I know some people do a lot of AND OR XOR etc etc, but these are just Markov Chain rules - a very rare class of MC rule, it would not be so useful with just these patterns, and in fact should be more an emergent class I think. For example they get the AI doing if cat is red AND left within 1 minute ago, do X, so I think that allows it to check some conditions and do A or B stuff, which is the tail end prediction, however, like I said there is many types of possible rules like this, Markov Chains cover 80% of them, so you needs the simple ~10 pattern mechanisms and then tech it the rest or let it emerge its own rules.
You need the how to build a ladder linkages, this is easily seeable on openAI.com it can fill in the rest of images, which is really hard to do. Then you need the goal decider part to decide will it fill in the rest of the ladder, or heart, or house, or what? This is the goal that what is it's goal/prompt to complete? Is it want to invent new homes, AGI, faster HDDs, paperclips? So this controls the prompt it is given to even begin with. There is a motor to sensory control, that you can ask it to get you the first and last letters of a sentence. This is fed like a prompt like GPT-2 takes in, but would use motor actions to control text notepad inputs to itself. This is rarer pattern mechanisms that are learnt, it is pretty much the same it feeds it new data it desires - that it predicts/ or listens to. ....So to me AGI is mostly the DALL-E thing but with a cherry on top called RL to control what prompt it starts with each morning. It's motors are a different way to collect on topic data. Most rare behavior emerges, it only needs simple pattern mechanisms. It should do a LOT at first, like a markov chain, as you add more pattern mechanisms it does even more, and emerges on its own. So for me, the x AND y ? thing and motor control is not agency, they are the fill in the ladder and what ladder to complete....You can see how simple my code is right now and it can predict a lot of completions, there is no massive codebase. To get it to more the things like x AND y ? would require it to look at the context given in a different way so it can predict the likely thing/ what prompts it would predict to desire to start talking about (always trying to improve its agenda topic, I went from immortality to AGI to etc specializing my field domain). ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T5417cc95d981211e-M248fc6bf81ab7d2588214cf6 Delivery options: https://agi.topicbox.com/groups/agi/subscription
