Notes and Summaries

AI summaries

With so much being invested in increasing our use of genAI, it is now almost impossible to read any digital text without being offered an AI-generated summary of key points.

Since most of what I read has abstracts or executive summaries, this is of little use to me personally, but it does bear some examination. What is the cognitive function of such summaries? They are sold as 'time saving', but is that really what they are doing?

An abstract or executive summary allows me to decide whether I want (or need) to read the document. It is more informative than a title and is written by the authors to promote their work, to give me a reason to read it. It is not written as a substitute for reading it, but as a form of marketing or promotion, which does allow the reader to quickly identify what may not be relevant to their purposes. So it does save time in the sense of helping reduce time wasted reading something only to find it did not contain what was being sought by the reader.

It is possible that some people may use an AI-generated summary like this, but that is not the market being sought by those encouraging us to save time with them. They want us to be tempted to save time by reading the summary instead of reading the document.¹ Getting an AI-summary is much more akin to asking another student if you can copy their notes for a lecture you want to skip.

Digression on note-taking

The persistence well into the 21st century of the University lecture should be puzzling. When European universities were founded in the middle ages, the transmission and demonstration of knowledge was largely oral, so one person talking to a group was an obvious efficiency in transmission.

At some point during the late medieval or early renaissance (someone will know roughly when - if that is you, please inform and correct me) commonplace books became affordable and thus widespread among people who might attend lectures. This led to a culture of personal note-taking as a tool for knowledge acquisition and retention, and much was written then as now on how to do this well.

Of course, taking notes in lectures or speeches has been around for millennia (it is how we know so much about Aristotle's thought) but when it becomes an expectation that every student will take notes, something changes. The lecture now has two products, a cognitive state and some notes, which bear a complicated relationship. Sometimes the notes play the role of a dictated textbook. Sometimes they are an aid to later recall. Sometimes a means of acquiring greater understanding during the lecture.

The same is true of notes taken when reading. They may be an extraction of information into a personal textbook, an aid to recall or an aid to comprehension. Notes which are an aid to comprehension will almost always include questions or reflections originating with the notetaker.

Now that almost all university lectures are recorded with transcripts and the slides or handouts circulated to all, it seems that the only remaining function for notetaking in lectures is as an aid to comprehension. One might draw a similar conclusion about reading, now we have widely available, searchable, annotatable digital texts.

Reading as thinking critically

Clearly an AI-generated summary of a document cannot be an aid to comprehension of that document, because there is no process of reading it in order to achieve comprehension. Nor, obviously, can it be an aid to recall. Its closest cousin in traditional notetaking is the personal textbook giving introductory overviews of things you want to learn about. However, unlike a real textbook it is not filtered through the understanding of an expert and unlike notes on what you have read yourself, its creation does not involve a level of understanding on your part, it is like borrowing someone else's notes or a cribsheet.

The first problem there might look like a technical problem: better and better LLMs will replicate the understanding of experts. I doubt that but here is not the place to challenge it. The second problem is more interesting: in order to create your own notes you have to read and understand what you have read to some degree. We can take notes on what we read because reading itself is a cognitive process.

Is this one we can outsource to an AI? Is it something that we no longer need to be able to do, like mental arithmetic? Let's explore that analogy further.

Some people enjoy mental arithmetic. It may also be an essential pedagogical stage in teaching maths. It can sometimes be useful when calculators are not handy or spotting a mistake on a bill that needs checking another way. But it is hard to argue that there is any intrinsic value in doing a calculation in your head rather than on a calculator, and it introduces additional sources of error.

Reading unsummarised documents, however, is very different. It can be enjoyable. It may have pedagogic value. It may still be useful. But it also has an intrinsic value. When you read you enter into another person's thought processes and follow them through those twists and turns. You think their thoughts, in their way, in their order, at their pace. With a summary you may learn what they think, in the sense of where their thinking process ended up, but you don't know how they think. And if you don't discover how other people think, you will always be trapped in your own idiosyncratic mind.

Reading also involves a form of reflective thought: as you follow another's thought process you are still present, with your own views, opinions, experience and knowledge. So you think with another but also about their thinking. This also happens when you listen to someone attentively. It is the very fundamental form of exercising a critical faculty, and once refined can be effectively turned upon your own thinking to improve that.

AI-generated summaries can certainly save a lot of effort and time, but they thereby lose the value of that process they shortcut. For most people in the humanities, this is obvious. Now is the time for active resistance of a narrative which promotes time-saving over using our most valuable intellectual capacities.