Note-taking and "second brain" tools (Obsidian, Roam, Logseq) are very good at storing what we know, but they offer little help in reasoning about it. In these tools, the meaning of a link between two notes lives only in the author's head — a line between "A" and "B" could mean supports, causes, contradicts, or nothing at all.
My claim is that if notes carry explicit types (problem, hypothesis, conclusion) and links carry explicit meaning (leads to, supports, contradicts), the structure of one's own reasoning becomes inspectable: a user can see which conclusions rest on unverified hypotheses, which questions block a decision, and where each thought came from.
This should be relevant to LessWrong because it is an attempt to make one's own reasoning legible — to make visible the moment where a probable explanation is treated as an established fact, or where an important decision rests on a hypothesis nobody checked.
I built a prototype to test this hypothesis, and I do not yet know whether the typing actually helps or just adds friction. This post lays out the idea, gives a concrete example, and marks where I expect it to break.
If I had only one word to describe a state in which a couple of friends from my generation and I find ourselves most of the time, that would be it - * overwhelmed*. This makes sense though, as the amount of information immediately accessible to us has increased dramatically and the Internet has played a significant role in it. Online news, YouTube, social networks compete for the attention of a modern human on a daily basis, and, most of the time they succeed. That being said, we have to draw reasonable conclusions and make sound decisions in our day-to-day life, which can be a complicated task as the external informational noise coming from media can become an internal noise that produces annoying questions and cyclic thoughts that don't lead anywhere.
The "second brain" tools like Obsidian, Roam Research, Logseq, and Tana can store information and partially offload it from the user's brain, thus the name. Writing down notes is an effective way to archive knowledge, but it provides limited assistance in reasoning and the decision-making process. Though, there are techniques like Zettelkasten that can organize and structure Obsidian notes. The main idea of the Zettelkasten approach is that atomic notes can be linked to each other. The main drawback of Zettelkasten, in my opinion, is that it provides very limited semantics and the kind of relation between the notes is left to the user, which can introduce friction points when working with a large number of notes. Externalizing information using "second brain" tools reduces the burden of remembering it, but does not solve the harder problem of deciding what matters, how ideas relate, and which unresolved questions deserve attention. Below is a depiction of the Zettelkasten-connected notes.
The semantics of notes live mostly in the author's head, which makes it hard to revisit or critique your own reasoning months later.
In my attempts to fight the overwhelm I gave Obsidian and Zettelkasten a try, but, due to limited semantics the approach did not fit me. As a result, I have come up with an alternative approach to notes and links between them. The approach is simple - notes should have types and meaningful connections between them. General notes become ideas, conclusions, problems and hypotheses, and links between them acquire meaning - leads to, derived from, part of, supports etc. According to my assumption, by adding semantics to regular notes, we get a technique that might help to reason efficiently, if used and interpreted correctly. Typed relationships do not determine what matters automatically, however, they can help to identify reasoning bottlenecks - unresolved questions on which several conclusions depend, weak hypotheses supporting important decisions, or unaddressed contradictions. The example of such graph can be as follows:
Problem: I feel overwhelmed
leads to
Question: What is the main cause?
leads to
**Hypothesis A: **The overwhelm is caused by external noise
Hypothesis B: It is caused by too many unresolved commitments
A leads to
**Idea: **Cut down the information originating from media
B leads to
**Idea: **Finish the tasks I can close by the end of the week
The graph does not answer the question for me. By adding a second hypothesis next to the first, it stops me from acting on the most obvious explanation before checking whether the competing one matters more.
Externalizing cyclic questions may help us distinguish between questions that require evidence, questions that require action, and questions that can remain unresolved for now. Though, there is a risk that additional complication of node types and their relations will cause excessive friction in the thought-documentation process. Below is a depiction of the same notes but with added semantics.
This is an experimental approach and its effectiveness is an open question. I built a small implementation of this idea called Arguma to see whether the representation is useful in practice. The main difference between my implementation and existing popular implementations like Kialo is that Kialo primarily models claims and their supporting or opposing reasons while Arguma experiments with a broader set of cognitive roles such as questions, problems, hypotheses, ideas, and conclusions and with relationships beyond support and opposition. Readers can see the proposed implementation by following the website https://arguma.io. The tool is in an early access state. One feature of Arguma that can help in collaborative reasoning is the ability to share the reasoning chain consisting of multiple logically interconnected thoughts. This is the example of a shared chain:
https://arguma.io/share.html?t=3e29e336f7045dc29dfb1d1faa9e3d9c908e97c0ccabdfd4
Anyone is welcome to try it out and leave feedback which I will be happy to receive.
Arguma is my attempt to test whether typed thoughts and explicit relationships make reasoning easier. The relevant questions are whether users notice unsupported conclusions more easily, identify unresolved questions that block decisions, and reconstruct their reasoning later—or whether maintaining the structure costs more effort than it saves.
Thanks for reading!