Hello,
I would like to make a game where a robot/AI takes in input of all the parts it can see and you can move around and it will record that data and each round it will do a recreation of the scenes it experienced using AI and the more scenes it experiences the more realistic the “dreams” will be so like if it only experiences one scene with a character moving around and pushing a ball around then when it dreams it will look very shaggy and everything will be all over the place but as iterations go on everything looks more and more in order
How would I make this script
Are there any existing neural network APIs/libraries i could use and learn to make this work
AI’s don’t have consciousness. It’d be like saying a CFD simulation has consciousness. All an AI is (assuming we are talking about ML) is it gets inputs and tweaks these things called weights to get as close to a target. A algorithm solver in a sense, they make their own algorithms for problems. I think there is this one called neural net but even then it is just OOP you still are coding the synapses yourself. Speaking on the math behind it you basically get the delta change of this activation function then apply it to your weights. But how should my neural network look like? Well you can either guess or optimize it by using genetic algorithms. Speaking on your more realistic dreams if u mean like generating one from previous data we used what is called a seq2seq model
I reccomend you read these