Slow AI

The utopians tell you the machine will save us: cure the diseases, fix the climate, teach every child. The doomers tell you it will end us, and soon. Between them sit the companies, who benefit from both stories, because both stories agree that the technology is too vast and too inevitable for an ordinary person to judge. Your job, in every version, is to adapt. I do not accept that. AI is a technology, built by companies, trained on our words, sold back to us with a margin. It is a catalyst: it amplifies what is already there, good and bad, at a speed we have never seen. It can be judged, and you are qualified to judge it. This book is a set of tools for doing exactly that.

Speed is the mechanism of most AI harm. The answer that arrives in three seconds arrives before your judgement does. The draft that writes itself skips the thinking the draft was for. The feed that refreshes forever outruns your capacity to ask whether any of it was worth your evening. In every case, the harm is in what the speed bypasses. Slowness puts your judgement back in the loop. A pause before the prompt. A question after the answer. A check before you sign your name to something a machine produced.

When the tool tells you your idea is excellent, ask it for the strongest case against what you just said. The quality of your work will improve immediately, and you will feel the design resist you, which is its own education.

notice where your warmth is going. Every hour of emotional labour you spend on a system that cannot receive it is an hour that a human being in your life did not get. If this chapter moves you to write one kind message tonight, send it to a person, who can miss you back.

An early-learning specialist I have collaborated with has a lovely answer for those nights, and it is an AI prompt.1 Ask the machine for three sentences of bedtime story, and require it to stop before the resolution, so that parent and child have to finish the tale together. The machine supplies a seed: a bear who cannot wake the carousel horses, a bunny holding a mysterious golden ticket. The child does the inventing, which decades of research on dialogic reading says is where the developmental payoff lives. Let the tool complete the arc and it has stolen the best part of the workout. Make it stop early and it has started a conversation, and the conversation is the point.

What separates those applications from the wreckage of the last chapter, and from the harms still to come in the next three? Look back at the bedtime story and you can read off the whole pattern. Four properties, and they organise everything that follows. The task is bounded. Three sentences, stop before the resolution. Good AI use has edges. The catastrophes mentioned elsewhere begin where the task is open-ended: be my friend, do my thinking, fill my feed. A human with judgement stands downstream. The parent is in the room, steering; the child is inventing. Where the human downstream is attentive and qualified, machine error is caught and machine volume becomes a gift. The comparison class is honest. The alternative, at the end of that long day, was an exhausted parent with an empty tank. Judge any AI system against what is actually available in that place, at that moment. It is the standard I draw from Eric Topol’s landmark review of medical AI,2 and it generalises: the comparison class determines the ethics. ‘Compared to what?’ will sort the genuine cases from the hype for the rest of your life. The thinking stays yours. The machine supplies the seed. The people supply the story. Every good case leaves the humans more capable, or at least no less capable. The harmful ones leave the user smaller.

AI as a sparring partner. When you have drafted an argument, ask the machine for the strongest case against it. Ask it what a hostile reviewer notices, what you have assumed, whose perspective is missing. Gary Klein’s research on pre-mortems showed years before AI that structured imagining of failure measurably improves decisions;12 a language model makes the exercise available on demand, at midnight, before the meeting.

The four tests fit in your pocket. Bounded task. Human downstream. Honest comparison. The thinking stays yours.

So the real skill of working with AI is knowing where the frontier runs through your own job. Nobody can hand you that map. The frontier moves with every model release, and it is different for a lawyer, a nurse, and a marketing manager. The only reliable way to hold the map is to keep doing enough of the work yourself that you can tell, from the inside, when an answer smells wrong.

Here is the honest arithmetic that almost nobody performs. A well-structured email takes three minutes to write. Prompting a machine to write it, reading the result, noticing the tone is off, re-prompting, and editing the second attempt to sound like you can take ten. For some tasks AI is a genuine time machine. For others it adds a layer of mediation between you and your own thinking that costs more than it saves, and the only way to know which is which is to time it honestly, task by task.

Saved time was refilled immediately with more tasks, broader responsibilities, and longer hours, while the conversational interface blurred the line between working and resting. They warned that the downstream costs of this pattern are cognitive fatigue, burnout, and weakened decision-making, and they gave it the right name: intensification.5 Parkinson’s law, that work expands to fill the time available, has found its perfect accelerant. If AI frees seven hours of your week and you spend all seven on more work, the machine has made your employer more productive and you more tired. Using some of those hours to read a book, or call someone you love, is a decision, and nobody will make it for you.

