Every time you see your cat - or a new cat, they look similar but never the 
same image is fed into the camera. The brain averages/Mixes these to create a 
concept node "cat" that looks in between all seen cats - like in text there is 
only a discrete feature 'cat'. If some or one is common however, you save the 
episodic memory of a more specific image or even the exact image if you repeat 
it in your brain.

Same for movies, you may see a series of images of someone walking, but 
different each time... so you create a concept movie node that is in between 
all seen walks.

For a single image, a cat is made of face, body, tail, legs, and a face is made 
of eyes etc, so we build a concept node for each feature up the hierarchy, 
concept nodes made of concept nodes.

Even text is continuous, you'll see the same thing in different ways ex. "I was 
walking my cat", "I walking was dog my", "my dog I was walking". In vision, 
layer 1 is all shades of pixel brightness ex. 0.0, 0.1 ... 1.0, same for text: 
vocab, but the way they fit together varies.

Look at this below, if a brain sees this image feature "line", it can recognize 
it up-side-down because all pixels are the same if we ignore order, and if we 
look at all pairs of 2 pixels we will see they are only 1 pixel over too much, 
same for 4 pixel groups. They are relatively close.
https://ibb.co/hmsXHn3


The more times the brain sees a feature the more stronger the neuron 
connections become. If we observe:
cat ate
cat ate
cat ate
cat slept
Then if see "cat _", we predict the more frequent future *more*, because more 
often it is the future (75% of the time in this example).


The brain is all about merging patterns. You use vision + sound to predict what 
will occur next or what something is. For example the image will look a bit 
like a cat node, and sound will weight in on it more because both are linked in 
close time.
------------------------------------------
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T181e21f4aba061f3-M6ae0d808a86c5bc17c1891e8
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Reply via email to