Content Architecture
I have lots of papers/books from different authors all relating to systems thinking / complexity / creativity / ideation. The search “systems thinking” OR complexity OR creativity reveals 40045 entries of which 1480 are pdfs. (Technical Note - these are all full text indexed with a product called Recoll so I can extract and copy sub-sets of the full collection according to some search/selection criteria).
These different papers present rather different ideas (and principles) on common topics which mean the ideas in the papers (ie their intrinsic methods) give different approaches in detail and more significantly in principle of how to do things.
The collection can be easily ‘sliced’ into collections from one author, for example the search term ”^ stafford beer”p200 ext:pdf produces a collection of 55 of Stafford Beers works including his books and including Espejo & Harnden book on VSM. The collection is not the definitive collection of Beer’s work but I hope a wide and eclectic representative sample. (On the website you can see this list by clicking on the right arrow - > - under the author under voices see https://tog.idok.me/Voices/Stafford-Beer)
(Note: In Recoll syntax ”^ stafford beer”p200 looks for Stafford Beer within 200 words of the start of the document. This is a good proxy for ‘scientific papers’ where the author appears early in the document or in the abstract. As a random observation less than 10% of pdfs seem to have the metadata filled with useful information such as author name even though this is a long established pdf capability).
AI Architecture
With these ‘representative’ collections from different authors we can use Google’s NotebookLM technology to produce all the LLM search/summary/structuring/re-writing facilities but LIMITED TO THE SOURCE MATERIAL ONLY. This limitation is vital instead of a wide search the outputs are specific to the collection and different collections produce a nuanced answer FROM THEIR OWN PERSPECTIVE to a given question. For example I enclose the TOG answer and James Wilk’s answer to the question “what is complexity and how can you deal with it”. These are at the website for example start with the What is Complexity index or try examples such as the response from Geoffrey Vickers or Derek Cabrera. It even extend to provide an answer from the unstructured notes from a discussion group MOM.
The important bit of this is that by keeping the authors separated as different NotebookLM sources we do not get the destructive mixing and dilution of using all the different perspectives ‘all at once’. It is an architecture which maintain the essential difference and distinct point of view without it being subjected to some consensus or averaging process - a process which because of its statistical nature often/always produces is driven by an artefact such as repetition/use of jargon/popularity/groupthink rather than the distinction and merit of the ideas.
Only once we have the different perspective developed in each author’s niche ideas (ie their POV or perspective) we can risk any combination, this is basis of summaries such as ‘compare & contrast’ the different guidance on complexity etc but using as its source material the answer to the same question to maintain the distinctive qualities & aspects of each Point of View.
In technical terms it is a crude ‘hand cranked’ version of the ‘holy grail’ of agentic AI but seems to work well here.
Implementation Architecture
The website is based on Obsidian a local notebook perfect for this sort of thing and the basis of a website with lots of automated navigation such as hyperlinks and the ability to slice different authors etc etc. The current site has the different authors under Voices as in https://tog.idok.me/Voices/Alicia-Juarrero also included here is the introduction produced by NotebookLM of all the papers loaded for each author’s niche.
In terms of progress I ‘hand cranked’ the site as it is but then ran out of NotebookLM queries (there is a daily limit) so I have had to change account and reload all the content. This I can automate but is it still getting back to where I was.
Future …
If I (or you) can think of some other principled queries - such as -
- what is the role of the observer and how is this managed
- how are boundaries defined and how is the distinction between the system and the environment maintained
- where do models and metaphor fit with the thinking, how can these be verified and explained
- how are separate perspectives maintained in this approach and are there any techniques to combine or reconcile different perspective
- how do you deal with different timescales of operations, tactics, strategy
- etc etc
Of course other sources and authors (Voices) can be included if I have a collection of their material.
Bugs and features …
The first thing I noted was the sensitivity to the source materials. For example I made a mistake on data load and included a Boyd OODA paper and a Cynefin paper in the Dettmer (Theory of Constraints) collection of MANY documents.
I was somewhat shocked to see the references to Cynefin in TOC until I spotted the error - clearly the use of complexity is a carpet bombing technique in Cynefin which obscures the weaker more nuanced signals in the collection. This is something to be aware of and a vivid demonstration of why a statistical consensus is dangerous (the point made earlier about using niches).
In some cases the answers to the questions for a collection might be fairly ‘weak’ - ie Cynefin bangs on about complexity, Vickers mentions it only vaguely. But by keeping the niches separate we strive to maintain a balance. There was one response in the source material - Pirsig - https://tog.idok.me/What-is-Complexity/Robert-Pirsig - which recommended Replacing “Cause” with “Value” to indicate preference in a complex system.
The techniques consider all the author’s papers equally with no weight given to more recent sources, this may be significant if people change their minds or evolve their thinking.
Hull …
When you get to grips with the wealth of detail and content a few key themes which are related to principles emerge. For example Wilk dismisses models where so many of the other authors adopt a model even before starting.
This is a ‘fork in the road’ which - if you don’t appreciate different perspectives - will be decided almost before you even start. Anything connected with MC Jackson presupposes models and systems - you are down the rabbit hole before you have time to think
And finally … and in response to Geoff - I could try any sources with the same technique but there is so many sources I am not sure where to start.
And postscript - if/when this website develop and there are more enquiries added we could make it a ‘thing’ used to write a paper or two and as an established service for TOG (The Other Group). I loved the way the LLM picked up “Mastering the Muddle” which seemed to capture the emphasis and focus on principles perfectly.