Wow Matt good generated data you coughed up. So useful. Let me intake it all.
These companies with trillions of dollars still never made AGI because the money doesn't help, it's about how many thinkers are thinking away in their office chairs. The only way money can help generate AGI is by making a more brute force AI or by making more data (ex. a fleet of Google cars). But we already have enough data and manual data and we need to design the AGI on our computers or paper, not a few supercomputers because nobody but a few Google members has those. On Sunday, December 01, 2019, at 8:51 PM, Matt Mahoney wrote: > But this approach doesn't even make sense. Our whole economy is based on job > specialization. It is far more efficient to organize machines like we > organize people, each doing a specific task. Everyone making progress in AI > is doing narrow AI, and really this is the only practical approach. Instead > of trying to automate a million different jobs all at once, you'll have more > success automating one job. That's going to be hard enough, given that all > the low hanging fruit has been picked. We have to remember what we want. Do we want cement? Nebulas? HDDs? Kids? Gusting wind? Huh? We want survival per Darwinian. We emerge, breed, persist, through repairing data information. Our evolution is exponential. The cell phones, big data, software, cars, is what increased information repair. Just like a cell or brain, faster. The more data; the more diverse patterns you learn, and the more specialized you learn (towards data science information technology, repair), the more you are a "know it all", because patterns allow you to solve more and more unseen data questions with missing data. The more we repair/reGenerate, the faster we can do it again, hence evolution is exponential. We don't need to specialize in fashion, HDDs, monitors, ... you said humans have a lot of specialty. All we need is agents to build i bigger better model and these agents will discover how to repair the data we have and evolve to the next stage. A self recursion of data. If we have a trillion AGI minds on a new nanotech computer grid thinking away, we do want them to solve different question and not be clones. You say all humans do a different specialized task, and so should the many AGIs, it works, it hard enough. Hmm, each one of the agents in said group is helping the data repair, they learn, they update the related nearby agents what they learn. More diverse data memory and a wider diverse friend network frameworks is better. They offload data or orders to sub memory/friend nodes. The bottom nodes are the workers. The end goal is survival, equilibrium, where we instantly replace back any data that is out of place - if a man dies he is replaced at a job, if a cell or lost it is replaced. Immortal, globally, it is and can't die. We emerge from nothing btw. So you're right, diversely tasked AGIs with diverse wisdom and manipulation ability (with most layer agents having the same wisdom and tasks but different approaches) making up a global hierarchy of offload is optimal. When I/we/you make the first or few AGIs, they will have some questions/data they try to repair/answer, maybe the same question in fact, it will be the most important questions, all we need do is make 1 AGI and then we can have it answer any question - and importantly any wall that arises during its answer. Therefore while you're correct, we do need a general AGI to do a single problem. Because natural language / patterns are what all data is, the more data - the better it can solve the problem in various ways - you can discover an answer to a question is ether to use a car or a house to hide behind if on the run - requiring you to know about cars plus houses plus hiding. Say we did automate each task/problem/job - calculator, building cars, driving cars, crawling your brain using nanotech organisms (most advanced technology possible), clearly there's too many and they overlap on-the-job, the nanobot task has many tasks along the way. And a lot of things we don't know how to make yet. How do you ever, or in the future, intend to automate that? You have each human and agent doing individual tasks....no cooperation....you add higher government layers bosses or incentives....cooperation emerges....this hierarchy we built here is a friend connection cooperation with aligned propagation waves. It's needed. Nanotech uses this. As well, a single agent in this hierarchy needs a hierarchy...think about it you have a society of many differently tasked humans as Matt said and many many many very precise tasked brittle ANI agents - the men work on problems like cars, roads, processors, the ANIs work on very simple problems like math, driving, but the ANIs while like us are specialized each with own task - are less capable. We need many specialized AGIs TO-DO 1 task. These many ANIs in our example here are like many many friend nodes of the bottom most layer of the friend hierarchy i mentioned, the hierarchy of coop friends has agents that build bigger stronger groups - bigger groups are more flexible - well guess what - each agent making up his friend hierarchy is a node and is made of a huge hierarchy of wisdom - our ANIs all over earth specialized are shallow bottom nodes of wisdom, bad, we need AGIs, all friends working together as well. Diverse data patterns, diverse friend connections, lead to super intelligence. More context leads to more context takeoff. So, yes we need many specialized agents, but, they need to be AGIs, and they need to work together. We need bigger data brains, bigger groups. It's a fractal of hierarchies that are extremely flexible/general. Heterarchies too, with friends you have boss parent nodes and also related nodes that can take your task and do it for you. You may know cat = dog and dogs run so you can answer your Question like ths > "Cats _" > "Cats run". ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T7116d9f5c1f0551d-Mbe905673f770436e18a4cd09 Delivery options: https://agi.topicbox.com/groups/agi/subscription
