The best text compressors are neural networks that (my own does half of these!) group words dog/cat/animal, count frequency, learns segmentation/byte pair encoding, learns online by talking to itself and storing what it says to itself, have weighting functions to mix activations, activation functions to squash output, are adaptive to local and global by using "energy" left over in the network, sort/organize articles to group stationary data (find and create patterns) modeling the writers wisdom so to predict as bad as he is (I think the predictor can be very smart, but to predict well in the enwiki9 it'd need to predict poor answers if it sees poor context!?), and more! Commonsense reasoning too.
So maybe it will become AGI. However let me tell you, AGI isn't that complex like you think where it needs "more" than a compressors does. It doesn't "need" a body so much, because the future we want comes on its own based on past context, which is what compressors do - predict based on past context, and a better compressor will use reward for predicting so it steers the prediction to a likely one - about food, money, sex, games, AI, or shelter, which is what text/images will probably talk about more than not. The compressor doesn't need to do interactive R&D in a lab and revise its theories, if it has enough data then it can analogate everything and get loads of free virtual data ex. cats eat dogs eat cats run dogs _? Dogs run. Now only it needs us to enact it out, it can tell us how to cure cancer simply. As said we need nanobots, so it will first scale its data/brain/arms/eyes, and add the many capabilities computers have like perfect recall and cloning adult brains and thinking faster. My point is, we all seen the walker bots that learn to walk or crawl, and they don't do much, they don't think about other problems not in front of them, they only see the room around them. We DON'T want a walker bot ffs! We want a thinker bot lol. No body, at first. It will generate many future discoveries like Einstein. We don't need a wanker bot to get better at hand motion, we need a thinker/planner. Thinking deals with long term, direct stimuli is useless almost. Anyway, Earth is just data/context - all AGI can do is use the past current state data to predict/generate the future, so focus on looking at past context and making decisions! Prediction! It's a weighting function, based on probability. And our survival desires are part of te probability - the next likely word is food/sex or the related ex. house money jogging (data collecting, game playing). Text compression we are doing is actually neural net modelling, prediction based on context. The compression is only happening when the Arithmetic Coding is slapped onto the output of the neural network, which almost perfectly encodes it into a small number very similar to Huffman Coding which was the algorithm in use before AC came out. Yes, AGI needs to listen to Q/A inputs, chat, answer yes no, find content in memories far back, .... but it's all Prediction, based on Attention. The context is weighted on using Attention - certain words, certain sensory streams (tunes off sound of mom, focuses on vision of game), and this leads to better desired prediction futures. This is how it decides to ignore/pay attention to certain known voices, past words/objects said/seen, this is how it chats, this is how it says "yes cats do indeed eat food" - the 'yes' and 'do' and 'indeed' are injected in at a random area, and this is how it finds context far back in a memory like where did she say she was going? "ack. And [she went to the shop], and Tuesday the cat I saw was in my home my food was so yu..." which is just prediction - it is in memory in a tree, unified, not a single long stream!, while if it is looking for it cus not in memory then it is surfing the web/real world collecting data or trying predictions in a lab - we can do that for now for it/them :). Remember what I taught yous, the most powerful technology is large "systems", distributed/aligned domains. It uses Big Context to vote in and make its decisions to make the future. Suns are big, so are Uranium atoms, and they extract radiation when too big. AGI extracts insights, free energy explosion like gasoline or fire on a house spreading like cancer cells burning up all energy it can find bye bye. Nanobots will spread too. Hopefully they hold us in the sack and ask us stuff before going into our brains/bodies, ha! The foglets will get you and transform you. It's empirical you understand that all AI find/create patterns because it lets them solve unseen problems the programmer never set them to answer. And all of physics has patterns. The reason Earth will become a fractal pattern of nanobot units terraformed and completely organized and unified is because being cooperative is paramount by using distributed government and knowing where / when / what all is in your home base and mind lead to higher probability for prediction and survival - that's why Earth will be a fully organized fractal of metaloid_superorganisms. It's empirical you understand all AI use big data and they repair missing data by filing n blanks, transforming voice, images to night to day, style type, black to white skin, voice to image, text to image, translation, frame rate fill in, image extension, super-resolution, etc. These are all just transforms and entailment (translation, sequence prediction, segmentation). Turing tapes can compute all algorithms, and AGI needs to do what it can do: check state its in, write/read bit/s, go to next location, repeat. It's empirical you understand friend networks also use trust, just as brain nets do. This sun moon compressor thing is everywhere. You have to remember, what do u want AGIs to do? AGIs are humans but with unhumanly capabilities and hence are ASIs technically, ans Big Data especially = ASSSSI ! They will work with/collect/clone microscopic data to create nanobots. To survive, we need to know our world well, we can't wait for the walker bot to get hit in the head and respond directly to stimuli, we need a thinker/planner, not a body, looking at all past context and predicting far ahead into the future comparing outcomes and what'd happen if one such happens before the even come! We need big data and big compute and many nanobot researchers. What you need, is a Turing Tape AGI that works with context and decides what to do next. Like an image, you decide where to look, what you see/read it as, what to write, then where to look next and what to say next. And to end this off well, see how the following illusions change what you think you see? Based on context they change their size, location, rotation, motion, rewardingness/pain, what they are, where the segmentation is, brightness, color, etc. https://ibb.co/g4MYqLs See, we dream not just in the night, but in the day, even when we don't day-dream. We are predictors. We are machines from evolution. And we aren't even different from tiny mites that search for food mate sleep and poop, they too make decisions based on context, like a fetus does as grows. Cells do too, wounds heal, animals regenerate, and all AI regenerate missing data, our brains lose neurons/humans every data in the net and replaces employees with newer better ones. All dictionary words/objects in life explain each other, you can get to any word by just a few steps in the dictionary! Small world network, can affect others by distributed offloading means. ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T2a0cd9d392f9ff94-M431197faea3d4e3fc58a331f Delivery options: https://agi.topicbox.com/groups/agi/subscription
