Very interesting topic. By image compression, we of course mean Pattern Mining. A totally random sentence or dataset has no re-occurrence of the same symbols, and therefore no pattern nor semantics. You'd never have a solution for any problem if Earth was always random. With images, it is easy to find the same symbol appear (the shade of brightness), same for text - letters and bits, but exponentially rarer to find larger pairs of the same shades. You begin to use similar patch matching technology. Through these 'words', you can glean semantics cat=turtle even if look too different.
It's often said that text isn't rich enough to build AGI. Need vision. Better: 3D block vision. Yet vision has so much noise! And more text = better prediction anyway, I've tested it myself. Could it be that vision does have a much more rich aka more patterns in it than text, and even more noise? In that case vision would be still more rich than text even though it has a lot of garbage. Ex. patterns|noise for text/vision may look as follows: 80%|20%, 20%|80%. A single image may be worth a million sentences, but there may be a lot more storage demand and less patterns compared to noise found per image. Another question is do we have enough text data? AGI will early in need microscopic data, and someone will need to see or feel it to capture the objects. We don't talk about microscopic structures enough on the internet. We do have some pictures though, and a lot of diverse relating text too still. My conclusion is that text will do us fine a fair bit until we need to start describing microscopic images to the AGIs or let them see our images themselves using visual cortices. And I believe 1GB of vision has more noise than 1GB of text, but I believe that that 1GB of vision has a lot of structured data/structures that isn't clearly/intuitively find able in text and therefore that itself is extra data too, and quality data at that. So the question remains: 1) Is vision a pattern-richer environment for AGI than text? Or is it noisier? 2) Does vision have more unique and clear quality patterns than text has? These could be worth more than 4,000 patterns, and if they are what we want, are also worth more than useless patterns. ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T2a0cd9d392f9ff94-Md6a9e5e67d0342c1a040c57e Delivery options: https://agi.topicbox.com/groups/agi/subscription
