Trying to get to this all day. "The greater the generalization of a network, the more entangled its neurons and as a consequence the network becomes less interpretable.". The article says our brain net is too complex to look at and should have it talk to us instead. And cast from the high dimensionality of it more intuitive shadows, like 4D to 3D, like how a 3D block casts 2D shadow.
Well, a more accurate neural network / car / system "model" is smallest and still explains most of the data. It is general / multi-purpose, it solves all sorts of problems because they are the same problem really. To do this, you need to merge patterns, to e-merge. Merging creates hierarchies / clusters of clusters, because whatever you merge can be used for other merges ex. you never store the same word feature more than once, and that is used to create a phrase feature or semantic feature or task sequence feature. And if you had a real life machine, ex. a motor, funnel, brush, arm, bag, electricity line etc, you'd make hierarchies and re-use parts, ex. the vacuum or a hammer or man could do all sorts of things, the man/ hammer could make up a star ship or house or car or hammer or man (man+hammer=new machine), and could have sequences re-used too ex. a man with hammer makes wall, then lifts it up to a story, and combine this with drive off and get more material // or wait while they take it an hour so drive over here and do other job. The most powerful things are the longest living technology and that is one that is small but deadly, like a nanobot advanced squad unit. This entangling is merging, but it doesn't make it uninterpretable. It makes all the data make sense if viewed! That's what it is after all, a model. The reason it seems messy to look at to yous is simple. Take a video, or 3D car, or tree trie like https://ibb.co/p22LNrN, and store it on a computer storage disc and look at it, it'll not make any sense! Unless you know how to view it! Same for DNA, we look at it and see strings, but we aren't looking at it correctly. My AGI is not only 100% understood in how it works/learns, but also after learning the WHOLE network can be observed to see what it's thinking, what it wants to do, what it knows, and how it's task sequences are functioning/going etc. Where you guys are at is this: you don't know how AI works, you just use backprop etc!!! No wonder yous can't read it or know how to make it. *You's aren't making the REAL bread, you're making FAKE bread. I actually KNOW how to make a good NN without backprop.* ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T1b8c0c3b7933a51f-M981812111677bc876b2b514f Delivery options: https://agi.topicbox.com/groups/agi/subscription
