Experiment: Mapping Cognitive Agency Under AI Amplification

Over the past few days, I conducted a small experiment in hybrid concept mapping. The goal was straightforward: take a loose cluster of existing notes from my TiddlyWiki and transform them into a structured map that could withstand scrutiny.  These notes (42 of them) were all tagged "LS-Cognitive". The LS stands for Learning Sprint.  

The working title of the resulting draft map is:

Maintaining Cognitive Agency and Brain Capital Under Conditions of AI Amplification

The process unfolded in stages.

First, I extracted key concepts from existing tiddlers and generated a provisional graph structure using AI (Ace Knowledge Graph, accessible as an app within ChatGPT Plus). This initial pass surfaced nodes such as Cognitive Agency, Cognitive Control, Cognitive Offloading, Human–Machine Symbiosis, Brain Capital, and Brain Economy. The resulting structure was technically coherent but conceptually uneven. This is the first time I manage to do something like this combining my TiddlyWiki content with mapping so I was impressed no matter how imperfect the map was. 

The Ace Knowledge Graph performed three useful operations:

1. Entity Extraction
   It identified recurring conceptual terms across the tiddlers and normalized them into discrete nodes.

2. Relationship Inference
   It generated directional edges based on linguistic and semantic cues in the text. These edges were labeled with descriptive phrases derived from the underlying content.

3. Structural Flattening
   It produced a graph representation that made implicit assumptions visible. Concepts that felt intuitively connected were now linked explicitly.

What Ace did not do was evaluate coherence. It did not ask whether the abstraction levels were aligned, whether certain nodes were redundant, or whether a central thesis existed. It generated a plausible structure.

Next, I imported the structure into CmapTools (my favorite concept mapping software) using a simple tab-delimited text file (other import formats failed). Here again, AI did then necessary format conversions.  The import preserved linking phrases and allowed manual rearrangement. Eliminating crossing lines forced clarification of hierarchy and exposed abstraction mismatches.  The node/concept definitions were manually added as info notes. 

At that point, the real work began. I removed some nodes, questioned ambiguous terms such as “learnability,” and examined whether each concept genuinely contributed to the central thesis. Some terms migrated to side modules. “Aging Well,” for example, remains connected by a dotted line, signaling a future expansion into lifespan cognition rather than a core dependency.  Not surprisingly, the map isn't entirely coherent because the original notes were collected over time without any concern around a central question or well-defined Learning Sprint. They only had one thing in common, they were related to "cognition". 

This map is not intended as a comprehensive theory. It functions as an interrogation device. Each node can be a starting point for further reading and research as well as future elaborations of the map. The map evolves as a record of structured reflection rather than a finished system.

Going forward, this experiment could become a template. Each Learning Sprint may generate its own evolving concept map, versioned and periodically reassessed. The value lies in making the architecture of thinking visible and open to revision.

Evolution of my Concept Mapping Practice

I started mapping using Inspiration, then relatively quickly turned to CmapTools. More recently I have explored the mapping functionalities associated with TiddlyMap.  The tools and their specific functionalities have played a key role in the development of my practice around mapping, and my evolving needs have played a key role in how I sometimes push the tools well beyond their original purpose.  I pushed CmapTools to become a gigantic map of maps, well beyond a sustainable system.  I am using TiddlyMap to mimic the capabilities of a knowledge graph.

In developing my mapping practice, I also acquired a few bad habits. Perhaps they were useful adaptation rather than just bad habits but now I am reconsidering my approach.

Examining even just the collection of maps that I have made public on my site, it is clear that I have used concept mapping mostly as a personal thinking tool. When I have used mapping within organizations, including at NASA and Chemonics, I had to either establish a clear framework (especially at NASA) so that the maps could be easily readable and usable. The NASA mapping approach was presented as early as 2016 at the APQC KM conference.

The connecting words are key to concept maps. They make the maps more readable and clarify the meaning of the connection.  These connecting words become the "predicate" in the world of ontologies and Knowledge Graphs. There is a subject (one node representing a concept), an object (another node representing a concept), and a predicate (connecting words) articulating the relationship between the two nodes.

In my early concept mapping practice, I either omitted the connecting words or used a few key words repetitively (and, such as, including). 

Most of the maps I created at NASA were not really concept maps because the nodes were not individual concepts but rather entire phrases. The unit of analysis for a node was at the higher level of "complete thought" and not individual concept. I called them knowledge maps or KMAPs, and eventually insight maps on my own website.  This allowed me to put much more content on a map than what is possible by identifying and mapping the connections between individual concepts. 

The connecting words and the way concepts are connected becomes more and more important as we move towards more "readable" maps or graphs. In a simple map, the connecting line and arrow may be relatively easy to interpret but that is not always the case and something more standardized is needed at scale. In particular, if we want these types of maps to be machine readable, then a formal ontology with very well defined "predicates" or connecting words becomes essential.  Ontology tools/software also become essential to manage larger ontologies.

I thought it would be interesting to take some of my old maps and see if they can be made more readable with greater attention to the connecting words and a more granular approach to concepts. It's also possible that what I was trying to express with those early maps isn't meant for a real concept map.

The first map I ever posted when I was still using Inspiration as a mapping tool, can easily be improved and turned into a proper concept map or even a knowledge graph. I have now recreated that early map using a more formal mapping technique inspired by ontologies, with specific classes of concepts (visually distinct on the map with different colors) and well-defined, consistent connecting words.  See Map #34 in the Insight Maps collection (and below). 

The original map included the movies' release year in parenthesis.  This type of data can be added as an annotation in the map, visible when hovering over each movie node.  Other properties could be added, including a link to the IMDb record for example.  In the related ontology, the "releasedYear" could simply be a property of the movie class.  

Not every map requires a formal ontology, but understanding ontologies and learning more about knowledge graph is helping me to refine my mapping practice. 

The Concept Mapping label provides access to some related posts over the years.