🌿  back

20.96 YRS

conducting an ideoscopy

Just over two months passed since I started using the conceptarium as a storage medium for my thoughts and as a source of serendipity in my workflows. Earlier this week, I also released the first stable version of the ideoscope, which makes today a perfect moment to do a first deep dive into the cognitive analytics derived from my thoughts. I’ll first list a few a priori hypotheses about how I think my thought process has been like in the past weeks before firing up the ideoscope. After that, I’ll go through each section and subsection of the dashboard, test my hypotheses, and derive new insights. In short, I aim to conduct my first ideoscopy.

a priori hypotheses

There are several things which I think happened to my thought process recently which I’m curious whether they will show up in the stats and visualizations. First, I implemented some experimental reflection practices in the afternoon. Those are sessions defined by an overarching question prompt (e.g. How can Kohonen self-organizing maps be applied in the real world?) for which I try to come up with speculative answers while making use of the conceptarium and other tools. I gained a bunch of ideas and insights while engaging in those sessions, and, of course, I also saved them into the conceptarium. I expect this to show up as elevated memetic birth rate levels in the afternoon.

Second, I experimented with a tweak to my informational diet which could be described as informational fasting. Basically, I massively reduced my social media usage to around one hour per week, and increased my exposure to my own thoughts through the conceptarium. This exercise has been in preparation for my upcoming project, which will focus on quantifying the nutritional value of online content as food for thought, in an attempt to optimize my informational diet. I essentially cut down on haphazard content consumption to make space for some more informed “meal prepping” in advance. Anyway, I think that this exercise has caused a reduction in the memetic variability and drift I experienced, because I wasn’t nudged that much in this and that direction by external outlets, and therefore was a bit more stationary and single-minded than usual. That, however, might have changed in the past week or so when I fell down the unlikely and irresistible rabbit hole of modular synthesizers. For context, modular synths are highly customizable instruments for creating electronic music using (mostly) analog sound effects.

Third, I feel that I got more comfortable with saving thoughts to the conceptarium, I adopt it more and more in my thinking. This means I roughly settled on a particular writing style for my entries, with slightly longer entries than before. I expect this to reflect in a stabilization of the metrics under the linguistics section. I might have also been more excited about how promising the conceptarium-related workflows seem to be, so the general sentiment might also have gone up a bit. Again, those are just thoughts I currently have about my recent thoughts before seeing the latest data. Time for that now.

The top-left panel suggests that there are days when dozens of ideas are saved, while there are also days when barely no new ideas are saved. Interestingly enough, it seems that when there are new ideas being generated, there are many of them, tiny cascades of insight around certain pockets of conceptual space. The top-right panel indicates a peculiar pattern: on Wednesdays and in the weekends, the birth rate of ideas is way lower. This might be explained by the fact that those days consist of more chores and errands being done. For instance, on Sundays I mostly go through what I call the accumulator, a batch of tiny meaningless tasks collected through the week. On Wednesdays, I currently have a lot of onsite activities at uni, which means quite some getting around to different places and working on (not extremely exciting compared to thoughtware) team projects.

The bottom-left panel does indicate a bit of the tendency I expected regarding the reflection practices, somewhat elevated levels of memetic birth rate throughout the afternoon. The bottom-right panel combines the weekday view with the hourly one into a weekly-planner-like snapshot. This one seems to reiterate some of the previous takes, while also indicated somewhat elevated levels around online lectures at the beginning of the week earlier in the day. Being challenged with new problems and possibilities inevitably leads to new ideas.

The left panel here simply shows a cumulative view of the birth rate by day discussed above. Somewhat predictibly, in the first days of using the conceptarium,I was overly excited about saving ideas in it. My intuition about which thoughts deserve to be persisted and which ones should just be let go wasn’t that well calibrated. That, however, started to change in the following weeks as I developed more sustainable heuristics for classifying ideas as worthy of defeating death. What’s more, the general slope of the population size curve appears to be slowly increasing lately, which might mean that becoming more fluent in exploiting my knowledge base is starting to influence my thinking. The population pyramid on the right indicates that most of my most active thoughts are ones which have been created a longer time ago, a pretty “aged” population of ideas.

As predicted in one of the a priori hypotheses, there was a pretty low plateau both for memetic variability and drift in the past weeks, which likely reflects the more inward thought patterns about my current approach to thoughtware. I haven’t changed my focus much (i.e. drift), and haven’t considered very diverse things at once (i.e. variability), compared to a few weeks prior. However, the promising rise in the past week might be explained by the unlikely detour into the fascinating world of modular synthesizers. When looking at the top-right panels for a larger timescale, the same tendency is amplified and more evident. To bring up the memetic variability and drift, increased exposure to curated online content might be a sensible intervention.

