Cognitive Variability: Towards more Diverse and Adaptable Thinking
InfraNodus text analysis tool is designed to induce cognitive variability: the ability to approach an idea from multiple perspectives, to discover general patterns but also nuances, to alternate between exploring and making sense. A higher level of cognitive variability makes any discourse more diverse and open, helping avoid cognitive bias and totalitarian thinking. Below we explain how it works.
Why Cognitive Variability Is Important?
Too often people get stuck with a certain idea or a pattern of thinking. InfraNodus can be used as a mind antivirus against obsessive loops — biased ideas, mundane patterns, totalitarian thinking, propaganda, and narrow-mindedness — proposing a kind of thinking that is pan~archic: spanning a range of states and modalities.
Using text network analysis, InfraNodus identifies the structure of the discourse through the measures of network modularity, distribution of influence, and narrative variability. It then uses its algorithms and AI to steer this structure toward a more adaptive and open state.
If the discourse is too biased toward a certain idea, InfraNodus will steer it toward diversification. If the discourse is too dispersed, InfraNodus will steer it toward more focus.
In plain terms, if InfraNodus were a conversational partner (which you can experience with the AI Chat App), it would support a conversation for some time, but then also change the subject and go on a tangent towards a different topic. After some time, it will bring the conversation back to the central ideas, stay there for a while exploring the nuance, and then shift the subject to a new topic again.
How Cognitive Variability Works?
When we talk about variability, there are different ways to describe it. As we are dealing with discourse, we should take into account structural, narraitive, and semantic properties.
Semantic variability is the easiest to define: it is the change in the semantic meaning of the words as we shift from one concept to the next. Structural variability can be identified through text network measures such as modularity (which concepts and entities tend to occur in the same context) and entropy (how well the key elements of meaning are distributed across the whole discourse). Narrative variability is less intuitive. As we shift from one concept to the next, we should somehow measure the change that occurs as we're reading a text.
One possible parameter could be word vector distance in a semantic space (using LLM embeddings). This distance, however, would be based on the global context (all the texts that have been written before and that our model has been trained on) and doesn't take the specifics of text into account.
That's why we use the measure of influence (betweenness centrality) of the network's nodes (representing the words). Betweenness centrality shows how important a word is for a specific context. High variability, in this case, means that we're jumping between the important and peripheral ideas. Low variability means that we either remain in the periphery or reiterating the main concepts only. Variability in this measure reflects the semantic distance we have to traverse from word to word as we're reading the text.
Using the approach above, we identify 4 states of variability, which are comprised of 8 stages. Each state and every stage correponds to a certain combination of discourse structure parameters as measured through network science algorithms that we use. These are:
- • How densely connected the discourse is
- • How modular it is
- • Concentration of the main ideas in the main topical cluster
- • Distribution of influence across the discoursive network clusters and nodes
- • Degree distribution
- • Dynamics of influence in the narrative
To represent these 4 states and 8 stages we plot a two-dimensional graph. On the X axis is the spectrum of intent, on the Y axis is the spectrum of scale. The intent can alternate between exploration and focus. The scale can alternate between small scale (zooming in) and big scale (zooming out). As a result, we get 4 possible combinations of intent and scale. Depending on our position in the graph, we can distinguish more states. Some of the intent / scale parameters are also interchangeable. For example, across the intent scale we can also think in terms of construction and deconstruction.
In this representation, we can see 4 states of structural / semantic variability that define the state of the discourse:
- biased
- focused
- diversified
- dispersed
Biased state is characterized by a high concentration of ideas in a few clusters, low modularity, and high degree distribution. Focused state is characterized by a high concentration of ideas in a few clusters, high modularity, and low degree distribution. Diversified state is characterized by a low concentration of ideas in a few clusters, high modularity, and high degree distribution. Dispersed state is characterized by a low concentration of ideas in a few clusters, low modularity, and low degree distribution.
To quantify narrative variability, we use terminology used in DFA methodology (detrended fluctuation analysis), which is commonly used for identifying HRV (heart-rate variability) in medical science. This is a robust approach to providing quantifiable data on variability and it also distinguishes four different states (uniform, regular, fractal, complex), which correspond to four different types of variability (repetitive, stationary, self-similar across scale, and non-stationary).
Cognitive Variability Stages: from Bias to Dispersion
Now that we presented the theory behind the approach, here is how it is implemented into InfraNodus, which identifies 8 stages of discourse evolution (indicated with numbers on the scheme above).
1→2 #bias #growth #vector #exponent:
If the discourse is too biased toward a certain group of concepts (stage 1 to 2 on the graph), InfraNodus will highlight the less represented ideas and propose you to explore them and to connect them (50%/50% explore / focus ratio and 50%/50% zoom in / out ratio at the stage 2).
2→3 #focus #saturation #plateau:
This will bring the discourse toward a focused state (stage 2 to 3). Higher connectedness, focus on multiplicities, increasing the scale, zooming out, linking ideas more than exploring (80% to 20%).
Now we can start zooming in again, decreasing the scale, connecting ideas within a smaller area, deepening the focus inside the existing structure (stage 3 to 4 on the graph).
4→5 #diversification #assimilation #release:
Once the discourse has become even more "focused" (stage 4 to 5), InfraNodus will also suggest to develop the specificities, zoom in further, go bigger scale, and, thus, reduce (global) focus and increase the proportion of exploration again (to 20%).
It will then propose to diversify the discourse by zooming out and exploring more clusters, developing each of them further (stage 5-6: moving between locally related ideas and also jumping across the structural gaps in the graph).
6→7 #dispersion #reorganization:
At some point, we shift toward dispersion by giving more weight to the exploration process, rather than focus (state 6-7: dispersion).
Once the scale is large enough, we zoom into the small scale again while exploring until we find a new concept we'd like to develop (stage 7-8).
8→1 #bias #genesis:
We then focus on it and develop it further again (stages 8 to 1).
Webinar on Cognitive Variability
In this webinar, we explore the concept of cognitive variability and how it can be used to enhance your thinking and creativity.
Demo: Cognitive Variability for Ecological Thinking
In this demo, I show how cognitive variability promotes ecological thinking and mind-viral immunity.
Try It Yourself
To try cognitive variability on your own ideas or texts, please, sign up for an account or log in if you already have one:
Sign Up Log In Join Our Team