TLDR?

Sorry, this one’s a story arc, you’re in it for the long haul.

As a self-respecting nerd I’m a big fan of the late Sir Terry Pratchett.

As a (somewhat) self-respecting teacher, I always had a special place in my heart for this part of the first Science of Discworld book:

“As humans, we have invented lots of useful kinds of lie. As well as lies-to-children … there are lies-to-bosses lies-to-patients and for all sorts of reasons, lies-to-ourselves.”
— The Science of Discworld, more at the L-Space Wiki

Telling lies is bad, right? That’s what we were all (hopefully) taught as kids, so what are these sorts of lies and why are they any different? Is this just another case of grown-ups and “rules for thee and not for me”? Of course that’s not the case. Really what we’re talking about is just appropriate summarisation: feeding people the right level of information for the right situation.

For students this should be about scaffolding, and avoiding things that might damage later understanding like forming lasting misconceptions, or misunderstandings of the whole domain of knowledge.

For leadership this should be about giving them a reason to agree with you providing enough context to make the summary make sense and make informed higher-level decisions.

For ourselves? Well I dunno, that’s between you and you.

Turtles all the way down

“So I think there are these huge dangers. I mean I’ll mention one more which worries me as an educator, which is I think there is a real risk that generative AI will destroy education. …

And here what I mean is not so much the fact that students won’t write their own essays. …

But it’s really that no one will read anything anymore. This is what we’re doing. It worries me. You know when you open up a document in Adobe Acrobat Reader it says this looks like a long document. Would you like me to summarise it for you? …

Search now they give you the AI summary first … And the AI summaries are often inaccurate. But what worries me even more than that actually is the fact that you never go to the text itself, you just go to the summary. And so education will be a matter of just accumulating summaries. And that’s what will damage people’s ability to actually understand what they’re talking about.”

Note: The podcast doesn’t come with a transcript, so any errors here are mine for trying to transcribe while my kids were nattering away next to me the whole time.

— Professor Tim Crane, on ABC's The Philosopher's Zone Podcast

As an educator I think there are some real risks to traditional education from AI and LLMs, but I don’t think summarisation is the heart of it. In fact, education is at its core a summarisation machine:

  • Teachers provide summaries to students in the form of lessons and notes.
  • Students provide summaries of those summaries to teachers, with a summary of independent learning sprinkled on top.
  • Teachers provide summaries of those summaries of summaries to students, parents, and authorities in the form of feedback, marks and grades, reporting, and competencies.
  • Students then pass on those summaries to employers or further education as transcripts, references, or some letters before or after their names.

Summarisation may be part of the issue for education, but the imperative to prove that you’re educated is at the root. We have a top-down, curriculum-driven approach to education that likes arranging students into neat little conceptual boxes like “humanities” and “science”. How does the arrangement work? Well of course we reduce years of knowledge and skills to a list of checkboxes with an arbitrary threshold for success (optionally graded on a curve if you’re feeling frisky). We’re forced into this continual process of lossy compression because our system requires one thing from education: a binary “educated” or “not educated”.

Am I proposing all of that gets thrown away? Hell no. Everything would fall apart if we spent all of our time trying to validate every single thing independently. Well, actually, I would like to throw away quite a few choice bits and pieces… 🤭

But how do we communicate?

Summarisation raises all sorts of tricky situations in everyday communication:

  • Communicating nuance in text has been a problem forever, even with these incredibly clear and well understood emojis in our arsenal 🍆💅🍑🥵✨.
  • Regional usage of terms and generational (ugh, sorry) lingo introduces all sorts of challenges.

…and then there are visualisations.

Make Me Data Literate

…this is a great opportunity to link to Dr Linda McIver’s lovely podcast Make Me Data Literate in which she interviews all sorts of wonderful data nerds of various stripes.

One of Linda’s standard questions is “What’s the first question you ask when you look at graphs in the media?” and this one has been living in my head for a while, because there are the obvious answers of whether any of the regular visualisation sins have been committed, but visualisations are summaries. Even if (the font size of the ‘if’ may vary with the source and purpose of the piece) everything has been done in good faith and with adequate attention to detail, are we equipped to understand the summary presented to us?

I was reading a post by Daisy Christodoulou a while back titled What if transferable skills don’t exist? and it’s been eating at me ever since. As a Digital Technologies teacher I’ve always had a lot of confidence in the value of the learning area, even if only for the focus on developing skills in problem decomposition. Now I’m starting to doubt some of this. How much do we rely on good domain knowledge in order to effectively analyse and decompose problems? As an adult it’s hard to bring all of the accreted knowledge and experience out of the dark recesses of the back brain and into the light to evaluate how much they help us analyse problems we encounter.

If I look at a graph in the media and there are no obvious mistakes or trickery at play, and it comes from a fairly reputable source, am I equipped to understand where the nuance is hiding, or whether authors made any mistakes? Depending on the domain, maybe not. I think this goes for any sort of communication, visual or not. As a global collective, we’ve become quite adept at appearing confidently knowledgeable and authoritative, and LLMs, being trained on collective human works are carrying on this fine tradition, except now carrying that added implied legitimacy of “but algorithms!”

On the other hand…

This point may be moot, since societies are increasingly turning to authoritative feelings rather than knowledge.

So do I have a point?

To bring this back to education, maybe transferable skills like “critical thinking” don’t exist, and all this time I’ve just been teaching domain knowledge wearing a fancy hat. Firmly in the “lies to ourselves” category there. This uncomfortable feeling has been growing while I’ve been writing this post.

If there’s anything that I think truly endangers the idea of education, it isn’t summarisation alone, it isn’t unwillingness to read source texts, it isn’t even submitting AI-generated work and plagiarism (although it is very much in the ballpark of academic integrity). It’s the lies:

  • The lies upwards about our own thoughts and analysis.
  • The lies downwards from AI-generated materials and feedback that lack the nuance coming from genuinely understanding the domain.
  • The lies to ourselves that reading a summarisation of a likely poorly formulated question to a black box system counts as knowing something.

Lies to children are grounded in building learners up. Lies from AI are just building a gilded cage for ourselves.

I have only made this blog post longer because I have not had the time to make it shorter.
— Apologies to Blaise Pascal