Hundreds of CPU cores and GPU stacks. A very large deep neural network (or a variant of DNNs). Thousands of watts. Energy. Time.
What does it take to have an energy efficient and reasonably sized AI? What are the thought experiments required here? I am not alluding to the mainstream neuromorphic computing research. Graphene dendritic transistors? Memristors based on 2-dimensional materials? No. They are exciting advances in the field, but the hypothetical at hand is quite different. It is - If one envisioned a world having an artificially designed intelligent system like a human (or even a rodent) brain, what would that system look like? What would that system contain? Components. Network architectures. Algorithms. Even if they sound fictional, at least, can we imagine the details as much as possible?
What does it take to have an energy efficient and reasonably sized AI? What are the thought experiments required here? I am not alluding to the mainstream neuromorphic computing research. Graphene dendritic transistors? Memristors based on 2-dimensional materials? No. They are exciting advances in the field, but the hypothetical at hand is quite different. It is - If one envisioned a world having an artificially designed intelligent system like a human (or even a rodent) brain, what would that system look like? What would that system contain? Components. Network architectures. Algorithms. Even if they sound fictional, at least, can we imagine the details as much as possible?