When I say unique/ clear/ what we want at the end of my post above, this would be because of Large Structures in an image(s). They are rarer to be described in text and are fully arranged in an image with a large structure. It makes images, which may be more noisier than text, worth as much as text.
I'd say use text for now (images are just as good but harder to handle), and make sure to describe microscopic images to your AGI in full detail. Later they can see them for themselves when need larger data throughput than humans can provide. They can tell us how to give them vision. Humans made up text to describe vision. And we know vision is the engine for discovery, much faster than trial and error using motors (that's primitive AI). 1D text is like vision, but we do use it to describe vision, meaning we say what is around each other or what comes next or what is what. When we say those things, maybe we should treat them as not a sequence but a single object that is a sequence. For example I say "The cat was eating the other cat's food and [the cat has water around itself with a party hat on its head and a cigarette sticking out of its mouth while wearing pink pajamas], and decided to collapse to the floor and finish eating its catnip.". This large section in [] can be excluded from a match and still be a match. Though maybe this is just local similarity segmentation ex. "the file '7445785656853467474347547' was sold on" Or "The article was trash, bad review here, didn't like it, coffee was too good, loved it, and to top it off the clerk fired me, best place ever. The shop next door was great, good food, good time, nice hours.". There's many of these, don't call it good/bad review classification (i forget the word...). Anyway, images and text share a sequence of flow in time - changes to structures, and a deeper description of parts of that sequence, nesting. We must segment properly, a object description is of the same thing and is a pattern therefore. ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T2a0cd9d392f9ff94-Mc43ab2e268db7bf805ee1112 Delivery options: https://agi.topicbox.com/groups/agi/subscription
