TL;CR: Too Long, Can’t Read

There is a fairly new solution to being sent something long: don’t read it.

I do this all the time. When a report comes through and I want to know what it says, I give it to AI and ask for the important bits. I might then ask questions, challenge something that looks odd, or go back to the original if I actually care about the detail. Quite often I come away knowing more than I would have if the alternative was a PDF sitting unread in a tab for three weeks.

This is not limited to AI. We skim articles, listen to podcasts at 1.5x speed, watch explainers instead of lectures and learn things from people on TikTok who can make a subject interesting in ninety seconds. I probably read fewer books than I did a few years ago, but I am almost certainly more informed. There are obvious reasons to be suspicious of short-form media, but there is also something slightly ridiculous about pretending that knowledge becomes less valuable when it arrives vertically.

Still, I wonder if something is changing underneath all of this. TL;DR originally meant too long, didn’t read: the information was there, you just couldn’t be bothered. We are moving towards a world where not reading something needn’t mean not knowing what it says at all. Someone, or increasingly something, can read it for you.

Kids these days, 370 BC

This sort of anxiety is almost comically old. In Plato’s Phaedrus, Socrates recounts a warning about writing itself: if people can store knowledge externally, they will exercise their memories less and may acquire the appearance of wisdom without understanding. It is essentially a complaint that a new technology allows people to outsource part of their thinking.

The obvious problem for modern technological pessimists is that Socrates was worried about writing. If he had won, this essay would be extremely difficult to distribute.

Writing did weaken the need to remember certain things, but that turned out to be a reasonable trade. It also allowed ideas to survive their authors, made complex arguments possible and eventually helped put an extraordinary amount of knowledge within reach of ordinary people. The arrival of books, radio, television, Google, YouTube and Wikipedia each changed what we needed to know ourselves and what we could reasonably expect to retrieve from somewhere else.

AI may just be the next version of that. I don’t particularly want to memorise the provisions of a tax code when I can look them up. Nor do I think someone has failed intellectually because they learned how a nuclear reactor works from a good YouTube video rather than a textbook.

The difficulty is working out which mental tasks are like remembering a phone number, safely outsourced, and which ones are part of understanding itself.

That question is becoming more relevant because our attention really does seem to be moving towards faster and more fragmented forms of information. There are signs, at least, that sustained reading is under pressure. The National Literacy Trust surveyed 76,131 children and young people in the UK in early 2024. Among 8–18-year-olds, 34.6% said they enjoyed reading in their free time and 20.5% said they read something daily in their free time, both the lowest figures the organisation had recorded in nearly two decades.

It is tempting to look at figures like these, glance at a teenager scrolling TikTok and declare the case closed. I don’t think it is that simple. Reading enjoyment is not intelligence, and a young person who watches a good history documentary, listens to a politics podcast and spends an evening disappearing down a Wikipedia hole may learn considerably more than someone dutifully reading a bad novel. Even the word reading is becoming a slightly awkward proxy for the thing we actually care about, which is whether people can understand complicated ideas and form reasonable judgements about them.

That matters because complicated ideas are not exactly disappearing.

Understanding your mortgage, pension or energy bill requires some basic numeracy. Understanding an election may involve polling error, public finances and competing statistical claims. Climate change, artificial intelligence, monetary policy, migration and modern warfare are all fields in which it is possible to sound informed after five minutes and remain completely wrong.

PISA gives at least some reason for concern about the underlying skills. Nearly 700,000 15-year-olds across 81 countries and economies took part in PISA 2022. Across 35 OECD countries with comparable results, average reading performance fell by around 10 points between 2018 and 2022. Covid was an obvious part of the shock, although the OECD’s longer-term analysis shows that the decline in reading had begun before the pandemic.

There is an odd mismatch here. The world is requiring more interpretation just as our information environment is becoming extraordinarily good at removing the need to linger over anything.

Huxley had a point

E. M. Forster got surprisingly close to this in The Machine Stops, written in 1909. His characters live largely alone, communicating through screens and receiving knowledge, entertainment and social contact through an enormous technological system. The eeriest part is not that he anticipated video calls. It is that people increasingly know the world through the Machine’s representation of it rather than through direct experience.

Our version is less dramatic but recognisable. A study is published. A newspaper writes about the study. Someone posts the article on X. Someone else screenshots the post. A creator makes a video about the screenshot, and by the time it arrives on your phone the original research is several floors below you.

This has always happened to some extent. Newspapers exist partly because nobody has time to sit in Parliament, read every government document and personally telephone the Treasury. Expertise itself depends on division of labour. The internet did, however, create the peculiar possibility that the primary source can be sitting one click away while almost nobody clicks it.

