22 comments

[ 5.1 ms ] story [ 63.9 ms ] thread
This is going a little over my head. Can anyone with more expertise help summarize / ELI5?
My undergrad is in Neurobiology, but I don't work in the field, but I can give you my understanding.

This article is basically an article that reviews the current theories regarding the neuroscience of intelligence. It's saying that there seems to be evidence of 'g' (which you could call IQ, but is the variance in cognitive abilities) that dictates the efficiency of our brain as a network of networks. It describes the brain as a interconnected global network of local networks that have discrete responsibilities. The reason these local networks to handle specific things is because it's more efficient to process in close proximity. And that the communication between these 'nodes' and the ability to tap into stored memory and intuition is described by 'g'.

like capsules?
more like flexible cognitive performance depends on the neurons in the brain being able to rapidly and efficiently re-organize into segregated task-relevant functional networks.

A good workplace analogy would be a large team that can flexibly reorganize itself in a task-relevant way.

That matches what I sussed out growing up with a high functioning older brother and friends of his, also special needs. Some things they understood or could do easily as anyone else. Others not so much. interestingly one of my brothers friends could spell and write perfectly, way above average for his age.

My assumption was some parts of they brains didn't develop normally which made it much more difficult for them to learn certain tasks. I've also run into people that have other deficits, friend didn't drive because of spacial deficits. But had a PhD in math. Bonus my brother drives.

their theory subdivides the brain into "local" networks, which are tightly connected and relatively stable, and "random" global connections, which connect various local regions in a more unstable manner. they propose that there are two kinds of intelligence, crystallized / persistent intelligence, and fluid intelligence. their perspective is that intelligence is governed not by specific regions of the brain, or specific networks, or specific overlap between networks, but by dynamic reorganization of various connectivity networks

note that this is an opinion article explaining an emerging theory. the theory / field of study is based in computational network models based on data from fmri and diffusion tensor mri and other methods. they do not focus on cellular or molecular biology, genetics, etc. their perspective focuses on brain organization in terms of network efficiency

Graphs (networks, webs) are a data structure we have not exploited enough. Tons of research is going into using graphs for artificial intelligence, and for understanding human intelligence (as in this article). But what about using knowledge graphs to augment human intelligence, like a prosthetic?

This is the power of Google search. It uses a knowledge graph that models the world (with an emphasis on the internet). The graph is big, but the view of it offered to users is minuscule -- in part to keep the interface as simple as possible, and in part for economic reasons.

There is open source software that lets people keep their own knowledge graphs. In Semantic Synchrony [1] you can keep a knowledge graph and merge it with others' knowledge graph. Joshua Shinavier (who wrote Semantic Synchrony) and I share a graph with over 400,000 nodes, and most views load in the blink of an eye.

A sister project, Digraphs with Text[2], offers a more flexible system of expression: It generalizes the graph, allowing relationships to involve more than two members, and allowing relationships to be members of other relationships. It also offers a search facility very much like natural language: To search, for instance, for reasons neurons need vitamin B, you would use a query like "(neurons #need vitamin B) #because /it". (The # mark indicates a joint between members of a relationship.)

[1] https://github.com/synchrony/smsn/wiki/ [2] https://github.com/JeffreyBenjaminBrown/digraphs-with-text

The "-omics" fields and neuroscience have started exploiting graph structures to describe metabolic interactions, gene regulatory networks, and organic neural networks, among other things. fMRI imaging, for example, can be described as representing a temporal network of oxygen levels in the brain.

The idea of creating novel purpose-built graphs and embedding ourselves in them, in addition to modeling existing networks using graph structures, is interesting. I think social media and other networking technologies are the most current examples, and the first decade of the internet was probably the best example of an intentionally-constructed dynamic knowledge graph.

wow thanks for this, making something like this is what I consider my life's work but I had never heard of these projects.
Thanks for the links. Will experiment with these.

I have been building a personal knowledge graph based on concept maps[0] using cmap tools[1].

[0] https://en.m.wikipedia.org/wiki/Concept_map

[1] https://cmap.ihmc.us

It would be interesting to visualize a SmSn knowledge graph as a Concept Map. I feel that text buffers are best for viewing and editing a graph when you are seated at a keyboard, but graphical views have their place, as well.
Are you sure that Google uses an explicit knowledge graph, and not an implicit one (derived from all web pages combined)?
Google has an explicit knowledge graph. I saw it at a meetup in LA once, for a minute; then the guy showing us felt guilty and put it away. It looked like RDF, with a tree of predicates. Each predicate was written in the form "wordOrPhrase/wordOrPhrase/wordOrPhrase/...".
(comment deleted)
This would actually indicate that spearman's g cannot distributed ~Gaussian. CLT wouldn't apply, anyhow, but positive feedback effects (a la small world RG) would apply. Very SFI sort of thing
A psychology professor once said to me that throughout history our theories of the mind tend to analogize the dominant research paradigm of the time. At one time it was chemistry, then physics, and now it's computer science.

I write that because it suggests that our theories of mind depend our own perspectives to a great degree - perhaps in their conclusions, or in how we describe them, or in our choice of research. (I wish I could remember the chemistry or physics analogies ATM.)

It is possible that the successive paradigms are objectively more powerful. So, a matter of perspective, but not perspectives of equal value, and also not diminished by the expectation of future paradigms we don't have yet.
That's exactly the first thing I thought as well.

This article is relevant: https://aeon.co/essays/your-brain-does-not-process-informati...