Can we help you deal with information overload?

The experience of information overload is nothing new for us humans.

Every so often a new technology comes along and challenges us to find new ways of curating and filtering the world around us.

This time it’s generative AI.

As Jonathan wrote in his recent essay, “Sword and shield: AI, democracy and the fight for the future of information”, “there are more things than ever to subscribe to and ultimately more things to read, listen to, watch and digest.” And, as the same essay highlighted, the chances are generative AI will create more fragmentation, more content – and more overload.

So, we’ll have to cope with more noise in the AI age.

Which is why I’ve been wondering if AI can help us find signals.

Our first experiment is to see if we can use AI to help us create a personalised daily news digest for Jonathan that reads relevant unread emails, scrapes RSS feeds and checks the links he sends via WhatsApp.

Sure, it won’t be perfect – generative AI is not some silver bullet - but if it can help us pick up more signals from information we might not otherwise have consumed, its got to be worth a try?

How I plan to use AI to build a personalised daily digest

First things first. We are focusing on things that would otherwise go unread. This avoids the challenge of AI making automated decisions that mean a person misses something they’d otherwise see… something we definitely want to avoid, particularly given how prone generative AI is to bias.

1. Defining what might be important information (Spotting signal)

To distinguish someone’s “signal” from their noise, we need to know what actually interests them.

To get this information, I propose to combine human and generative AI: I’ll interview people, collect their writing, and use the transcripts and generative AI to create a prompt that I hope fill filter what is relevant – or not – for a person.

The first person to go through the process will be Jonathan himself! Usefully he has already written an essay which offers really clear data on what he is interested in (the one mentioned earlier - “Sword and shield: AI, democracy and the fight for the future of information”). We’ll use that to create a filtering prompt that we hope will tag articles to the five different themes he discussed in his essay.

2. Capturing relevant information (Filtering Noise)

The next stage will be to redirect all the content that is currently being missed to a central repository, where it can be assessed and filtered. We need to get a good flow and range of content…we won’t be solving the problem if we’re not picking out the top x% of content that is otherwise missed. Once content is being fed in, the generative AI prompt will do its filtering.

3. Share the content (The end product)

To get the filtered content back to Jonathan, or anyone else who joins us for the journey, we’ll need to send some form of digest. An email would seem to make sense. But there’s a second important thing we’ll need to build here: a feedback mechanism.

Any good AI algorithm has feedback – and though a generative prompt is not quite the same as a Facebook algorithm, I’ll need to know when it gets things right or wrong - and why. There are some better types of AI out there for learning but as this is an experiment we’re going with the easy route….

How does this sound to you?

If we can get this right then we’ll be in a position to help people stay better informed on relevant topics by showing them content they’d otherwise miss without taking up any of their time.

Would you like to get involved in the experiment and test the process out?

We’re looking for a few people who’d be interested and would say whether they see promise, and under what conditions.

This is not about us creating a new product – the fact is that Readwise Reader is 90% there already, and may over time get to 100%. But it is about attempting to prove, or disprove, the generative AI promise of saving time and tackling a problem that we feel plagues people working in progressive causes. And others too, but that’s not our focus or passion.

What would we need from you to participate? Not that much… It would take a few calls, some commitment to be part of the testing while we go, and agreeing to write something about the process and idea would be good to.

Nick Scott

P.S If you’re wondering who I am, you can find out a bit more about me here.