In one sense, this is just another layer of abstraction. Modern life is built on them. You do not need to understand TCP/IP to send an email or monetary policy to use a bank account. Abstraction is what makes complicated systems usable. The problem is when we can no longer move back down a layer when we need to: when the summary starts to become the source.

AI adds another layer because it can now do some of the reading itself.

This is already happening, although less dramatically than some coverage suggests. Reuters Institute’s 2026 Digital News Report found that 10% of people across its surveyed markets had used a standalone AI chatbot for news in the previous week, up from 7% a year earlier, rising to 16% among under-35s. Among those using chatbots for news, the most common use was asking follow-up questions; 34% used them to summarise news and 30% to make news easier to understand.

That sounds less like people abandoning information than changing how they interrogate it.

Huxley and Orwell offer a neat shorthand. Orwell worried that information would be kept from us. Huxley worried that we would become too entertained to care.

Orwell: you can’t read it. Huxley: you don’t want to read it. 2026: don’t worry, the AI read it for you.

But Forster gets closer to the interesting part. The information still exists. You can access almost all of it. You just increasingly encounter it through something else.

And I don’t actually think the conclusion should be pessimistic.

AI may turn out to be one of the best tools we have ever invented for dealing with a world that has become too complicated for any individual to understand.

Take a 140-page council budget. The vast majority of residents were not going to read it before ChatGPT existed. If they can now ask why spending rose, which services were cut, how debt changed and where a suspicious number came from, that is not the death of literacy. It is a fairly remarkable improvement in access to public information.

The same applies to scientific papers, contracts and financial documents. Being able to ask unlimited stupid questions without embarrassment is useful. A good AI can translate specialist language into normal English, explain the bit you didn’t understand and keep trying until you do. There is a plausible future in which people read fewer primary documents but understand considerably more of what is inside them.

This is where the distinction becomes harder than simply deciding whether AI is good or bad. There is a difference between using a machine to get into something difficult and using it to avoid having to understand the thing at all, but in practice those behaviours sit on a continuum. I use AI for both. Sometimes it makes me investigate something much more deeply than I otherwise would. Sometimes I ask for the summary, take the answer and move on.

The danger isn’t summarisation itself. We have always relied on other people to summarise things for us. The danger is relying on a summary without knowing who produced it, what it omitted, or how to inspect the thing underneath when it actually matters. And that ability will not be evenly distributed: people who know when and how to drop down a layer will have an advantage over those who only ever see the interpretation.

The consequences probably depend less on whether we technically “read” than on whether we retain the ability and inclination to interrogate what we are being told. A calculator is incredibly useful if you understand numbers; it becomes more dangerous if you have no way of noticing that the answer is obviously wrong. Something similar may be true of AI and information.

Who reads the world?

That is where this becomes more than an argument about attention spans.

Mass literacy changed the distribution of power partly because it allowed far more people to encounter information without somebody else having to tell them what it said. You could read a law, contract, newspaper or political argument yourself. Most people did not spend their evenings reading legislation, obviously, and intermediaries never disappeared. But the ability to check mattered.

If more of our information now comes through layers of interpretation, the people and systems doing the interpreting become more important. This is not automatically sinister. A good journalist can understand a subject better than I do; a good teacher can save me weeks; a good AI can make an inaccessible document useful. But summarising something always means deciding what deserves to survive the summary.

Software makes this particularly interesting because it scales. The assumptions, mistakes and omissions of something used by hundreds of millions of people can become part of the information infrastructure itself. That does not mean AI is secretly telling everyone what to think. It means that who builds these systems, how they work and how easily we can look beneath them begins to matter rather a lot.

None of this makes me want to return to a world before TikTok, YouTube or AI. I like all three. I am probably more informed because they exist, and I suspect many other people are too. Nor is there much point telling everyone to read every original source; civilisation works because we don’t have to.

But outsourcing intellectual work does not make the work disappear. It moves it somewhere else. That makes it worth paying slightly more attention to what we choose to outsource, who or what we outsource it to, and whether we could still recognise when the resulting explanation is nonsense.

We have spent centuries making written information more widely available. We may now be entering a period in which access is no longer the difficult part. The harder question is what happens to a society where almost everyone can read, almost everything is written down, but much of the actual reading is increasingly done on our behalf.

That society need not be less intelligent or less informed. It may be the opposite. But it would be a society increasingly dependent on its interpreters, which makes understanding who they are, and occasionally checking their work, rather important.