Medical educators have started using the term ‘never-skilling’ for trainees who lean on AI so early and so completely that the underlying judgement never forms at all.11 Nothing faded. There was nothing to fade. Think of the junior analyst who has only ever produced a model by asking the machine for the formulas, or the new manager whose first difficult email was drafted by a chatbot before they had ever found their own words for a hard conversation. The scaffolding went up before anything was built behind it. If you manage juniors, this is now among the most serious questions on your desk: if AI does all the work your trainees used to learn on, where exactly do your seniors come from in ten years?

At this point in the argument, most advice collapses into willpower. Use AI less. Be more disciplined. Resist. You have probably already tried it, and it did not hold. The philosopher Avigail Ferdman explains why: she argues that AI is creating what she calls capacity-hostile environments: settings where the design of the tools actively works against the growth of human capacities.12 A capacity, in her terms, is something you build by exercising it, the way a muscle grows under load. A capacity-hostile environment removes the load, and frames the removal as a kindness. Look at how your tools are built. The button that finishes your sentence is bigger than the empty page. The summary arrives before you have read the document. ‘Improve this writing’ sits one click away at all times. The default, everywhere, is the answer instead of the method, and defaults are designed by teams whose success metric is your engagement. Your willpower is fighting a building full of professionals whose job is to make the effortless choice irresistible. You were always going to lose that fight on your own. The remarkable thing is that we keep telling people the fight is theirs alone to win.

Reverse the order. Do the thinking first, then let the AI react to it. Write the rough draft, then ask for critique. Make your diagnosis, then check it against the tool.

Demote the answer. Change your standing prompts so the tool gives you options, methods, and counter-arguments instead of finished products. A sparring partner builds your judgement with every exchange, and that is the role you are hiring for.

Take a task you consider ‘your’ work and walk through its steps: did I do the thinking, did we share it, or did I accept the machine’s output without engaging?

for the steps you cannot afford to lose, do the task unaided on a schedule: read the paper before you summarise it, draft the email before you ask for help, find the route without the map. Once a month keeps the skill warm, and it doubles as your early-warning system for the jagged frontier, because you can only smell a wrong answer in a domain you still practise.

For one week, time your AI-assisted tasks honestly, including the prompting, the reading, the checking, and the fixing. Keep the uses that genuinely pay. Drop the ones that only felt fast. Then decide, deliberately, where the recovered time goes, before your calendar decides for you. The tools will keep improving, and the better they get, the stronger every pull in this chapter becomes. The verification tax shrinks as accuracy improves, which makes trust easier, which makes checking rarer, which makes the errors that remain more dangerous. The frontier expands, which invites you further out onto it. This is why the practice matters more as the technology matures.

Speed does the first damage. When an answer arrives in three seconds, it arrives before your own thinking has started, and the thinking that would have happened in the gap simply never occurs. Whole categories of insight live in that gap. I have watched it in my own work: the question I would have refined, the connection I would have made on the walk to the kettle, pre-empted by a fluent paragraph that was good enough to end the search. Friction is where thought happens, and these systems are engineered to remove it.

So the discipline is one question, asked at the moment the tool offers to do the thing: does this need to exist? Answered honestly, it thins the pile instead of adding to it. One thing that needed making is worth more than ten that did not. Post when you have something to say, and let the feed go hungry when you do not.

When people complain about AI writing, they point at the surface. The balanced sentence. The rule of three. The word ‘delve’. None of those is the problem, and writers have used every one of those techniques for centuries because they land. The machine copies them for exactly the same reason. Slop lives underneath the technique. Slop is content with no agency, no depth, and no purpose. Nobody needed to say it. Nobody chose it. It was made because making it was possible, to fill a slot and feed a feed.

A 2026 paper by Kommers and colleagues, ‘Why Slop Matters’, gives the most useful anatomy of it I have read.4 Slop has three marks: Superficial competence. It looks right. The lighting is good, the sentences scan, the voice is confident. Underneath there is very little. Asymmetric effort. It took seconds to make and minutes to consume. The cost has moved off the maker and onto you. Mass producibility. There is never one of these. There are a thousand behind it, and a thousand behind those.

