One way to characterize Personal Knowledge Management (PKM) is to suggest it involves the analysis of actual and potential tactics applied between encountering information and the application of that information. I came to this topic with a background in the development and evaluation of technology tools for academic studying which involves considerable overlap with PKM. I think it fair to say that studying offers an advantage to interested parties because it has a superior theoretical framework and a large volume of theory-driven research. PKM seems to have developed within a framework I could describe as logical rather than research-based, but it is related to methods of considerable longevity (e.g., commonplace books, note-taking within procedural systems such as the Zettelkasten).
This post was prompted by the announcement and availability of a new version of Mem.ai. There are many digital note-taking tools available, but for some time I have concentrated on two – Obsidian and Mem.ai. My rationale has been that I wanted to invest sufficient time in creating and using a personal knowledge management system so that I could offer credible comments on the tools I use and the tactics that are recommended and that I have employed. Part of this involves building a significant collection of notes over an extended period of time. Many recommended practices cannot really be evaluated with a small body of material used for a short period of time.
When I started using Mem it was because I wanted to explore how AI could be applied within a PKM system. With time, Obsidian extensions allowed several different ways to add AI to Obsidian so there was no longer a unique difference, but I have continued to use both nonetheless.
Comparing Obsidian and Mem.AI
When comparing how Obsidian and Mem serve writers between reading and writing, there are distinct approaches each platform takes to facilitate the transition from note-taking to writing.
Obsidian
Obsidian is known for its flexibility and emphasis on linking notes to create a network of ideas. It supports a bottom-up approach to writing, where notes are interconnected through backlinks and tags, allowing users to discover relationships between ideas organically. This method aligns with the slip-box or Zettelkasten approach, which encourages the creation of permanent notes that can stand alone and be easily integrated into future projects. Obsidian’s use of markdown files and its ability to handle large volumes of notes make it a powerful tool for writers who prefer a structured yet flexible environment for developing their ideas.
Mem
Mem, on the other hand, focuses on enhancing the linking capability through AI-driven suggestions. It extends beyond manual tagging and keyword searches by proposing related ideas and documents, which can come from the user’s own mems or those shared by team members. This AI-driven approach aims to improve the retrieval and linking of information, making it easier for writers to access relevant content and insights. Mem’s design is centered around the concept of a “second brain,” where storage, retrieval, and linking are optimized to support the writing process.
Key Differences
- Linking and Organization: Obsidian relies on manual linking and tagging, while Mem uses AI to suggest connections.
- Flexibility vs. Automation: Obsidian offers more flexibility in how notes are organized and linked, whereas Mem provides automated suggestions to enhance the linking process.
- User Experience: Obsidian’s interface is more suited to users who prefer a hands-on approach to organizing their notes, while Mem’s AI features cater to those who appreciate automated assistance in discovering connections.
Both platforms offer unique advantages depending on the writer’s preferences and workflow. Obsidian is ideal for those who enjoy a more manual and customizable approach, while Mem provides a more automated and AI-enhanced experience.
When is the process the product?
Part of the marketing for the original Mem.ai made the argument that the AI capabilities freed users from some of the process requirements of other note-taking tools. The differentiation of notes into folders and the connecting of notes by manual links was not necessary. You could search and chat with your notes to accomplish your goals. Such capabilities were there (@ in Mem to create a link instead of the [[]] in Obsidian), but were claimed to be unnecessary.
The AI can do it for you is what concerns practitioners in some domains for some purposes. Educators may be concerned that students use AI to complete homework assignments. Writing assignments can easily and reasonably be completed by giving an AI tool a prompt. With writing there are two interrelated problems. As a skill writing needs to be learned, so practicing the subskills (procedures) involved in skilled writing are not practiced when the work is done by the AI. A separate concern is that writing is a way to process the content that can be the focus of the assigned writing task and this processing does not happen when the AI provides and assembles the content. There are counters to these concerns as AI can contribute in different ways allowing some subskills that are involved to be ignored so that others can be emphasized, but this possibiity is making my example unnecessarily complicated.
With note-taking, I think of the argument for what I am calling the manual approach is based on the assumed value of generative cognitive processing. I describe a generative activity as an external task that is likely to increase the probability of an internal (cognitive) process. When proposing an example of a generative activity, I use questions. In theory, connecting new concepts with experiences is an important learning process. Individuals may or may not do this on their own. If I request that they provide an example of concept XXX, it is fairly likely they will think and come up with something. Hence, questions function as generative activities
The organization of notes into folders or categories and the searching for connections to be made permanent with links involves thinking and decision-making that is less likely without the commitment to tasks that require such thinking. These actions may also serve generative functions. While educational researchers have proposed and evaluated many manual processing activities associated with note-taking as part of studying, to my knowledge such research does not exist for some of the procedures recommended by recent, digital note-taking gurus (see an earlier post on the lack of such research). So, unlike the abundant research on the benefits of provided and self-generated questions, the specific activities associated with digital (and manual) note-taking skills are largely untested. This is partly the reason I continue to duplicate my collection of notes within both Obsidian and Mem. Personal experience is a weak research tool, but better than nothing.
This is what I mean by questioning whether the processing requirements of the various note-taking tools strongly contributes to the eventual application. The recent development of systems such as Obsidian and Mem seem more likely driven by the long-term use of information in comparison to what might be associated with academic studying, Purpose and length of the exposure to use processes may be important differentiators. What is interesting about Mem is that it has come out with the argument that AI can eliminate many of the activities focused on and debated by Obsidian users.
Summary
This post attempts to identify and differentiate two note-taking and note-using approaches that can be associated with two specific products. While both systems can now be used in the same ways, the proposed differences are interesting. How important are the manual actions AI can eliminate? I will offer one observed advantage to the AI capabilities that can be applied with either system, with the large collection of notes I have now accumulated, I have found that AI prompts surface useful notes I would not have identified based on the manual links I had accumulated. I suppose there might have been benefit in a continuation of exploration by the use of links, tags, and search, but I must deal with the reality I could not necessarily make the effort. Perhaps continuing to use both and adding links to connections identified by AI makes the most sense.
