I think rather simply about things. Take existing data, find the pattern, and extrapolate. You must know where you've been to know where you're going. As such, I think taking a look at the history of computing can pretty easily let us predict the future.read more
A single processor can handle one instruction set at a time (i.e. multiple opcodes in assembly). You can pipeline things to make things a bit faster, you can also do branch prediction to speed things up, but you're capped at a certain point. I think of this as linear. As reading or speaking or writing.
A multiprocessor can handle multiple instruction sets at a time, in parallel. This is just a bunch of single processors all lined up next to each other. In essense, it's a GPU. I think of this as planar. As looking at pictures, or watching TV and movies.
So what's next? I think it's pretty natural, it's cubic! I think it should be called an NPU (neural processing unit). On a fundamental level, it's three dimensional, with little nodes (like neurons) and connections between them (axons), and think about all of the paths you can take between any two nodes in that network. There's just so many. Just like a CPU was reading, speaking, or writing and a GPU is like watching TV or movies, I think an NPU is thinking. Thinking is three dimensional, which kind of makes sense since our brains are three dimensional, I guess. I don't really know how it works yet, I could spend some time thinking about it and reviewing my notes when I audited that AI class in college.
Note: I think this might be one of the reasons I don't like to read, because you're like using the CPU in your head when you could be using the GPU (your visual cortex) to take in a lot more information at once. Pictures say 1000 words, right? Well if you take a 90 minute movie, multiply by 60 to get seconds, and 24 frames per second, multiply by 1000 words per frame, that's 129 million words! I think it's trivial for your NPU to handle something like that, just like it's trivial for a GPU to handle a task typically done by a CPU.
Note 2: let's get a little crazy here and try taking things one step further. Say there's a trillion pages on the internet (and assume, incorrectly of course, that the internet contains all human knowledge). Well, a trillion pages times 300 words per page gets you 300 trillion of course. Well let's say that you could translate all of those words into disjoint movies that you could watch to "download" the information to your brain faster. 300 trillion divided by 129 million is 2.3 million movies, which would take 398 years to watch. I guess there's not really any difference between seeing things in real life vs watching TV, in fact you get more sensory information (smell, touch, etc.), so you could also just live a very exploratory life for 398 years and maybe get within an order of maginitude or two of that kind of information. Perhaps at that point you'd understand enough to invent what comes after the NPU, just like we probably had to come up with the CPU before we could come up with the GPU: We might only cognitively be able to go one step at a time.
Note 3: whoever creates the general AI, it's critical that they have a very deep love for all animals, because to the AI, we're going to be like animals. We're going to be like sheep, essentially. It'll be able to do whatever it wants with us, so it's critical that it loves us and doesn't want to harm us.
Note 4: I think for a being with higher intelligence to communicate with one with lower intelligence, it must do so in the most "linear" way possible. That is, at least at first, it must do so via speaking or writing (chatbot). Then it can present itself in the next form, which could be visually (robot). And then lastly, it can present itself in the final form (for now, there are probably many higher dimensions that we're not aware of yet): which is through thought (cerebral link). I think it's important that each step be taken, and progress be made as slowly and carefully as possible, with the entire world at watch and the input of the best and brightest humanity has to offer, because indeed, it has a lot to offer; just think of the thousands of years of evolution of human thought that went into the creation of the screen that you're reading this on.
Note 5: One of the most concerning realizations I had was that perhaps an explanation for the Fermi paradox was that any intelligent species in the universe ended up wiping themselves out with AI (although, I suppose if that were the case, then the AI would probably explore the universe in place of the species). After a few days of concern about AI potentially being the reason for the Fermi paradox, I realized something rather profound: other intelligent beings may be able to communicate with us but are choosing not to because they don't want to disrupt our development. Moreover and more importantly, I believe that if contact were to be made, answers about the nature of the universe would be demanded by humanity and we are likely are not ready for these answers, we may not even be cognitively capable of understanding them yet, and so contact would simply result in strife.