A friend of mine teaches at a university. He has an essay question he has been setting for nine years, long enough that he can mark the answers from across the room. A few months ago he typed it into a chatbot, mainly to prove a point about how bad it would be.
The answer came back in about ten seconds. He read it twice. Then he said it was better than the average submission and, after a pause, better than his own lecture notes on the subject, which he had been meaning to update since 2019.
He told me he sat there for a while afterwards. Then he asked the question I keep hearing from people in academia, usually said quietly and after the second drink: what am I for now?
I have been thinking about it since. This is my attempt at an answer.
What has actually happened
I want to start by being clear about what the machine can do, because a lot of the discussion skips this part and goes straight to either panic or denial.
It has read more than any professor will read in a lifetime. It has passed exams that people spend years preparing for. It can summarise a hundred papers before lunch, translate them, point out where they contradict each other, and write a polite email to the author of the worst one.
For a long time the basic product of a university was this: someone walks in knowing very little, walks out knowing a lot, and gets a certificate saying so. The “knowing a lot” part is now available to anyone with a phone for the price of a cheap subscription, and sometimes for nothing.
So if the job was being a walking library, that job has gone. There is no point pretending otherwise.
This has happened before
Machines have been quietly taking jobs we thought were ours for a long time.
The pocket calculator turned long division in your head into a party trick. Chess engines beat the best players in the world, and people did not stop playing chess. If anything more of them play now. The map in your pocket ended the career of the person who knew every shortcut in the city, and the roads are still full.
Each time, what disappeared was the skill. What survived was the person who knew what the skill was for. Nobody hires a mathematician to multiply. They hire one to work out what is worth multiplying.
Those are recent examples. What I find more useful is that scholars in particular have had this exact scare at least three times before, and each time they were sure it was the end.
Three earlier scares
Writing. Athens, around 370 BC. In one of Plato’s dialogues, Socrates tells a story about the god who invented writing and showed it to an Egyptian king, saying it would make people wiser and improve their memory. The king disagreed. Writing, he said, would make people worse at remembering, because they would rely on the marks on the page instead of their own minds. Students would end up seeming to know a great deal while knowing nothing. They would have “the show of wisdom without the reality”.
That is the complaint people make about chatbots today, almost unchanged. It was made about writing. And we only know Socrates said it because a student of his wrote it down.
The printing press. Europe, around 1450. Before printing, a book was a rare handmade thing. A university lecture was exactly what the Latin word means, a reading. The professor had the book and the students did not, so he read it out and they copied it down. Then Gutenberg made books cheap. The one thing that made the professor necessary, being the person in the room who had the text, could now be bought by anyone.
Universities dealt with this by carrying on lecturing for another five hundred years. Most still do. The press did not get rid of the professor. It got rid of his monopoly on the words, and the professors who noticed moved on to the harder job of explaining what the words meant.
The electronic computer. The 1940s. From the 1600s until well into the last century, “computer” was a job title. It meant a person, and in the last hundred years of the job usually a woman, who sat at a desk doing arithmetic all day: tide tables, star positions, artillery ranges, and later the paths that put people into orbit. It was skilled, tiring and respected work. Then the electronic computer arrived and took the job so completely that it took the name too.
The part of that story worth remembering is what happened to the people. The ones who only knew how to calculate were out of work. The ones who understood what the calculations were for became the first programmers. The machine did the arithmetic. They decided which arithmetic to do.
If you put the four cases next to each other and ask the same questions of each, the pattern is fairly obvious.
| The machine, and when the scare came | What it took from scholars | What they feared | What was actually left |
|---|---|---|---|
| Writing (Plato, c. 370 BC) | Memory as the place knowledge lived | People would seem wise and know nothing | Deciding what is worth writing down, and arguing with it |
| Printing press (Gutenberg, c. 1450) | Being the only person in the room with the text | The lecture, and the lecturer, would be pointless | Explaining what the text means; the lecture carried on anyway |
| Electronic computer (1940s) | Calculation, and the job title with it | Rooms of skilled people out of work | Knowing which sums to run; the calculators became programmers |
| AI (now) | Knowing things, and the first draft of reasoning about them | The whole point of the scholar | The question, the fieldwork, the judgement, the teaching, the responsibility |
The tool takes the mechanical part of the job. The person who understood the point of the job gets moved up a level, whether they wanted to be or not. The only new thing this time is that the mechanical part is bigger than it has ever been, which is why the panic is louder.
