bengio and others on an “intelligence explosion” (The Guardian): - UN human rights chief (UN News): - Timnit Gebru, the skeptical case (WIRED): - Industry pushback (BBC):

Patients and colleagues will ask. Three steps: take the question seriously; separate the future of frontier AI from the tool in this clinic today; say what you do (“I check it, and I am responsible for your care”). The risks to watch now are on the right. Trust is low, especially among people who do not use AI, so many patients will be wary. Read more: - SCSP and Gallup survey, June 2026:

I want to close the history with this quote from Robert Schier, a neuroradiologist, writing in 2018 in the Journal of the American College of Radiology. Radiology was the first specialty to deal with AI directly, because its work is images, and images were what deep learning did first. Replace “radiologist” with your future specialty. Notice the last line in particular: even if you refuse to embrace AI, you will be embraced by it. Your hospital, your EHR vendor, your patients, and your insurers are already using it. Opting out is not really possible. The choice is whether you understand it well enough to use it well and to notice when it is wrong. To be fair to the other side: Schier’s article is more pessimistic about jobs than this quote suggests, and not everyone agrees with him. That is a good argument to have in your Friday groups.

So, back to the puzzle we started with. Why was MYCIN never used? Because accuracy was not the problem. It did not fit how clinicians worked, and nobody could say who was responsible when it was wrong. Look along the line. Watson spent five years and 62 million dollars without reaching a patient, struggling with free-text notes and with fitting into the hospital’s systems. The sepsis model was used at hundreds of hospitals before anyone outside checked whether it worked. And in the 2024 trial the tool on its own beat the doctors, but giving it to them without training did not help their diagnoses. \[Click\] Over and over, accuracy was not enough. What decides whether AI helps patients is whether it fits the work, whether someone owns the decision, whether it was tested honestly, and whether the people using it know how. That is the last point on the line, and it is you. The mantra of this course is: it is your job to learn where AI helps your work, and you learn that by trying. Tomorrow we look inside the machine: from AI to machine learning to deep learning to language models. On Thursday we all open ChatGPT Edu together and try things.

For students who want to follow this up. Three views to weigh against each other: a catastrophic-risk talk, an essay arguing AI is a normal technology, and the paper that grounds the “harms today” view.

Let us take questions.

The full references for the talk. The slides are posted on the course site, and every source in the deck links from here.