— title: Generative AI and Books of Sand date: 2024-08-13 printupdated: false gardenstate: sapling —

Google’s use of generative AI has you save all sorts of info in a heap, and has gen AI make sense of it for you. (See https://youtu.be/KoN_bcDmhR4?feature=shared)

The heap is a Book Of Sand. Having the machine parse through it and generate a guess as to what you want based on that info isn’t the same thing as surfacing that info, and it’s also not the same thing as organising that info so it can be perused by the person generating it.

Without the option to peruse, we can’t make connections we don’t anticipate. The surprise and romance of that is lost.

Without getting the info from the heap directly surfaced, we’re at best receiving a lossy version of our ideas, in a format we can’t control. It’s like the difference between typing notes and writing them: writing the notes gives the flexibility to draw, or to embellish words to make them stand out, in a way that is more flexible and powerful than ascii or Microsoft Word can provide.

Offering genAI this way is lazy: it avoids the effort required to actually design an interface for the heap or organise the info in a way that’s useful for users.

Offering genAI this way is also a lost opportunity to design it in a manner that’s collaborative. For example, the AI could dynamically organise notes, files, photos and other data objects in such a way as to allow a user to answer their questions themselves. The AI wouldn’t provide the answer, but it’d provide tools to help somebody answer.

Using genAI designed this way means ceding control over the answers and how they’re arrived at, accepting a lossy response that might not be accurate, and throwing knowledge into a heap that you have no way to retrieve save a special deity with magic powers which retrieves things from the heap for you, which can’t assist you in a flexible manner like a human assistant would; it can only summarise what it sees, and can only respond to prompts with those summaries.