For a while I’ve thought the linear form of information ingestion we experience today and have experienced for a long time (from the papyrus scroll to the webpage) is somewhat outdated. Information in the form of chunks of text stacked on top of each other may be easily human-readable, but this form does not contain the relations among the information besides the order in which they appear. Thus information as it is presented in any given article or book is preserved simply as an array of paragraphs/chapters/sections.

If we take each of these arrays as a node in a graph then the edges are citations among these different sources. From this is developed what might be called a tree of knowledge, or more aptly a knowledge network.

Each of these nodes, however, contains a wealth of knowledge. One book, one work, that the author may have presented as linear in its form may be not linear in its relationship to itself, meaning that each of these nodes should actually be understood as its own graph. If the nodes are graphs then they aren’t nodes at all but just more parts of the larger knowledge graph, and so we must reduce further to find a new node of information exchange.

In the digital sphere we are no longer constrained by the physical limitation of exchanging information book by book, article by article, but we could indeed atomize this exchange down to chapter by chapter, paragraph by paragraph, in some cases sentence by sentence, whichever form contains a fully developed and exchangeable thought.

So long as proper labeling is maintained, each of these information nodes can still be understood and ingested in their linear form; however, this new method of storage allows for a much more dynamic set of options for both ingestion, presentation, and collective understanding. These nodes can be connected, analyzed, or consumed via their semantic similarity, topically, via keyword or n-gram, etc. not just in their relation to other nodes within some given work, but in relation to all other nodes.

Wikipedia indeed has its own knowledge graph that contains the relations among various encyclopedic entries, a truly herculean effort to categorize every term, event, place, notable person, and genus into its proper place. However, what I am envisioning is not a network of encyclopedic entries but the totality of human thought. (I use the vague term ‘thought’ because how this information is stored is malleable, the most traditional form being thought composed [text], but could just as easily be thoughts depicted [images], or perhaps in the not so distant future thoughts as measured by neural activity [in which case we would truly begin to see knowledge as digital hive mind]).

Although I cannot say that the above vision was the impetus for the project I present below (it was in fact my having to quickly study for an exam), what I have created is a minor effort I completed in the spirit of the ideas heretofore presented.

The initial idea I suspect came from an Elon Musk quote I stumbled upon somewhere on X and likely thought little of at the moment, but for whatever reason stuck with me:

It is important to view knowledge as sort of a semantic tree — make sure you understand the fundamental principles, ie the trunk and big branches, before you get into the leaves/details or there is nothing for them to hang on to.

Now, the idea as knowledge as a semantic tree is not necessarily a new one, and seems, in fact, quite obvious to us now as knowledge graphs are what power most digital experiences to date. In typical fashion, Musk in this quote seems to deride those sorts of academics or traditional intellectuals who bury their heads in the sand of unrelated and arcane facts, which they have been forced to memorize for various forms of sophisticated regurgitation whether it be when taking an exam, writing a thesis, defending a dissertation, or lecturing to some pitiful undergrads…but perhaps that is merely my reading and Musk only means to say detailed facts are only as useful as their relation to other facts or, better yet, as parts of larger ideas.

Where my mind went upon reading this quote was that forming of this semantic graph while combing through all of these linear and often disjointed stores of knowledge takes a great deal of effort, and that perhaps it might be better for this graph to be already somewhat established, and then one need only ingest and learn this information via the rails of semantic relations.

So I began by creating a series of scripts that take in linear text (a book for instance) and split it into chunks. Each chunk is compared to every other chunk using cosine similarity (between the embeddings of each respective chunk) as well as a multi-n-gram analysis. If the combined score of these two methods is greater than some arbitrary threshold then an edge is created (in this case I use a directional edge…since the initial for loop runs in sequential order relations will mostly be from sequentially earlier nodes to later ones meaning that a degree of the linearity from the original document is preserved…much like a tree still has a degree of direction).

With this, one can then walk the graph and thereby read the book via semantic branches. In this case I had an AI assistant accompany me through the journey so each chunk of text I read is simultaneously explained and elaborated upon.

What I have here is merely a prototype. Much more sophisticated methods are required to extract the semantic relationships within a given text. Still, no matter the sophistication of the methods, the endeavor will always be a subjective one. What would be best is if these relationships did not need to be extracted, but rather if the way that information is composed was non linear. If this were the case, authors would compose books not by chapter or journalists by paragraph but rather they would begin with some general topic or question as a root and then branch out into many non-sequential and perhaps overlapping directions of thought. Such a method, I think, would make both the creation and ingestion of information to be far more entertaining and fruitful.

