Right now we have more information at our fingertips than at any point in history. The explosion of information drove the creation of internet search engines & directly led to the development of the Large Language Models which power modern AI models.

But is all of this information worthwhile if we can’t make sense of it?

In the early 2000s I was the Library Systems Administrator at Victoria University of Wellington. The library contained around 1 million catalogue items, across multiple physical sites. The main library is located in the Rankine Brown Building - a 1960s behemoth perched on a hillside overlooking New Zealand’s capital city & packed with stacks & stacks of books.

The technology we used a quarter of a century ago is pretty much the same as what we’re using today - we had databases, networks, servers, internet browsers & yes, we had Google. A big focus of what we did was working out ways to improve access to the information for students & scholars. But it didn’t end there - specialist librarians were on hand to help people make sense of the information they found.

My time in libraries introduced me to the concept of Data > Information > Knowledge > Wisdom. The idea is that we start with raw Data (numbers or words), structure it into Information (essays, reports etc.). We absorb that information & apply context to transform it into Knowledge. Finally, we develop Wisdom by applying our critical thinking & lived experience to our knowledge. This concept is a bit dated now (especially as “wisdom” is a bit ephemeral) - but it’s still a good foundation to explore how we learn.

Too Much Information

For many people the idea of a library is dated. We carry access to mind-boggling amounts of information in our pockets, why do we need to fossick through dusty bookcases?

While libraries have catalogues & their own search engines, the first mass-market technology that helped us navigate the expanding internet was search engines. If you wanted information on a specific topic you could enter that into “search” & it would return a list of related information.

We all know what happened next, the explosive growth of online information overwhelmed “search”. Sure Google can return 693,000 results for “pangolin diet” - but how many results will you look at, deciding that you’ve acquired enough knowledge about what pangolins eat?

While the economics of search engine optimisation & online advertising broke part of the information ecosystem, the desire to shape humanity’s beliefs broke another big part. Propaganda has been with us for millennia, but those wanting to influence our values & perceptions of reality could suddenly do it on a global scale. Those ahead of the game could use misinformation to direct people to buy whatever they were selling - be it a political belief, questionable lifestyle choice or snake oil cure all.

The underlying issue became how people turned Information into Knowledge. In a world awash with information how people were directed to consume specific information was key. Knowledge be damned, if we just turn the information firehose on people they’ll get used to whatever flavour of information we’re hitting them with.

When Search Became AI Chat

Context is key to understanding information. Who authored the information you’re consuming? What are their goals? How did you end up with this information in the first place?

Search increasingly left us awash in a sea of low quality information with very little in the way of context.

Then came AI models that had scraped much of the internet & could search the internet for up to date information.

Rather than a short search term you could write sentences, or even paragraphs to define exactly the information you’re looking for. AI models would return a well written summarisation & even (sometimes) supply citations to back up its response.

Surely - we’ve cracked it now! People quickly discovered that rather than a Google Search they could ask Claude, ChatGPT or Gemini a question & it would come back with information wrapped in context.

Too Little Context

This is where things get murky. Context is how we make sense of information, but a “context window” is a technical term for AI model short-term memory.

If we’ve fed most of the internet to powerful AI models & they can search the internet, doesn't that mean that they’re getting close to being an all-seeing digital oracle?

First the obvious - AI models notoriously struggle with separating fact from fiction. From Google’s AI suggestion to put glue on pizza, to Grok’s obsession with non-existent “white genocide”.

But there’s a bigger issue lurking under the hood of AI models - their short-term memory is surprisingly limited. You may have come across a chatbot “compacting” the conversation - this is where it summarises what you’ve talked about into a smaller form, to fit into its memory.

The deeper you go into a conversation the more the AI model forgets or over-simplifies. To the point that context can be lost, diluted or warped into something misleading or unrecognisable.

Even prioritised instructions & key concepts can be forgotten - to the point the model appears to be suffering from digital dementia.

Context is Critical

…Thinking that is!

Yes, that was a dad joke - but critical thinking & understanding context remain key human skills we can’t farm out to machines.

As much as AI models appear to be all-knowing oracles, their underlying architecture means that context & short-term memory are major limitations.

While we might think that librarians are in the rearview mirror of history their skills are needed more than ever. Not just in the professional sense, but in that we all use our internal librarians when transforming Data to Information to Knowledge to Wisdom.

Banner Photo by Jason Leung