Why did people leave a feed engineered by experts to keep them? Because Sora was honest by accident. Everything in it was machine-made and everyone knew it. There was nobody behind any of it: no friend who might post, no stranger who might be real, no chance that the next thing would matter to anyone you know. Strip out the humans and the feed loses the intermittent possibility of meaning, and it turns out that possibility was the hook all along.

what about the people who never got to practise critical thinking before the machine arrived offering to do it for them? You cannot lose a capacity you never built. My students are sharp, and they know the feed is farming them; what they have never been given is a fair fight. The fair fight is teachable, and cheaply: the three marks of slop and the incentive question later in this chapter are the whole starter syllabus, and I have watched a seminar room of nineteen-year-olds pick them up inside an hour. If you teach, parent, or manage anyone in this generation, teaching them to see the feed is worth more than any rule about screen time.

Upstream of every feed you resent is a person who pressed post on something that did not need to exist, and some days that person is you.

Restore the gap. When you must ask the machine, slow the exchange down to thinking speed. Ask it to reply with one question before any answer. Write your own answer to the question before you continue. This is a two-minute habit that rebuilds exactly the pause the interface was designed to remove. Practise boredom on purpose. The uncomfortable first ten minutes are the resistance working, and once you know that, they change character entirely. Attention is a capacity in Ferdman’s sense, from Chapter 3: it grows under load and withers without it.8 Reading a long book, watching dust motes, walking without a podcast: these are training, and they are training the one faculty every other chapter of this book depends on.

Plato, the actual one, taught that education draws out what is already in the student. He worked through questions. His method ran on the relationship between teacher and student, on dialogue, on the productive discomfort of being asked what you really mean.

My PhD is in atmospheric physics. I spent years measuring greenhouse gases from space, deep in the most funded, most respected corner of the academy. And I can tell you from inside that the technical skills got me the job, and something else entirely made me any good at it. Reading poetry trained me to notice what a dataset was not saying, to hold uncertainty open instead of resolving it early because resolution felt better, to read the same output three times and find a different meaning each time. Every critical decision I made as a scientist was an act of interpretation. I did not know that at the time. AI made it impossible to miss.

And the skills left standing on the far side, the ones no model produces on demand, turned out to be the ones we had spent twenty years telling students not to study. Judgement. Interpretation. The reading of what is absent. The capacity to hold two contradictory ideas without rushing to collapse them. The willingness to stay in ambiguity until it resolves honestly. These were filed under ‘soft skills’. They are the hardest skills there are, and the evidence for their new market value is now arriving from the least sentimental sources available.

AI is fluent. Fluency reads as quality to anyone who has never been taught to take a text apart. Goodness requires judgement, and judgement has to be built in a human, slowly, under load, which is precisely the load the tools now offer to carry for you.

A poem is completed by what the reader brings, a lifetime the text triggers and does not contain. The machine has no first encounter to remember. It reads every poem as a professor and no poem as a person.

The moat is us. The moat is the humanities: history, linguistics, ethics, poetry, the long patient training in what texts do and what they hide. We priced all of that as decoration. It was the load-bearing wall, and we found out the way you usually find out about load-bearing walls.

Where the task is bounded, the comparison honest, and a human with judgement stands downstream, AI is excellent, and refusing it can be its own failure of judgement. The technology finds proteins, reads scans, raises floors, opens doors for people the built world ignored. Chapter 2’s four tests are the whole toolkit in portable form, and the harm chapters exist so that you can apply them with your eyes open.

The machine’s content is engineered to be consumed the way sugar is engineered to be eaten. The supply is now infinite and the attention it feeds on is not. One question, asked of everything you consume and everything you make, cuts through the entire economy of slop: did this need to exist?

The question ‘did a machine write this?’ is almost always the wrong question. The right ones are older. Is it true? Is it any good? Did the person who signed it stand behind it?

This week, before you prompt an AI about anything that matters to you, write two sentences of your own answer first. Just two, before the machine gets its turn: what you actually think, and what you most need to know. Then prompt, and watch what changes. The machine’s answer will be better, because you gave it something real to sharpen against. Your reading of the answer will be sharper, because you now hold a position for it to differ from. And once or twice this week, you will find your two sentences were already enough, and you will close the tab and get on with your day.