I have a confession: my Second Brain hasn’t been working very well lately.

The problem is the “Resources” stack, one of the four main buckets that are part of my PARA organizational system.

It has ballooned to more than 100 individual notebooks, and I can’t even remember which ones exist. Which means the decision of where to put a new note is extremely taxing and time-consuming, requiring me to look through this massive list.

It hasn’t been working for the very same reason that I developed PARA in the first place – when you categorize information by topic, it eventually expands to a completely unmanageable number of overly specialized subcategories. Ideas get siloed and completely forgotten.

I took to Twitter and asked whether someone would be willing to help me come up with a better set of categories. I use a free service that automatically saves any article I “favorite” in Instapaper to a dedicated Evernote notebook. I’ve had it running automatically in the background for years, and recently realized I had collected nearly 300 of my favorite articles going back to May of 2015. I wanted to use this body of articles to better organize the ideas I drew from my reading.

Ergest Xheblati (@ergestxheblati) generously volunteered to use his knowledge of machine learning to give it a try. You can read his full technical report here, along with details of the algorithms and tools he used, the workflow file in case anyone wants to play with it, and other observations on his process. Here are the results in a Google Spreadsheet, with separate tabs for the raw output, an article-by-article index, and my own labels.

The reason I find this approach so exciting is that it is aligned with my fundamental approach to organizing information – discovering the topics within an existing body of knowledge in a bottom-up way, instead of trying to determine them upfront in a top-down way. The result is a fascinating window into a new way of organizing knowledge that uses the power of algorithms, rather than human effort.

Here’s the main takeaway: the algorithm identified 19 topics, with 1 to 94 articles in each of them. Here is a summary of the results, with labels I added to describe each topic:

This is a remarkable result, in my opinion, since this list is a highly accurate representation of my interests, and was created without any upfront categorization on my part. The topics are fairly broad, but also have clean edges, at least in my mind. The pairing of topics makes complete sense to me, even though they might be unorthodox. There are few enough that I can keep them all in mind and see them all at once, while being specific enough that it’s clear where a given piece of information goes.

 

I decided to reorganize my Resources stack according to this new list of topics. This was a test of the flexibility and plasticity of my Second Brain – in theory, it should be easy to change its structure, and yet I find that I rarely do so because I don’t know what to change it to. This experiment gave me a clear direction for improvement based on real-world data.

As I did this reorganization, I found that most of my existing Resource notebooks already fell into these topics to a large extent. This meant that I didn’t have to refile one note at a time, which would have taken forever. Instead, many smaller, more specific notebooks could easily be “collapsed” into the new notebooks. Here are some examples of how my existing notebooks fit into the new, broader topics:

  • “Books & writing” incorporated:
    • writing
    • publishing
  • “Design” incorporated:
    • design thinking
    • workspace design
  • “Software & product” incorporated:
    • agile
    • software
    • lean
  • “Business & strategy” incorporated:
    • strategy
    • sales
    • entrepreneurship
    • social entrepreneurship
    • personal branding
  • “Culture and creativity” incorporated:
    • creativity
    • culture
    • magick
  • “Knowledge management” incorporated:
    • crowdsourcing
    • extended cognition
    • group knowledge management
    • pkm
    • note-taking
    • learning
  • “Therapy & personal growth” incorporated:
    • ethics
    • pleasure
    • moods/emotions
  • “Systems & information” incorporated:
    • networks
    • emergence

Here’s what my Resources stack in Evernote looks like now, with 36 notebooks instead of 106 (the new topics are in ALL CAPS):

Here are a few other observations about the process:

1. There are actually two kinds of Resources notebooks

As you can see in the screenshot above, not every existing notebook could easily be recategorized. The ones in lower case I felt were more “functional,” more like true “resources” that deserved to be kept separate for specific use cases. For example, I know exactly where to look when I want to see examples of “annual reviews,” and don’t need those notes to mix with others.

2. Progressive Summarization makes broader topics possible

At first, I had some fear that dumping 100 notes from 6 notebooks together in one place might lead to overly broad collections that weren’t as easy to search.

On the other hand, I know my best work comes from connecting ideas across different domains. I want there to be some overlap and “bleeding” of ideas across the typical boundaries between fields. For example, there is a lot in common between “workflow design,” “workspace design,” and “product design,” and I want those notes to freely mix together. This is more likely to happen when I’m able to see those notes juxtaposed against each other in the same notebook.

Luckily, with the power of progressive summarization, I can very quickly scan a large number of notes, and within seconds determine if a given note is relevant to my needs. In other words, I can have the benefits of large, broad collections without paying the penalty of tediously re-reading one note at a time. And in fact, I found that even among my existing notebooks, I was already using overly broad topics such as “productivity” and “psychology.” So this new categorization didn’t really cost me anything.

2. This reorganization gave me the chance to make other changes

With any organizational system, there is the constant threat of inertia. The “way you do things” can easily become a rigid convention that limits change, not only in your note-taking but also in your work and life. It’s hard to find the time in daily work to make decisions about what to archive or move. But I found that since I was making this major change, I also had the chance to delete or archive or move many notes and even entire notebooks that no longer serve my purposes.

I archived notebooks on “time-tracking,” “habit mapping,” and “intermittent fasting,” among others, realizing that I was no longer interested in them. Other Resource notebooks I moved to Areas, such as “Music” and Cooking,” since they’ve become permanent, important, ongoing parts of my life.

3. Many Resource notes I realized were applicable to current projects

As I was moving notebooks, some notes caught my eye as I realized they could be useful in my current projects.

A note from an article on cryptocurrency, for example, I moved to a Project notebook for a cryptocurrency conference I’m speaking at this month. I think this points both to the value of what I’ve collected in the past, and the difficulty of discovering it within my previous bloated system. There is tremendous value in our notes – but it is up to us to discover and rediscover them.

Takeaway

This was a technically complex process, but as a proof-of-concept it points the way to a bright future: ideas organized by machines, instead of by humans.

The history of technology shows a continual evolution of humans doing less and less manual labor, and instead dedicating themselves to higher order creativity and problem solving. That evolution is now reaching the world of learning and knowledge, with the potential to free us from boring tasks that take up a huge percentage of our time as knowledge workers.

If we could develop this technology into a product or service, we could see patterns in our thinking continuously resurfaced, leading to new and even better insights. We could make our computers into true thought partners, able to not only store information, but analyze and even understand what it is telling us.

For now, the good news is that you don’t actually need a fancy machine learning algorithm to tell you what your main interests are. If you find that your Resources have proliferated beyond all reason, just look at the notebooks you already have, and sort them into 15-20 broader topics that are still specific enough to tell you where something goes.

We can’t afford to spend our precious time meticulously filing notes into hyper-specialized categories. Spend that time instead reading, capturing the best of what you come across, and creating new creative works that move your career and society forward.


Subscribe to Praxis, our members-only blog exploring the future of productivity, for just $10/month. Or follow us for free content via Twitter, Facebook, LinkedIn, or YouTube.