Resident agent: we are creating an AI to live inside the internet, or how we got to AI by giving up on the world

We can think of intelligence as a simulator. Given a certain number of inputs, an agent is able to predict which outputs will normally follow. The simulator has been trained on the repetition of a very large number of cases where similar inputs have been recorded. That allows it to estimate probabilities in a more or less reliable way. A human agent mig…