Both panels in this subsection indicate that the fitness distribution of ideas in the memetic ecology of my mind happens to be quite spread out, without having an obvious modality. What the meaning of that is beyond this superficial take, I’m not yet sure. The general fitness values are pretty compact, without any significant anomaly to tip everything off. This is encouraging in that the serendipity search parameter of the conceptarium can be used quite reliably, without expecting the activations of individual thoughts to render the search results chaotic. Additionally, it seems that there’s a right skew towards higher fitness values, which, together with the low memetic load value, might mean that I haven’t invested a lot of time into ideas which I’m now not making use of by thinking about them. This, however, might also indicate a too exploitative and not explorative enough foraging strategy through the space of online content.

Focusing solely on the language thoughts now, there’s a steady rise in the length of individual items, as measured through the estimated reading time of individual entries. Fortunately, the values seem to be leveling out. Readability also seems to stabilize lately, reaching an estimated reading level which is just above highscool graduates based on the Flesch-Kincaid scale. Together, those metrics might indicate the “maturity” of a certain writing style for conceptarium entries added in the moment. For me, those metrics do not lend themselves to particularly useful targets to tweak at the moment, but it’s still interesting and fun to perceive the development of my saving behavior this way.

The top-left panel following objectivity by week also suggests a stabilizing phase, because the metric is based on the use of certain keywords which tend to indicate whether a piece of text is all dry factuality or rather a more polarized opinion. This is also why at a broad level objectivity and sentiment seem to be inversely related. If I’m overly excited about a thing, I’m likely losing out on objectivity. If I wanted to tweak my thinking to be more rational and scientific, I might aim for higher objectivity scores in my thoughts. Conversely, if I aimed for going directly orthogonal to such ideas, I might head otherwise. At the moment, those are merely interesting affordances, and it’s exciting to now that such landmarks and references are available.

Zooming in a bit on the timeline of my interests, you can clearly pick out the last “modular synth” phase, preceded by quite a few days of thinking about reinforcement learning. Going even a bit earlier than that, around halfway through September, I seem to have thought a lot about basic linear algebra (e.g. “state cloud”, “principal component”, “orthonormal eigenvector”), which makes sense – that’s when I started working on conceptors with my thesis supervisor. It feels pretty powerful and novel to be able to see at a glance what I’ve been thinking about over time, even if the keyword extraction methods used here are pretty rudimentary.

Even more satisfying, in my view, is to see clusters of ideas taking shape in the low-dimensional projections. For instance, hovering over the dots on the left panel reveals that the bottom cluster is composed of many things Origami, while the cluster on the very right is made up of all things Dune and space opera. In contrast, the somewhat fuzzier cluster on the left is composed of ideas about linear algebra, Bayes, and lambda calculus, while the shy cluster taking shape in the top-right side of things is made of new ideas related to modular synths. The 3D projection turns out to be more difficult to navigate intuitively on a 2D screen. Some tweaks of the t-SNE algorithm might improve the results, but walking through the point cloud in VR might truly enhance the experience.

And so we get to the last subsection of the ideoscope, which also happens to be one of my favorites. It shows you just how much thought there is to explore out there, how many exotic ideas there are to grasp. It achieves this by offering you stats on the proportion of the whole space which you “conquered” through your ideas up to this point. The infinitesimal slice of the pie in the left panel acts as an antilibrary, reminding you just how much there is to know and learn out there, making you a bit humbler. The panel on the right, to which I resorted due to the computational complexity of sampling high-dimensional semantic space precisely, simply shows across how many dimensions of semantic space your thoughts are spread. If your collection of ideas would have been really complex, intricate, and going strong across many dimensions, many directions would be required to define the aggregate cloud. However, for me currently, the first principal component already explained more than a third of the variance in the data, which means that my ideas are spread along a tiny sliver of semantic space. A strong motivator to keep learning and reflecting on new learnings.


All in all, actually seeing those stats come to life and reflecting on what they indicate has been a really exciting and rewarding exercise. However, way more refined, rigorous, and structured procedures for interpreting such metrics is needed if they are to actually help inform workflows. I hope that several future iterations of the ideoscopy will help refine the process from a somewhat disorganized first glance to a more targeted analysis.