For a while I’ve thought the linear form of information ingestion we experience today and have experienced for a long time (from the papyrus scroll to the webpage) is somewhat outdated. Information in the form of chunks of text stacked on top of each other may be easily human-readable, but this form does not contain the relations among the information besides the order in which they appear. Thus information as it is presented in any given article or book is preserved simply as an array of paragraphs/chapters/sections.

If we take each of these arrays as a node in a graph then the edges are citations among these different sources. From this is developed what might be called a tree of knowledge, or more aptly a knowledge network.

Each of these nodes, however, contains a wealth of knowledge. One book, one work, that the author may have presented as linear in its form may be not linear in its relationship to itself, meaning that each of these nodes should actually be understood as its own graph. If the nodes are graphs then they aren’t nodes at all but just more parts of the larger knowledge graph, and so we must reduce further to find a new node of information exchange.

In the digital sphere we are no longer constrained by the physical limitation of exchanging information book by book, article by article, but we could indeed atomize this exchange down to chapter by chapter, paragraph by paragraph, in some cases sentence by sentence, whichever form contains a fully developed and exchangeable thought.

So long as proper labeling is maintained, each of these information nodes can still be understood and ingested in their linear form; however, this new method of storage allows for a much more dynamic set of options for both ingestion, presentation, and collective understanding. These nodes can be connected, analyzed, or consumed via their semantic similarity, topically, via keyword or n-gram, etc. not just in their relation to other nodes within some given work, but in relation to all other nodes.

Wikipedia indeed has its own knowledge graph that contains the relations among various encyclopedic entries, a truly herculean effort to categorize every term, event, place, notable person, and genus into its proper place. However, what I am envisioning is not a network of encyclopedic entries but the totality of human thought. (I use the vague term ‘thought’ because how this information is stored is malleable, the most traditional form being thought composed [text], but could just as easily be thoughts depicted [images], or perhaps in the not so distant future thoughts as measured by neural activity [in which case we would truly begin to see knowledge as digital hive mind]).

Although I cannot say that the above vision was the impetus for the project I present below (it was in fact my having to quickly study for an exam), what I have created is a minor effort I completed in the spirit of the ideas heretofore presented.

The initial idea I suspect came from an Elon Musk quote I stumbled upon somewhere on X and likely thought little of at the moment, but for whatever reason stuck with me:

It is important to view knowledge as sort of a semantic tree — make sure you understand the fundamental principles, ie the trunk and big branches, before you get into the leaves/details or there is nothing for them to hang on to.

Now, the idea as knowledge as a semantic tree is not necessarily a new one, and seems, in fact, quite obvious to us now as knowledge graphs are what power most digital experiences to date. In typical fashion, Musk in this quote seems to deride those sorts of academics or traditional intellectuals who bury their heads in the sand of unrelated and arcane facts, which they have been forced to memorize for various forms of sophisticated regurgitation whether it be when taking an exam, writing a thesis, defending a dissertation, or lecturing to some pitiful undergrads…but perhaps that is merely my reading and Musk only means to say detailed facts are only as useful as their relation to other facts or, better yet, as parts of larger ideas.

Where my mind went upon reading this quote was that forming of this semantic graph while combing through all of these linear and often disjointed stores of knowledge takes a great deal of effort, and that perhaps it might be better for this graph to be already somewhat established, and then one need only ingest and learn this information via the rails of semantic relations.

So I began by creating a series of scripts that take in linear text (a book for instance) and split it into chunks. Each chunk is compared to every other chunk using cosine similarity (between the embeddings of each respective chunk) as well as a multi-n-gram analysis. If the combined score of these two methods is greater than some arbitrary threshold then an edge is created (in this case I use a directional edge…since the initial for loop runs in sequential order relations will mostly be from sequentially earlier nodes to later ones meaning that a degree of the linearity from the original document is preserved…much like a tree still has a degree of direction).

With this, one can then walk the graph and thereby read the book via semantic branches. In this case I had an AI assistant accompany me through the journey so each chunk of text I read is simultaneously explained and elaborated upon.

What I have here is merely a prototype. Much more sophisticated methods are required to extract the semantic relationships within a given text. Still, no matter the sophistication of the methods, the endeavor will always be a subjective one. What would be best is if these relationships did not need to be extracted, but rather if the way that information is composed was non linear. If this were the case, authors would compose books not by chapter or journalists by paragraph but rather they would begin with some general topic or question as a root and then branch out into many non-sequential and perhaps overlapping directions of thought. Such a method, I think, would make both the creation and ingestion of information to be far more entertaining and fruitful.