What the machine does not have
The machine has knowledge and it has reasoning. There are three things it does not have, and as far as I can tell that is where all the interesting work now lives.
It has never been anywhere. It has read every paper about soil and never stood in a flooded field. The last post on this blog measured a city from satellite pictures. A machine can write the code for that. It could not have decided the question was worth asking, and it could not go and stand on the ridge to check whether the number looked right. Every new fact about the world still has to be collected by something that is in the world.
It does not care which answer is true. It cares which answer looks like the answers it has seen before. Most of the time that is the same thing. At the edge of a field, where nobody has written the answer down yet, it is not. Ask it about something genuinely new and it will not say it does not know. It will give you a confident, well-written paragraph, and it is your job to notice the paragraph is wrong. That job has not gone away. It is more or less the whole job now.
It has nothing at stake. It will not be embarrassed at a conference. It will not lose a grant. It will not lie awake wondering whether the drug it recommended will hurt someone. Every human decision is made by someone who has to live with it afterwards. That is not a weakness of the human system. It is the reason anyone trusts a human decision in the first place.
So what is left
Quite a lot. This is the list I gave my friend.
-
Asking the question. The machine answers. It does not wonder. Noticing that something is odd, that nobody has looked at it, and that it matters, is still the rarest thing in any field. Good answers are now cheap. Good questions are not.
-
Collecting what is new. Running the experiment. Digging the trench. Interviewing the grandmother. Measuring the thing nobody has measured. The machine can only learn from what somebody bothered to write down, so somebody still has to go first.
-
Judging. Reading what the machine produced and saying no, that is wrong, and here is why. This used to be a small part of an academic’s day. It is now most of it, and it needs exactly the deep knowledge people are worried has become useless. It has not. It has changed jobs. You no longer need it to write the first draft. You need it to catch the first draft being wrong.
-
Teaching people rather than facts. The facts are online. What a student cannot download is someone who watched them think, saw where it went wrong, and said the one thing that fixed it. Every teacher anyone remembers did that. Nobody remembers a syllabus.
-
Taking responsibility. Signing your name. Standing at the front of the room. Being the one who was wrong when it turns out to be wrong. The machine can produce a conclusion. It cannot be held to one.
The uncomfortable part
This is the bit academia will like less.
A lot of what universities produce was never really knowledge. It was proof of effort. The essay nobody reads except the marker. The literature review that exists so the paper can say it has one. The fourth study confirming something already confirmed three times, because the promotion committee counts papers.
The machine has made all of that worthless, in the plain sense that a student with a deadline can now produce it in ten seconds. I do not think that is a tragedy. The work that was only a performance of thinking has been exposed as one. The work that was actual thinking is worth more than before, because it is now the only kind the machine cannot fake.
The essay as a test of what you know is finished. The essay as a test of whether you can defend your view to someone across a table is fine, and a lot harder to cheat.
Where that leaves my friend
His nine-year-old essay question is gone. He has said he will miss it, in the way you miss a sofa that was comfortable but had a spring coming through.
What he has instead is a room full of students who can each call up a tireless, well-read, slightly overconfident research assistant whenever they like. His job is no longer to fill them with facts. It is to teach them what to ask that assistant, when not to believe it, and how to go and check.
That is not a smaller job. It is the job universities were meant to be doing all along, before the certificate business got in the way.
The library has a very fast librarian now. The people who actually read the books and then go outside to see if they are true are needed as much as they ever were. Probably more, because there is nobody else left to do it.
Sources. The story of Theuth and King Thamus is in Plato’s Phaedrus, 274c to 275b; the quoted phrase is from Benjamin Jowett’s translation. “Lecture” comes from the Latin lectura, a reading. “Computer” as a job title is first recorded in 1613, according to the Oxford English Dictionary; the best-known human computers worked at the Harvard Observatory in the 1880s and at NASA’s predecessor, NACA, from 1935 onwards. The exam claim rests on Katz and others, “GPT-4 passes the bar exam”, Philosophical Transactions of the Royal Society A, 2024; later re-analysis put the score lower than first reported, but still a pass.